2022-12-07 04:36:17,368 INFO [train.py:941] (3/4) Training started 2022-12-07 04:36:17,369 INFO [train.py:951] (3/4) Device: cuda:3 2022-12-07 04:36:17,415 INFO [lexicon.py:168] (3/4) Loading pre-compiled data/lang_char/Linv.pt 2022-12-07 04:36:17,423 INFO [train.py:962] (3/4) {'best_train_loss': inf, 'best_valid_loss': inf, 'best_train_epoch': -1, 'best_valid_epoch': -1, 'batch_idx_train': 0, 'log_interval': 100, 'reset_interval': 200, 'valid_interval': 3000, 'feature_dim': 80, 'subsampling_factor': 4, 'warm_step': 2000, 'env_info': {'k2-version': '1.23', 'k2-build-type': 'Release', 'k2-with-cuda': True, 'k2-git-sha1': 'b2ce63f3940018e7b433c43fd802fc50ab006a76', 'k2-git-date': 'Wed Nov 23 08:43:43 2022', 'lhotse-version': '1.9.0.dev+git.97bf4b0.dirty', 'torch-version': '1.10.0+cu102', 'torch-cuda-available': True, 'torch-cuda-version': '10.2', 'python-version': '3.8', 'icefall-git-branch': 'ali_meeting', 'icefall-git-sha1': 'f13cf61-dirty', 'icefall-git-date': 'Tue Dec 6 03:34:27 2022', 'icefall-path': '/exp/draj/mini_scale_2022/icefall', 'k2-path': '/exp/draj/mini_scale_2022/k2/k2/python/k2/__init__.py', 'lhotse-path': '/exp/draj/mini_scale_2022/lhotse/lhotse/__init__.py', 'hostname': 'r8n04', 'IP address': '10.1.8.4'}, 'world_size': 4, 'master_port': 12354, 'tensorboard': True, 'num_epochs': 20, 'start_epoch': 1, 'start_batch': 0, 'exp_dir': PosixPath('pruned_transducer_stateless7/exp/v1'), 'lang_dir': 'data/lang_char', 'base_lr': 0.05, 'lr_batches': 5000, 'lr_epochs': 3.5, 'context_size': 2, 'prune_range': 5, 'lm_scale': 0.25, 'am_scale': 0.0, 'simple_loss_scale': 0.5, 'seed': 42, 'print_diagnostics': False, 'inf_check': False, 'save_every_n': 5000, 'keep_last_k': 10, 'average_period': 200, 'use_fp16': True, 'num_encoder_layers': '2,4,3,2,4', 'feedforward_dims': '1024,1024,2048,2048,1024', 'nhead': '8,8,8,8,8', 'encoder_dims': '384,384,384,384,384', 'attention_dims': '192,192,192,192,192', 'encoder_unmasked_dims': '256,256,256,256,256', 'zipformer_downsampling_factors': '1,2,4,8,2', 'cnn_module_kernels': '31,31,31,31,31', 'decoder_dim': 512, 'joiner_dim': 512, 'manifest_dir': PosixPath('data/manifests'), 'enable_musan': True, 'concatenate_cuts': False, 'duration_factor': 1.0, 'gap': 1.0, 'max_duration': 300, 'max_cuts': 100, 'num_buckets': 50, 'on_the_fly_feats': False, 'shuffle': True, 'num_workers': 8, 'enable_spec_aug': True, 'spec_aug_time_warp_factor': 80, 'blank_id': 0, 'vocab_size': 3290} 2022-12-07 04:36:17,423 INFO [train.py:964] (3/4) About to create model 2022-12-07 04:36:17,822 INFO [zipformer.py:179] (3/4) At encoder stack 4, which has downsampling_factor=2, we will combine the outputs of layers 1 and 3, with downsampling_factors=2 and 8. 2022-12-07 04:36:17,865 INFO [train.py:968] (3/4) Number of model parameters: 75734561 2022-12-07 04:36:22,649 INFO [train.py:983] (3/4) Using DDP 2022-12-07 04:36:23,011 INFO [asr_datamodule.py:357] (3/4) About to get AMI train cuts 2022-12-07 04:36:23,015 INFO [asr_datamodule.py:204] (3/4) About to get Musan cuts 2022-12-07 04:36:23,016 INFO [asr_datamodule.py:208] (3/4) Enable MUSAN 2022-12-07 04:36:24,404 INFO [asr_datamodule.py:232] (3/4) Enable SpecAugment 2022-12-07 04:36:24,405 INFO [asr_datamodule.py:233] (3/4) Time warp factor: 80 2022-12-07 04:36:24,405 INFO [asr_datamodule.py:246] (3/4) About to create train dataset 2022-12-07 04:36:24,405 INFO [asr_datamodule.py:259] (3/4) Using DynamicBucketingSampler. 2022-12-07 04:36:24,760 INFO [asr_datamodule.py:268] (3/4) About to create train dataloader 2022-12-07 04:36:24,760 INFO [asr_datamodule.py:381] (3/4) About to get AliMeeting IHM eval cuts 2022-12-07 04:36:24,762 INFO [asr_datamodule.py:300] (3/4) About to create dev dataset 2022-12-07 04:36:24,950 INFO [asr_datamodule.py:315] (3/4) About to create dev dataloader 2022-12-07 04:36:55,259 INFO [train.py:873] (3/4) Epoch 1, batch 0, loss[loss=5.144, simple_loss=4.675, pruned_loss=4.674, over 3908.00 frames. ], tot_loss[loss=5.144, simple_loss=4.675, pruned_loss=4.674, over 3908.00 frames. ], batch size: 100, lr: 2.50e-02, grad_scale: 2.0 2022-12-07 04:36:55,259 INFO [train.py:896] (3/4) Computing validation loss 2022-12-07 04:37:02,329 INFO [train.py:905] (3/4) Epoch 1, validation: loss=4.832, simple_loss=4.375, pruned_loss=4.552, over 857387.00 frames. 2022-12-07 04:37:02,330 INFO [train.py:906] (3/4) Maximum memory allocated so far is 2969MB 2022-12-07 04:37:05,271 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=5.0, num_to_drop=2, layers_to_drop={0, 1} 2022-12-07 04:37:05,405 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=88.27 vs. limit=5.0 2022-12-07 04:37:10,013 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=8.45 vs. limit=2.0 2022-12-07 04:37:18,880 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 04:37:32,955 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=199.00 vs. limit=5.0 2022-12-07 04:37:40,099 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.0676, 5.0679, 5.0631, 5.0571, 4.3353, 4.9878, 5.0169, 5.0215], device='cuda:3'), covar=tensor([0.0010, 0.0010, 0.0010, 0.0014, 0.0023, 0.0011, 0.0006, 0.0009], device='cuda:3'), in_proj_covar=tensor([0.0009, 0.0009, 0.0009, 0.0009, 0.0009, 0.0009, 0.0009, 0.0009], device='cuda:3'), out_proj_covar=tensor([9.1956e-06, 9.2484e-06, 9.0974e-06, 9.3481e-06, 9.2369e-06, 9.2645e-06, 9.1997e-06, 9.3760e-06], device='cuda:3') 2022-12-07 04:37:44,025 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=24.11 vs. limit=2.0 2022-12-07 04:37:44,440 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=7.59 vs. limit=2.0 2022-12-07 04:37:48,807 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=155.64 vs. limit=5.0 2022-12-07 04:38:01,934 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 04:38:03,516 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=8.96 vs. limit=2.0 2022-12-07 04:38:13,699 INFO [train.py:873] (3/4) Epoch 1, batch 100, loss[loss=0.5442, simple_loss=0.4676, pruned_loss=0.608, over 14430.00 frames. ], tot_loss[loss=0.9373, simple_loss=0.8416, pruned_loss=0.8748, over 882741.45 frames. ], batch size: 53, lr: 3.00e-02, grad_scale: 0.125 2022-12-07 04:38:16,997 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.426e+01 1.096e+02 2.039e+02 4.080e+02 7.158e+03, threshold=4.078e+02, percent-clipped=0.0 2022-12-07 04:38:28,320 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=8.99 vs. limit=2.0 2022-12-07 04:38:34,167 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=4.01 vs. limit=2.0 2022-12-07 04:38:42,191 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=144.0, num_to_drop=2, layers_to_drop={0, 2} 2022-12-07 04:38:50,194 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0464, 3.8608, 4.5594, 4.5271, 4.5406, 4.5445, 2.2619, 4.5445], device='cuda:3'), covar=tensor([0.0127, 0.0105, 0.0013, 0.0028, 0.0023, 0.0018, 0.0147, 0.0016], device='cuda:3'), in_proj_covar=tensor([0.0009, 0.0009, 0.0009, 0.0009, 0.0009, 0.0009, 0.0009, 0.0009], device='cuda:3'), out_proj_covar=tensor([8.7688e-06, 8.8157e-06, 8.7562e-06, 8.9692e-06, 8.8314e-06, 8.8501e-06, 8.7994e-06, 8.9325e-06], device='cuda:3') 2022-12-07 04:39:03,237 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9619, 3.9557, 3.9602, 3.9560, 3.9654, 3.9550, 3.9518, 3.9433], device='cuda:3'), covar=tensor([0.0026, 0.0034, 0.0026, 0.0025, 0.0017, 0.0035, 0.0045, 0.0028], device='cuda:3'), in_proj_covar=tensor([0.0009, 0.0009, 0.0009, 0.0009, 0.0009, 0.0009, 0.0009, 0.0009], device='cuda:3'), out_proj_covar=tensor([8.9646e-06, 8.9072e-06, 9.1700e-06, 8.9149e-06, 9.1003e-06, 9.0624e-06, 9.0270e-06, 9.0327e-06], device='cuda:3') 2022-12-07 04:39:21,525 INFO [train.py:873] (3/4) Epoch 1, batch 200, loss[loss=0.4925, simple_loss=0.4189, pruned_loss=0.4971, over 14277.00 frames. ], tot_loss[loss=0.6813, simple_loss=0.6005, pruned_loss=0.6645, over 1282833.12 frames. ], batch size: 28, lr: 3.50e-02, grad_scale: 0.25 2022-12-07 04:39:23,973 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.78 vs. limit=2.0 2022-12-07 04:39:24,231 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.8957, 5.8982, 5.8913, 5.8976, 5.8945, 5.8982, 5.8956, 5.8982], device='cuda:3'), covar=tensor([9.1041e-05, 2.0450e-04, 1.1932e-04, 2.3581e-04, 1.5626e-04, 7.9711e-05, 1.6427e-04, 1.6971e-04], device='cuda:3'), in_proj_covar=tensor([0.0009, 0.0009, 0.0009, 0.0009, 0.0009, 0.0009, 0.0009, 0.0009], device='cuda:3'), out_proj_covar=tensor([9.5295e-06, 9.4062e-06, 9.4614e-06, 9.0863e-06, 9.5359e-06, 9.1151e-06, 9.3046e-06, 9.2061e-06], device='cuda:3') 2022-12-07 04:39:24,801 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.748e+01 6.194e+01 9.239e+01 1.604e+02 3.274e+02, threshold=1.848e+02, percent-clipped=0.0 2022-12-07 04:39:58,317 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.1242, 5.1257, 5.1235, 5.1171, 5.1237, 5.1259, 5.1175, 5.1261], device='cuda:3'), covar=tensor([0.0006, 0.0007, 0.0004, 0.0012, 0.0004, 0.0006, 0.0022, 0.0005], device='cuda:3'), in_proj_covar=tensor([0.0009, 0.0009, 0.0009, 0.0009, 0.0009, 0.0009, 0.0009, 0.0009], device='cuda:3'), out_proj_covar=tensor([9.2885e-06, 9.3887e-06, 9.2758e-06, 9.0204e-06, 9.3760e-06, 9.1759e-06, 9.2756e-06, 9.1543e-06], device='cuda:3') 2022-12-07 04:40:05,949 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.63 vs. limit=2.0 2022-12-07 04:40:22,373 WARNING [optim.py:389] (3/4) Scaling gradients by 0.06466581672430038, model_norm_threshold=184.7753143310547 2022-12-07 04:40:22,528 INFO [optim.py:451] (3/4) Parameter Dominanting tot_sumsq module.encoder.encoder_embed.conv.6.weight with proportion 0.54, where dominant_sumsq=(grad_sumsq*orig_rms_sq)=4.438e+06, grad_sumsq = 3.527e+09, orig_rms_sq=1.258e-03 2022-12-07 04:40:27,614 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 04:40:30,367 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=300.0, num_to_drop=2, layers_to_drop={0, 3} 2022-12-07 04:40:31,030 INFO [train.py:873] (3/4) Epoch 1, batch 300, loss[loss=0.4554, simple_loss=0.3842, pruned_loss=0.425, over 14330.00 frames. ], tot_loss[loss=0.5742, simple_loss=0.4987, pruned_loss=0.5537, over 1584547.69 frames. ], batch size: 55, lr: 4.00e-02, grad_scale: 0.5 2022-12-07 04:40:34,363 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.885e+01 5.232e+01 7.085e+01 9.806e+01 2.857e+03, threshold=1.417e+02, percent-clipped=2.0 2022-12-07 04:41:12,170 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=357.0, num_to_drop=2, layers_to_drop={1, 2} 2022-12-07 04:41:29,512 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=9.00 vs. limit=5.0 2022-12-07 04:41:33,212 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=387.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 04:41:43,793 INFO [train.py:873] (3/4) Epoch 1, batch 400, loss[loss=0.4618, simple_loss=0.3858, pruned_loss=0.4092, over 14175.00 frames. ], tot_loss[loss=0.5249, simple_loss=0.4502, pruned_loss=0.4933, over 1722837.51 frames. ], batch size: 80, lr: 4.50e-02, grad_scale: 1.0 2022-12-07 04:41:47,370 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.807e+01 5.021e+01 6.661e+01 9.478e+01 2.713e+02, threshold=1.332e+02, percent-clipped=6.0 2022-12-07 04:42:10,540 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=439.0, num_to_drop=2, layers_to_drop={0, 3} 2022-12-07 04:42:14,539 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1966, 2.0852, 2.3115, 2.3643, 2.2102, 2.2916, 2.2922, 1.9775], device='cuda:3'), covar=tensor([0.0434, 0.0279, 0.0234, 0.0479, 0.1325, 0.0544, 0.0556, 0.3262], device='cuda:3'), in_proj_covar=tensor([0.0010, 0.0010, 0.0010, 0.0010, 0.0010, 0.0010, 0.0010, 0.0011], device='cuda:3'), out_proj_covar=tensor([9.9724e-06, 1.0100e-05, 1.0115e-05, 9.9234e-06, 1.0277e-05, 1.0112e-05, 9.8646e-06, 1.0978e-05], device='cuda:3') 2022-12-07 04:42:16,720 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=9.02 vs. limit=5.0 2022-12-07 04:42:17,106 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=448.0, num_to_drop=2, layers_to_drop={2, 3} 2022-12-07 04:42:31,848 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.26 vs. limit=5.0 2022-12-07 04:42:55,428 INFO [train.py:873] (3/4) Epoch 1, batch 500, loss[loss=0.4618, simple_loss=0.3726, pruned_loss=0.4173, over 14461.00 frames. ], tot_loss[loss=0.4983, simple_loss=0.422, pruned_loss=0.4538, over 1827341.55 frames. ], batch size: 27, lr: 4.99e-02, grad_scale: 1.0 2022-12-07 04:42:59,043 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.328e+01 4.979e+01 6.873e+01 9.259e+01 1.851e+02, threshold=1.375e+02, percent-clipped=7.0 2022-12-07 04:43:10,763 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7617, 2.9675, 3.0679, 2.8641, 3.0358, 3.5149, 3.1770, 3.6051], device='cuda:3'), covar=tensor([0.0970, 0.1227, 0.0882, 0.1117, 0.0919, 0.0308, 0.0761, 0.0476], device='cuda:3'), in_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0012, 0.0012, 0.0011, 0.0011, 0.0011], device='cuda:3'), out_proj_covar=tensor([1.0425e-05, 1.0975e-05, 1.0922e-05, 1.1497e-05, 1.1027e-05, 1.0245e-05, 1.0687e-05, 1.0154e-05], device='cuda:3') 2022-12-07 04:43:20,772 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.7461, 5.8052, 5.8125, 5.8192, 5.7986, 5.7685, 5.6202, 5.4406], device='cuda:3'), covar=tensor([0.0107, 0.0060, 0.0078, 0.0061, 0.0077, 0.0115, 0.0419, 0.0505], device='cuda:3'), in_proj_covar=tensor([0.0010, 0.0010, 0.0010, 0.0010, 0.0010, 0.0010, 0.0012, 0.0010], device='cuda:3'), out_proj_covar=tensor([9.6498e-06, 1.0231e-05, 9.6985e-06, 9.6288e-06, 9.7349e-06, 1.0111e-05, 1.0923e-05, 1.0098e-05], device='cuda:3') 2022-12-07 04:43:30,752 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=3.50 vs. limit=2.0 2022-12-07 04:43:35,419 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=3.73 vs. limit=2.0 2022-12-07 04:43:37,337 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=3.28 vs. limit=2.0 2022-12-07 04:43:38,493 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=562.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 04:43:42,879 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=568.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 04:43:52,946 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0597, 1.9101, 2.2680, 1.7857, 2.8425, 2.5468, 1.8426, 2.4713], device='cuda:3'), covar=tensor([0.2360, 0.1740, 0.1781, 0.2713, 0.0825, 0.1082, 0.3296, 0.1592], device='cuda:3'), in_proj_covar=tensor([0.0015, 0.0013, 0.0014, 0.0014, 0.0014, 0.0013, 0.0015, 0.0013], device='cuda:3'), out_proj_covar=tensor([1.2501e-05, 1.2556e-05, 1.2861e-05, 1.3834e-05, 1.2363e-05, 1.1911e-05, 1.3913e-05, 1.2605e-05], device='cuda:3') 2022-12-07 04:43:57,726 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=590.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 04:44:05,271 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=600.0, num_to_drop=2, layers_to_drop={0, 1} 2022-12-07 04:44:05,744 INFO [train.py:873] (3/4) Epoch 1, batch 600, loss[loss=0.326, simple_loss=0.2958, pruned_loss=0.2015, over 1245.00 frames. ], tot_loss[loss=0.4812, simple_loss=0.4025, pruned_loss=0.4243, over 1910099.65 frames. ], batch size: 100, lr: 4.98e-02, grad_scale: 1.0 2022-12-07 04:44:09,407 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.092e+01 6.985e+01 9.438e+01 1.366e+02 4.481e+02, threshold=1.888e+02, percent-clipped=23.0 2022-12-07 04:44:21,320 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=623.0, num_to_drop=2, layers_to_drop={0, 1} 2022-12-07 04:44:25,258 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=629.0, num_to_drop=2, layers_to_drop={0, 2} 2022-12-07 04:44:30,961 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=4.38 vs. limit=2.0 2022-12-07 04:44:34,401 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=4.46 vs. limit=2.0 2022-12-07 04:44:39,013 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=648.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 04:44:39,858 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4864, 2.3969, 2.6271, 2.6157, 2.2813, 2.4005, 2.2548, 2.5233], device='cuda:3'), covar=tensor([0.2162, 0.1446, 0.1723, 0.1462, 0.2011, 0.2437, 0.1905, 0.1645], device='cuda:3'), in_proj_covar=tensor([0.0014, 0.0013, 0.0013, 0.0013, 0.0015, 0.0015, 0.0014, 0.0014], device='cuda:3'), out_proj_covar=tensor([1.4077e-05, 1.2031e-05, 1.2294e-05, 1.2931e-05, 1.3839e-05, 1.4246e-05, 1.3143e-05, 1.3105e-05], device='cuda:3') 2022-12-07 04:44:41,217 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=651.0, num_to_drop=2, layers_to_drop={1, 3} 2022-12-07 04:44:41,813 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=652.0, num_to_drop=2, layers_to_drop={2, 3} 2022-12-07 04:45:15,578 INFO [train.py:873] (3/4) Epoch 1, batch 700, loss[loss=0.458, simple_loss=0.3735, pruned_loss=0.3609, over 14420.00 frames. ], tot_loss[loss=0.4697, simple_loss=0.3888, pruned_loss=0.401, over 1925867.23 frames. ], batch size: 53, lr: 4.98e-02, grad_scale: 1.0 2022-12-07 04:45:19,022 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.942e+01 9.279e+01 1.429e+02 2.357e+02 5.770e+02, threshold=2.857e+02, percent-clipped=36.0 2022-12-07 04:45:31,532 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=9.88 vs. limit=5.0 2022-12-07 04:45:42,042 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=739.0, num_to_drop=2, layers_to_drop={0, 1} 2022-12-07 04:45:45,172 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=743.0, num_to_drop=2, layers_to_drop={2, 3} 2022-12-07 04:46:15,754 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=787.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 04:46:25,998 INFO [train.py:873] (3/4) Epoch 1, batch 800, loss[loss=0.4354, simple_loss=0.3577, pruned_loss=0.3243, over 14281.00 frames. ], tot_loss[loss=0.4573, simple_loss=0.3762, pruned_loss=0.3755, over 1966482.35 frames. ], batch size: 31, lr: 4.97e-02, grad_scale: 2.0 2022-12-07 04:46:29,402 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.458e+01 1.291e+02 1.818e+02 2.513e+02 6.152e+02, threshold=3.636e+02, percent-clipped=18.0 2022-12-07 04:46:58,625 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=847.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 04:47:25,208 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.09 vs. limit=2.0 2022-12-07 04:47:27,029 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=7.20 vs. limit=5.0 2022-12-07 04:47:36,965 INFO [train.py:873] (3/4) Epoch 1, batch 900, loss[loss=0.4425, simple_loss=0.3656, pruned_loss=0.3149, over 14295.00 frames. ], tot_loss[loss=0.4432, simple_loss=0.3641, pruned_loss=0.3505, over 1891449.70 frames. ], batch size: 28, lr: 4.96e-02, grad_scale: 2.0 2022-12-07 04:47:40,263 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.004e+01 1.837e+02 2.554e+02 3.837e+02 9.711e+02, threshold=5.109e+02, percent-clipped=29.0 2022-12-07 04:47:41,893 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=908.0, num_to_drop=2, layers_to_drop={0, 2} 2022-12-07 04:47:48,595 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=918.0, num_to_drop=2, layers_to_drop={0, 3} 2022-12-07 04:47:52,726 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=924.0, num_to_drop=2, layers_to_drop={1, 3} 2022-12-07 04:47:59,348 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.00 vs. limit=2.0 2022-12-07 04:48:01,034 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2318, 3.6850, 3.3961, 3.7876, 3.7868, 3.9949, 3.5534, 3.5995], device='cuda:3'), covar=tensor([0.4014, 0.2054, 0.3907, 0.2243, 0.2219, 0.1684, 0.2098, 0.3274], device='cuda:3'), in_proj_covar=tensor([0.0033, 0.0031, 0.0032, 0.0032, 0.0032, 0.0029, 0.0031, 0.0030], device='cuda:3'), out_proj_covar=tensor([3.3214e-05, 2.8568e-05, 3.0660e-05, 2.6840e-05, 2.6278e-05, 2.5577e-05, 2.8253e-05, 2.7770e-05], device='cuda:3') 2022-12-07 04:48:08,950 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=946.0, num_to_drop=2, layers_to_drop={2, 3} 2022-12-07 04:48:13,017 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=952.0, num_to_drop=2, layers_to_drop={0, 1} 2022-12-07 04:48:39,053 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.08 vs. limit=2.0 2022-12-07 04:48:46,727 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1000.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 04:48:47,390 INFO [train.py:873] (3/4) Epoch 1, batch 1000, loss[loss=0.4221, simple_loss=0.3464, pruned_loss=0.2949, over 14191.00 frames. ], tot_loss[loss=0.4303, simple_loss=0.3536, pruned_loss=0.3269, over 1879897.14 frames. ], batch size: 94, lr: 4.95e-02, grad_scale: 2.0 2022-12-07 04:48:50,982 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.184e+02 2.449e+02 3.112e+02 4.355e+02 1.197e+03, threshold=6.224e+02, percent-clipped=18.0 2022-12-07 04:49:17,168 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1043.0, num_to_drop=2, layers_to_drop={1, 2} 2022-12-07 04:49:40,453 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.37 vs. limit=5.0 2022-12-07 04:49:51,011 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1091.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 04:49:58,566 INFO [train.py:873] (3/4) Epoch 1, batch 1100, loss[loss=0.4026, simple_loss=0.3337, pruned_loss=0.27, over 14270.00 frames. ], tot_loss[loss=0.4187, simple_loss=0.3446, pruned_loss=0.305, over 1970416.07 frames. ], batch size: 35, lr: 4.94e-02, grad_scale: 2.0 2022-12-07 04:50:02,057 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.129e+02 2.767e+02 3.669e+02 5.185e+02 1.491e+03, threshold=7.337e+02, percent-clipped=11.0 2022-12-07 04:50:35,571 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1153.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 04:51:09,752 INFO [train.py:873] (3/4) Epoch 1, batch 1200, loss[loss=0.3777, simple_loss=0.315, pruned_loss=0.2456, over 6926.00 frames. ], tot_loss[loss=0.4039, simple_loss=0.3338, pruned_loss=0.2835, over 2000565.02 frames. ], batch size: 100, lr: 4.93e-02, grad_scale: 4.0 2022-12-07 04:51:11,152 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1203.0, num_to_drop=2, layers_to_drop={0, 1} 2022-12-07 04:51:12,882 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.090e+02 2.727e+02 3.697e+02 4.642e+02 1.114e+03, threshold=7.394e+02, percent-clipped=5.0 2022-12-07 04:51:18,428 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1214.0, num_to_drop=2, layers_to_drop={0, 3} 2022-12-07 04:51:21,873 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1218.0, num_to_drop=2, layers_to_drop={1, 2} 2022-12-07 04:51:25,881 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1224.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 04:51:35,470 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.10 vs. limit=2.0 2022-12-07 04:51:41,200 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1246.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 04:51:45,480 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.60 vs. limit=2.0 2022-12-07 04:51:55,345 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1266.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 04:51:59,306 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1272.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 04:52:15,064 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1294.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 04:52:20,508 INFO [train.py:873] (3/4) Epoch 1, batch 1300, loss[loss=0.3683, simple_loss=0.3089, pruned_loss=0.2333, over 14167.00 frames. ], tot_loss[loss=0.3929, simple_loss=0.3257, pruned_loss=0.2675, over 1963425.46 frames. ], batch size: 29, lr: 4.92e-02, grad_scale: 4.0 2022-12-07 04:52:23,970 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.301e+02 2.849e+02 3.755e+02 4.812e+02 1.034e+03, threshold=7.510e+02, percent-clipped=8.0 2022-12-07 04:53:18,114 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.5518, 1.1460, 1.6024, 1.2027, 1.7256, 1.2394, 1.2844, 1.4320], device='cuda:3'), covar=tensor([0.0784, 0.3130, 0.0747, 0.2043, 0.0689, 0.1239, 0.1377, 0.0791], device='cuda:3'), in_proj_covar=tensor([0.0024, 0.0031, 0.0025, 0.0030, 0.0027, 0.0028, 0.0026, 0.0028], device='cuda:3'), out_proj_covar=tensor([1.8960e-05, 2.9451e-05, 1.8555e-05, 2.7072e-05, 2.2314e-05, 2.0940e-05, 2.0303e-05, 2.1333e-05], device='cuda:3') 2022-12-07 04:53:23,157 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3542, 2.2508, 2.7307, 2.7105, 2.3852, 2.5763, 2.1690, 2.6309], device='cuda:3'), covar=tensor([0.1130, 0.1308, 0.0754, 0.0718, 0.1192, 0.0921, 0.1682, 0.1046], device='cuda:3'), in_proj_covar=tensor([0.0033, 0.0033, 0.0030, 0.0035, 0.0031, 0.0030, 0.0032, 0.0032], device='cuda:3'), out_proj_covar=tensor([2.7349e-05, 2.7135e-05, 2.7596e-05, 2.9146e-05, 2.5188e-05, 2.5036e-05, 2.5897e-05, 2.8127e-05], device='cuda:3') 2022-12-07 04:53:33,568 INFO [train.py:873] (3/4) Epoch 1, batch 1400, loss[loss=0.246, simple_loss=0.2083, pruned_loss=0.1516, over 1242.00 frames. ], tot_loss[loss=0.3832, simple_loss=0.319, pruned_loss=0.2532, over 1964695.39 frames. ], batch size: 100, lr: 4.91e-02, grad_scale: 4.0 2022-12-07 04:53:36,992 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.548e+02 3.437e+02 4.813e+02 6.365e+02 1.224e+03, threshold=9.626e+02, percent-clipped=13.0 2022-12-07 04:53:57,852 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1434.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 04:53:58,219 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.01 vs. limit=2.0 2022-12-07 04:54:36,075 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2724, 2.2404, 2.1497, 2.2150, 2.2077, 2.2214, 2.1007, 2.1240], device='cuda:3'), covar=tensor([0.0272, 0.0401, 0.0399, 0.0427, 0.0361, 0.0322, 0.0390, 0.0430], device='cuda:3'), in_proj_covar=tensor([0.0024, 0.0027, 0.0025, 0.0024, 0.0025, 0.0025, 0.0025, 0.0028], device='cuda:3'), out_proj_covar=tensor([2.2671e-05, 2.5674e-05, 2.4649e-05, 2.2724e-05, 2.3087e-05, 2.2899e-05, 2.4113e-05, 2.6801e-05], device='cuda:3') 2022-12-07 04:54:41,778 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1495.0, num_to_drop=2, layers_to_drop={2, 3} 2022-12-07 04:54:46,084 INFO [train.py:873] (3/4) Epoch 1, batch 1500, loss[loss=0.3492, simple_loss=0.3008, pruned_loss=0.208, over 14511.00 frames. ], tot_loss[loss=0.3739, simple_loss=0.3121, pruned_loss=0.2412, over 1933055.00 frames. ], batch size: 49, lr: 4.89e-02, grad_scale: 4.0 2022-12-07 04:54:47,684 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1503.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 04:54:49,749 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.091e+01 2.709e+02 3.903e+02 5.151e+02 1.116e+03, threshold=7.805e+02, percent-clipped=3.0 2022-12-07 04:54:52,086 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1509.0, num_to_drop=2, layers_to_drop={1, 2} 2022-12-07 04:55:22,606 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1551.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 04:55:33,682 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1566.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 04:55:46,868 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.24 vs. limit=2.0 2022-12-07 04:55:58,347 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.12 vs. limit=2.0 2022-12-07 04:55:59,955 INFO [train.py:873] (3/4) Epoch 1, batch 1600, loss[loss=0.3708, simple_loss=0.3101, pruned_loss=0.2253, over 14284.00 frames. ], tot_loss[loss=0.3671, simple_loss=0.3076, pruned_loss=0.2311, over 2017453.23 frames. ], batch size: 39, lr: 4.88e-02, grad_scale: 8.0 2022-12-07 04:56:03,633 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.027e+02 3.341e+02 4.121e+02 5.397e+02 1.445e+03, threshold=8.242e+02, percent-clipped=8.0 2022-12-07 04:56:07,799 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1611.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 04:56:19,555 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1627.0, num_to_drop=2, layers_to_drop={1, 2} 2022-12-07 04:56:25,489 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.11 vs. limit=2.0 2022-12-07 04:56:39,401 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.53 vs. limit=2.0 2022-12-07 04:56:52,764 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1672.0, num_to_drop=2, layers_to_drop={0, 2} 2022-12-07 04:57:02,206 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3353, 4.4947, 3.9376, 4.1541, 4.2545, 3.9338, 4.6598, 4.4822], device='cuda:3'), covar=tensor([0.0479, 0.0477, 0.0584, 0.0529, 0.0578, 0.0524, 0.0297, 0.0381], device='cuda:3'), in_proj_covar=tensor([0.0036, 0.0033, 0.0039, 0.0031, 0.0037, 0.0033, 0.0033, 0.0034], device='cuda:3'), out_proj_covar=tensor([3.5182e-05, 3.4845e-05, 3.7531e-05, 3.2196e-05, 3.7017e-05, 3.2793e-05, 3.4026e-05, 3.3688e-05], device='cuda:3') 2022-12-07 04:57:13,820 INFO [train.py:873] (3/4) Epoch 1, batch 1700, loss[loss=0.3875, simple_loss=0.3148, pruned_loss=0.2387, over 8648.00 frames. ], tot_loss[loss=0.362, simple_loss=0.3039, pruned_loss=0.2238, over 1961189.21 frames. ], batch size: 100, lr: 4.86e-02, grad_scale: 8.0 2022-12-07 04:57:17,699 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.181e+02 3.504e+02 4.735e+02 5.720e+02 1.294e+03, threshold=9.470e+02, percent-clipped=11.0 2022-12-07 04:58:13,230 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.31 vs. limit=2.0 2022-12-07 04:58:21,826 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1790.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 04:58:29,845 INFO [train.py:873] (3/4) Epoch 1, batch 1800, loss[loss=0.3686, simple_loss=0.3004, pruned_loss=0.2235, over 6929.00 frames. ], tot_loss[loss=0.3599, simple_loss=0.3021, pruned_loss=0.219, over 1986892.53 frames. ], batch size: 100, lr: 4.85e-02, grad_scale: 8.0 2022-12-07 04:58:33,677 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.780e+01 3.438e+02 4.789e+02 6.464e+02 1.407e+03, threshold=9.578e+02, percent-clipped=4.0 2022-12-07 04:58:36,063 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1809.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 04:58:41,662 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=7.55 vs. limit=5.0 2022-12-07 04:58:57,938 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1838.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 04:59:12,091 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1857.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 04:59:12,154 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1857.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 04:59:12,480 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.58 vs. limit=2.0 2022-12-07 04:59:21,927 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1870.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 04:59:43,721 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1899.0, num_to_drop=2, layers_to_drop={1, 2} 2022-12-07 04:59:45,047 INFO [train.py:873] (3/4) Epoch 1, batch 1900, loss[loss=0.3579, simple_loss=0.3032, pruned_loss=0.2081, over 14675.00 frames. ], tot_loss[loss=0.356, simple_loss=0.2992, pruned_loss=0.2136, over 1995714.77 frames. ], batch size: 33, lr: 4.83e-02, grad_scale: 8.0 2022-12-07 04:59:48,578 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.566e+01 3.394e+02 4.356e+02 5.869e+02 1.320e+03, threshold=8.711e+02, percent-clipped=5.0 2022-12-07 04:59:58,509 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1918.0, num_to_drop=2, layers_to_drop={2, 3} 2022-12-07 05:00:01,452 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1922.0, num_to_drop=1, layers_to_drop={3} 2022-12-07 05:00:08,196 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1931.0, num_to_drop=2, layers_to_drop={2, 3} 2022-12-07 05:00:18,255 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.53 vs. limit=5.0 2022-12-07 05:00:34,813 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4435, 2.1869, 1.8535, 2.6204, 2.4244, 2.5645, 1.9368, 1.4964], device='cuda:3'), covar=tensor([0.0641, 0.1128, 0.1956, 0.0816, 0.0663, 0.0787, 0.0759, 0.5585], device='cuda:3'), in_proj_covar=tensor([0.0027, 0.0030, 0.0036, 0.0026, 0.0024, 0.0028, 0.0032, 0.0064], device='cuda:3'), out_proj_covar=tensor([1.5346e-05, 1.9237e-05, 2.6162e-05, 1.5902e-05, 1.5645e-05, 1.6745e-05, 1.9998e-05, 5.5749e-05], device='cuda:3') 2022-12-07 05:00:35,417 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1967.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:00:37,494 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.99 vs. limit=2.0 2022-12-07 05:01:01,360 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2166, 1.9791, 2.0971, 1.7326, 2.1937, 2.0161, 2.2092, 1.7000], device='cuda:3'), covar=tensor([0.0307, 0.0408, 0.0293, 0.0360, 0.0257, 0.0436, 0.0259, 0.0509], device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0020, 0.0022, 0.0020, 0.0020, 0.0023, 0.0019, 0.0021], device='cuda:3'), out_proj_covar=tensor([1.9102e-05, 1.7189e-05, 1.8144e-05, 1.6703e-05, 1.8796e-05, 1.7817e-05, 1.6521e-05, 1.7107e-05], device='cuda:3') 2022-12-07 05:01:01,972 INFO [train.py:873] (3/4) Epoch 1, batch 2000, loss[loss=0.3685, simple_loss=0.3068, pruned_loss=0.2151, over 11969.00 frames. ], tot_loss[loss=0.3529, simple_loss=0.2971, pruned_loss=0.2091, over 1956912.29 frames. ], batch size: 100, lr: 4.82e-02, grad_scale: 8.0 2022-12-07 05:01:02,130 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3127, 3.2516, 3.5082, 3.6767, 3.4355, 3.5724, 2.6848, 3.5280], device='cuda:3'), covar=tensor([0.0465, 0.0620, 0.0431, 0.0300, 0.0413, 0.0363, 0.0977, 0.0470], device='cuda:3'), in_proj_covar=tensor([0.0024, 0.0025, 0.0023, 0.0022, 0.0027, 0.0023, 0.0022, 0.0025], device='cuda:3'), out_proj_covar=tensor([2.0662e-05, 2.0824e-05, 1.9201e-05, 1.7541e-05, 2.2590e-05, 1.9212e-05, 1.9484e-05, 2.3018e-05], device='cuda:3') 2022-12-07 05:01:05,829 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.042e+02 3.554e+02 4.725e+02 6.724e+02 1.490e+03, threshold=9.451e+02, percent-clipped=5.0 2022-12-07 05:01:40,784 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2022-12-07 05:02:14,851 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2090.0, num_to_drop=2, layers_to_drop={0, 2} 2022-12-07 05:02:23,810 INFO [train.py:873] (3/4) Epoch 1, batch 2100, loss[loss=0.3226, simple_loss=0.2815, pruned_loss=0.1819, over 14188.00 frames. ], tot_loss[loss=0.3449, simple_loss=0.2928, pruned_loss=0.2013, over 2008525.42 frames. ], batch size: 99, lr: 4.80e-02, grad_scale: 16.0 2022-12-07 05:02:27,743 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.809e+01 3.029e+02 3.794e+02 4.821e+02 1.470e+03, threshold=7.588e+02, percent-clipped=1.0 2022-12-07 05:02:29,007 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.06 vs. limit=2.0 2022-12-07 05:02:42,241 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6574, 4.2174, 4.6565, 4.1180, 4.1452, 4.5685, 3.6624, 4.3916], device='cuda:3'), covar=tensor([0.0552, 0.0502, 0.0141, 0.0179, 0.0284, 0.0138, 0.0617, 0.0218], device='cuda:3'), in_proj_covar=tensor([0.0026, 0.0027, 0.0021, 0.0022, 0.0022, 0.0021, 0.0026, 0.0023], device='cuda:3'), out_proj_covar=tensor([2.1989e-05, 2.1685e-05, 1.7505e-05, 1.8282e-05, 1.6620e-05, 1.6397e-05, 2.1780e-05, 1.8297e-05], device='cuda:3') 2022-12-07 05:02:53,153 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2138.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:03:14,427 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.2990, 3.9630, 4.7149, 3.8344, 4.3947, 4.4397, 4.1596, 4.2049], device='cuda:3'), covar=tensor([0.0292, 0.0461, 0.0283, 0.0736, 0.0220, 0.0197, 0.0407, 0.0272], device='cuda:3'), in_proj_covar=tensor([0.0033, 0.0036, 0.0033, 0.0033, 0.0036, 0.0029, 0.0035, 0.0039], device='cuda:3'), out_proj_covar=tensor([2.9892e-05, 3.2882e-05, 3.1612e-05, 2.9952e-05, 3.4022e-05, 2.7136e-05, 3.0829e-05, 3.7593e-05], device='cuda:3') 2022-12-07 05:03:38,958 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2194.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:03:45,040 INFO [train.py:873] (3/4) Epoch 1, batch 2200, loss[loss=0.3975, simple_loss=0.3198, pruned_loss=0.2376, over 8594.00 frames. ], tot_loss[loss=0.3414, simple_loss=0.2907, pruned_loss=0.1977, over 2012381.99 frames. ], batch size: 100, lr: 4.78e-02, grad_scale: 16.0 2022-12-07 05:03:49,158 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.495e+02 3.028e+02 4.377e+02 7.241e+02 2.014e+03, threshold=8.755e+02, percent-clipped=23.0 2022-12-07 05:03:54,619 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2213.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:04:01,849 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2222.0, num_to_drop=2, layers_to_drop={0, 2} 2022-12-07 05:04:04,965 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2226.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:04:09,394 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.13 vs. limit=5.0 2022-12-07 05:04:30,765 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.67 vs. limit=5.0 2022-12-07 05:04:38,731 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2267.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 05:04:40,838 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.99 vs. limit=2.0 2022-12-07 05:04:41,029 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2270.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:05:06,465 INFO [train.py:873] (3/4) Epoch 1, batch 2300, loss[loss=0.379, simple_loss=0.3136, pruned_loss=0.2222, over 14262.00 frames. ], tot_loss[loss=0.3333, simple_loss=0.2866, pruned_loss=0.191, over 2074440.25 frames. ], batch size: 66, lr: 4.77e-02, grad_scale: 16.0 2022-12-07 05:05:10,739 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.291e+01 3.360e+02 4.447e+02 6.275e+02 1.931e+03, threshold=8.894e+02, percent-clipped=10.0 2022-12-07 05:05:18,525 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2315.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:05:33,404 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.52 vs. limit=5.0 2022-12-07 05:05:38,595 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2340.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 05:05:55,484 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=10.90 vs. limit=5.0 2022-12-07 05:06:05,352 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8524, 1.8281, 2.1646, 1.9354, 1.8167, 2.0117, 1.6271, 1.9647], device='cuda:3'), covar=tensor([0.0127, 0.0196, 0.0069, 0.0112, 0.0151, 0.0160, 0.0293, 0.0120], device='cuda:3'), in_proj_covar=tensor([0.0021, 0.0025, 0.0022, 0.0020, 0.0025, 0.0021, 0.0020, 0.0021], device='cuda:3'), out_proj_covar=tensor([1.8531e-05, 2.3630e-05, 1.9535e-05, 1.6684e-05, 2.1783e-05, 1.8383e-05, 1.9034e-05, 1.8928e-05], device='cuda:3') 2022-12-07 05:06:28,674 INFO [train.py:873] (3/4) Epoch 1, batch 2400, loss[loss=0.3699, simple_loss=0.3025, pruned_loss=0.2186, over 7763.00 frames. ], tot_loss[loss=0.3296, simple_loss=0.2848, pruned_loss=0.1878, over 2044813.52 frames. ], batch size: 100, lr: 4.75e-02, grad_scale: 16.0 2022-12-07 05:06:28,875 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2401.0, num_to_drop=2, layers_to_drop={2, 3} 2022-12-07 05:06:32,588 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.275e+02 3.236e+02 4.365e+02 6.045e+02 1.327e+03, threshold=8.731e+02, percent-clipped=8.0 2022-12-07 05:07:14,175 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6897, 1.6122, 1.6760, 1.5475, 1.7161, 1.5946, 1.8085, 1.4561], device='cuda:3'), covar=tensor([0.0203, 0.0271, 0.0184, 0.0328, 0.0151, 0.0226, 0.0145, 0.0183], device='cuda:3'), in_proj_covar=tensor([0.0017, 0.0017, 0.0018, 0.0016, 0.0016, 0.0019, 0.0015, 0.0017], device='cuda:3'), out_proj_covar=tensor([1.7084e-05, 1.3832e-05, 1.5136e-05, 1.3706e-05, 1.4313e-05, 1.5096e-05, 1.3047e-05, 1.4408e-05], device='cuda:3') 2022-12-07 05:07:44,796 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2494.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 05:07:50,098 INFO [train.py:873] (3/4) Epoch 1, batch 2500, loss[loss=0.2633, simple_loss=0.2106, pruned_loss=0.158, over 1206.00 frames. ], tot_loss[loss=0.3245, simple_loss=0.2821, pruned_loss=0.1838, over 2042355.66 frames. ], batch size: 100, lr: 4.73e-02, grad_scale: 16.0 2022-12-07 05:07:54,362 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.013e+02 3.632e+02 4.712e+02 6.167e+02 1.692e+03, threshold=9.425e+02, percent-clipped=8.0 2022-12-07 05:08:00,489 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2513.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:08:07,544 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.35 vs. limit=5.0 2022-12-07 05:08:11,295 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2526.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:08:23,743 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2542.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:08:39,733 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2561.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:08:50,008 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2574.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:08:54,482 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.86 vs. limit=2.0 2022-12-07 05:09:13,580 INFO [train.py:873] (3/4) Epoch 1, batch 2600, loss[loss=0.338, simple_loss=0.2901, pruned_loss=0.193, over 14230.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.2799, pruned_loss=0.1812, over 2070721.53 frames. ], batch size: 69, lr: 4.71e-02, grad_scale: 16.0 2022-12-07 05:09:17,460 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.018e+02 2.998e+02 4.312e+02 6.612e+02 1.607e+03, threshold=8.624e+02, percent-clipped=4.0 2022-12-07 05:09:35,629 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8158, 2.5795, 3.6522, 3.1523, 3.8881, 3.6209, 3.2785, 2.1268], device='cuda:3'), covar=tensor([0.0164, 0.2346, 0.0162, 0.0403, 0.0141, 0.0213, 0.0263, 0.2235], device='cuda:3'), in_proj_covar=tensor([0.0036, 0.0066, 0.0031, 0.0040, 0.0033, 0.0033, 0.0027, 0.0066], device='cuda:3'), out_proj_covar=tensor([2.1557e-05, 6.2001e-05, 1.7845e-05, 2.8181e-05, 2.0499e-05, 1.9081e-05, 1.9051e-05, 5.5888e-05], device='cuda:3') 2022-12-07 05:10:23,657 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2685.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 05:10:25,173 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.3949, 0.9200, 1.3519, 0.9902, 1.3497, 0.8250, 1.4005, 1.3847], device='cuda:3'), covar=tensor([0.0092, 0.0449, 0.0161, 0.0397, 0.0132, 0.0273, 0.0122, 0.0118], device='cuda:3'), in_proj_covar=tensor([0.0023, 0.0030, 0.0027, 0.0034, 0.0024, 0.0030, 0.0024, 0.0027], device='cuda:3'), out_proj_covar=tensor([1.7302e-05, 2.6518e-05, 2.0225e-05, 2.9709e-05, 1.8197e-05, 2.6010e-05, 2.0554e-05, 2.1450e-05], device='cuda:3') 2022-12-07 05:10:32,502 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2696.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 05:10:36,434 INFO [train.py:873] (3/4) Epoch 1, batch 2700, loss[loss=0.2918, simple_loss=0.2664, pruned_loss=0.1586, over 14561.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.2775, pruned_loss=0.1772, over 2064410.31 frames. ], batch size: 43, lr: 4.69e-02, grad_scale: 16.0 2022-12-07 05:10:40,372 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.321e+02 3.041e+02 4.444e+02 6.160e+02 1.367e+03, threshold=8.889e+02, percent-clipped=4.0 2022-12-07 05:11:11,058 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8734, 2.1588, 3.2241, 2.1104, 2.5662, 2.6628, 2.0528, 2.3092], device='cuda:3'), covar=tensor([0.1698, 0.1071, 0.0577, 0.0779, 0.0571, 0.0404, 0.0759, 0.0545], device='cuda:3'), in_proj_covar=tensor([0.0032, 0.0029, 0.0020, 0.0023, 0.0022, 0.0020, 0.0026, 0.0025], device='cuda:3'), out_proj_covar=tensor([3.1453e-05, 2.6597e-05, 1.7492e-05, 1.8070e-05, 1.8830e-05, 1.7182e-05, 2.5211e-05, 2.1818e-05], device='cuda:3') 2022-12-07 05:11:14,299 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2746.0, num_to_drop=2, layers_to_drop={0, 3} 2022-12-07 05:11:15,364 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.93 vs. limit=2.0 2022-12-07 05:11:54,662 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7915, 1.5767, 2.0633, 1.8023, 1.7391, 1.7590, 1.2735, 1.7068], device='cuda:3'), covar=tensor([0.0213, 0.0471, 0.0147, 0.0275, 0.0232, 0.0271, 0.0551, 0.0278], device='cuda:3'), in_proj_covar=tensor([0.0029, 0.0040, 0.0027, 0.0029, 0.0032, 0.0028, 0.0027, 0.0026], device='cuda:3'), out_proj_covar=tensor([2.7945e-05, 4.0720e-05, 2.8324e-05, 2.6315e-05, 3.0934e-05, 2.6658e-05, 2.7434e-05, 2.5794e-05], device='cuda:3') 2022-12-07 05:11:56,166 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2022-12-07 05:12:00,034 INFO [train.py:873] (3/4) Epoch 1, batch 2800, loss[loss=0.336, simple_loss=0.2992, pruned_loss=0.1864, over 14065.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.277, pruned_loss=0.177, over 2015857.25 frames. ], batch size: 29, lr: 4.67e-02, grad_scale: 8.0 2022-12-07 05:12:05,006 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.030e+02 3.640e+02 4.591e+02 6.544e+02 1.997e+03, threshold=9.181e+02, percent-clipped=12.0 2022-12-07 05:12:18,775 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2824.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 05:12:30,802 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8985, 1.6919, 2.1466, 1.8888, 1.6924, 1.9283, 1.1991, 1.8561], device='cuda:3'), covar=tensor([0.0144, 0.0400, 0.0088, 0.0241, 0.0207, 0.0146, 0.0588, 0.0205], device='cuda:3'), in_proj_covar=tensor([0.0030, 0.0041, 0.0027, 0.0030, 0.0034, 0.0028, 0.0028, 0.0027], device='cuda:3'), out_proj_covar=tensor([2.8365e-05, 4.2919e-05, 2.8545e-05, 2.8487e-05, 3.2927e-05, 2.6861e-05, 2.8983e-05, 2.7558e-05], device='cuda:3') 2022-12-07 05:13:09,599 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2885.0, num_to_drop=2, layers_to_drop={1, 3} 2022-12-07 05:13:22,398 INFO [train.py:873] (3/4) Epoch 1, batch 2900, loss[loss=0.3512, simple_loss=0.2978, pruned_loss=0.2023, over 13861.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.2764, pruned_loss=0.176, over 2022826.20 frames. ], batch size: 20, lr: 4.65e-02, grad_scale: 8.0 2022-12-07 05:13:27,202 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.262e+02 3.059e+02 4.261e+02 6.986e+02 2.302e+03, threshold=8.522e+02, percent-clipped=12.0 2022-12-07 05:13:48,129 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8227, 2.7669, 2.6148, 2.8846, 2.8697, 2.7207, 2.5393, 2.6178], device='cuda:3'), covar=tensor([0.0277, 0.0357, 0.0387, 0.0296, 0.0295, 0.0371, 0.0403, 0.0350], device='cuda:3'), in_proj_covar=tensor([0.0033, 0.0048, 0.0038, 0.0037, 0.0037, 0.0040, 0.0044, 0.0043], device='cuda:3'), out_proj_covar=tensor([3.8349e-05, 5.2794e-05, 3.9091e-05, 4.0506e-05, 3.9796e-05, 4.6792e-05, 4.8029e-05, 4.8029e-05], device='cuda:3') 2022-12-07 05:13:52,944 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2937.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 05:14:30,587 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.12 vs. limit=2.0 2022-12-07 05:14:43,379 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2996.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 05:14:45,384 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2998.0, num_to_drop=2, layers_to_drop={1, 3} 2022-12-07 05:14:46,533 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.98 vs. limit=2.0 2022-12-07 05:14:48,187 INFO [train.py:873] (3/4) Epoch 1, batch 3000, loss[loss=0.2803, simple_loss=0.2576, pruned_loss=0.1515, over 13994.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.2742, pruned_loss=0.1734, over 2044279.67 frames. ], batch size: 22, lr: 4.63e-02, grad_scale: 8.0 2022-12-07 05:14:48,187 INFO [train.py:896] (3/4) Computing validation loss 2022-12-07 05:14:56,536 INFO [train.py:905] (3/4) Epoch 1, validation: loss=0.2054, simple_loss=0.2366, pruned_loss=0.08706, over 857387.00 frames. 2022-12-07 05:14:56,537 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17131MB 2022-12-07 05:15:01,308 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.332e+02 3.340e+02 4.389e+02 5.804e+02 1.068e+03, threshold=8.777e+02, percent-clipped=8.0 2022-12-07 05:15:12,245 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3019.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 05:15:30,352 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3041.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:15:32,918 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3044.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 05:15:33,333 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.27 vs. limit=2.0 2022-12-07 05:16:04,060 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3080.0, num_to_drop=2, layers_to_drop={2, 3} 2022-12-07 05:16:21,383 INFO [train.py:873] (3/4) Epoch 1, batch 3100, loss[loss=0.3073, simple_loss=0.2404, pruned_loss=0.1871, over 1197.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.2726, pruned_loss=0.1713, over 2027272.24 frames. ], batch size: 100, lr: 4.61e-02, grad_scale: 8.0 2022-12-07 05:16:26,604 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.583e+02 3.492e+02 4.643e+02 6.451e+02 1.266e+03, threshold=9.287e+02, percent-clipped=10.0 2022-12-07 05:16:48,129 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7035, 1.9614, 3.8565, 3.6118, 3.4758, 3.7253, 2.1473, 3.7839], device='cuda:3'), covar=tensor([0.0159, 0.1367, 0.0197, 0.0248, 0.0236, 0.0171, 0.0714, 0.0191], device='cuda:3'), in_proj_covar=tensor([0.0031, 0.0047, 0.0028, 0.0033, 0.0036, 0.0031, 0.0030, 0.0029], device='cuda:3'), out_proj_covar=tensor([3.1433e-05, 5.0891e-05, 3.1599e-05, 3.3830e-05, 3.5781e-05, 3.1346e-05, 3.2946e-05, 3.0748e-05], device='cuda:3') 2022-12-07 05:16:48,248 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4512, 2.4361, 3.7119, 2.9036, 3.4583, 3.3387, 2.7859, 2.4614], device='cuda:3'), covar=tensor([0.0390, 0.3742, 0.0278, 0.0905, 0.0462, 0.0468, 0.0535, 0.3419], device='cuda:3'), in_proj_covar=tensor([0.0053, 0.0092, 0.0040, 0.0051, 0.0045, 0.0048, 0.0037, 0.0096], device='cuda:3'), out_proj_covar=tensor([3.0379e-05, 8.1471e-05, 2.2806e-05, 3.3723e-05, 2.6950e-05, 2.8252e-05, 2.3277e-05, 7.6814e-05], device='cuda:3') 2022-12-07 05:17:02,075 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.37 vs. limit=2.0 2022-12-07 05:17:08,731 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.03 vs. limit=5.0 2022-12-07 05:17:12,856 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.0356, 4.9831, 4.8342, 5.0889, 5.3359, 4.0438, 5.3529, 5.1434], device='cuda:3'), covar=tensor([0.0590, 0.0629, 0.0584, 0.0608, 0.0397, 0.0719, 0.0446, 0.0730], device='cuda:3'), in_proj_covar=tensor([0.0061, 0.0051, 0.0071, 0.0060, 0.0066, 0.0048, 0.0055, 0.0066], device='cuda:3'), out_proj_covar=tensor([7.3778e-05, 6.0664e-05, 8.1198e-05, 7.0587e-05, 7.2830e-05, 5.4629e-05, 7.4826e-05, 8.0637e-05], device='cuda:3') 2022-12-07 05:17:27,812 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3180.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:17:45,613 INFO [train.py:873] (3/4) Epoch 1, batch 3200, loss[loss=0.3351, simple_loss=0.2796, pruned_loss=0.1953, over 8606.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.2727, pruned_loss=0.1717, over 2004140.73 frames. ], batch size: 100, lr: 4.59e-02, grad_scale: 8.0 2022-12-07 05:17:50,388 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.484e+02 3.868e+02 5.118e+02 7.409e+02 2.192e+03, threshold=1.024e+03, percent-clipped=11.0 2022-12-07 05:18:36,237 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.97 vs. limit=2.0 2022-12-07 05:18:57,696 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3288.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 05:18:59,989 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1814, 1.7674, 2.6137, 2.1927, 2.4649, 2.4804, 1.4247, 1.8029], device='cuda:3'), covar=tensor([0.0358, 0.0491, 0.0300, 0.0249, 0.0276, 0.0220, 0.1478, 0.0786], device='cuda:3'), in_proj_covar=tensor([0.0035, 0.0033, 0.0035, 0.0034, 0.0033, 0.0031, 0.0049, 0.0038], device='cuda:3'), out_proj_covar=tensor([2.8252e-05, 3.0666e-05, 2.9774e-05, 2.5944e-05, 2.8287e-05, 2.4103e-05, 5.0374e-05, 3.5729e-05], device='cuda:3') 2022-12-07 05:19:01,652 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0569, 2.0278, 1.9847, 2.0751, 2.0344, 2.0213, 1.9530, 1.8810], device='cuda:3'), covar=tensor([0.0311, 0.0475, 0.0445, 0.0390, 0.0373, 0.0450, 0.0502, 0.0500], device='cuda:3'), in_proj_covar=tensor([0.0034, 0.0053, 0.0041, 0.0040, 0.0040, 0.0043, 0.0050, 0.0048], device='cuda:3'), out_proj_covar=tensor([4.2829e-05, 6.2222e-05, 4.5622e-05, 4.7377e-05, 4.5104e-05, 5.2806e-05, 5.9981e-05, 5.7572e-05], device='cuda:3') 2022-12-07 05:19:01,657 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3293.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:19:08,886 INFO [train.py:873] (3/4) Epoch 1, batch 3300, loss[loss=0.3547, simple_loss=0.2979, pruned_loss=0.2058, over 14127.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.27, pruned_loss=0.1684, over 2016095.40 frames. ], batch size: 99, lr: 4.57e-02, grad_scale: 8.0 2022-12-07 05:19:14,104 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.592e+02 2.921e+02 3.689e+02 4.978e+02 9.002e+02, threshold=7.379e+02, percent-clipped=0.0 2022-12-07 05:19:21,044 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.03 vs. limit=5.0 2022-12-07 05:19:21,978 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.14 vs. limit=2.0 2022-12-07 05:19:23,457 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.22 vs. limit=5.0 2022-12-07 05:19:42,816 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3341.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:19:49,922 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3349.0, num_to_drop=2, layers_to_drop={1, 3} 2022-12-07 05:20:05,299 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.91 vs. limit=2.0 2022-12-07 05:20:11,317 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3375.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:20:24,037 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3389.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:20:30,013 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4139, 3.8919, 3.9961, 4.3433, 4.2409, 3.1670, 4.6449, 4.3103], device='cuda:3'), covar=tensor([0.0595, 0.0841, 0.0794, 0.0638, 0.0569, 0.0840, 0.0465, 0.0942], device='cuda:3'), in_proj_covar=tensor([0.0061, 0.0054, 0.0073, 0.0063, 0.0068, 0.0050, 0.0057, 0.0067], device='cuda:3'), out_proj_covar=tensor([7.3494e-05, 6.5553e-05, 8.6668e-05, 7.6635e-05, 7.7188e-05, 5.8191e-05, 7.7934e-05, 8.3506e-05], device='cuda:3') 2022-12-07 05:20:34,759 INFO [train.py:873] (3/4) Epoch 1, batch 3400, loss[loss=0.3042, simple_loss=0.2601, pruned_loss=0.1741, over 5994.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.2694, pruned_loss=0.1682, over 1984727.25 frames. ], batch size: 100, lr: 4.55e-02, grad_scale: 8.0 2022-12-07 05:20:39,711 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.829e+02 3.330e+02 4.669e+02 8.149e+02 2.778e+03, threshold=9.337e+02, percent-clipped=27.0 2022-12-07 05:21:36,643 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.10 vs. limit=2.0 2022-12-07 05:21:43,573 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3480.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:22:01,457 INFO [train.py:873] (3/4) Epoch 1, batch 3500, loss[loss=0.273, simple_loss=0.2407, pruned_loss=0.1526, over 6020.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.2675, pruned_loss=0.1653, over 2004016.15 frames. ], batch size: 100, lr: 4.53e-02, grad_scale: 8.0 2022-12-07 05:22:06,643 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.311e+02 3.200e+02 4.145e+02 5.923e+02 9.666e+02, threshold=8.289e+02, percent-clipped=1.0 2022-12-07 05:22:24,422 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3528.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:22:40,193 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3546.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 05:23:21,087 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3593.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:23:28,041 INFO [train.py:873] (3/4) Epoch 1, batch 3600, loss[loss=0.2928, simple_loss=0.2711, pruned_loss=0.1572, over 14409.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.2664, pruned_loss=0.164, over 1959739.70 frames. ], batch size: 41, lr: 4.50e-02, grad_scale: 8.0 2022-12-07 05:23:28,289 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0944, 2.1955, 3.0785, 2.5595, 3.1777, 2.6092, 2.9046, 2.5788], device='cuda:3'), covar=tensor([0.0302, 0.2415, 0.0214, 0.0650, 0.0382, 0.0481, 0.0451, 0.1881], device='cuda:3'), in_proj_covar=tensor([0.0060, 0.0103, 0.0048, 0.0056, 0.0053, 0.0054, 0.0044, 0.0108], device='cuda:3'), out_proj_covar=tensor([3.3995e-05, 8.4297e-05, 2.6292e-05, 3.5044e-05, 3.1577e-05, 3.1445e-05, 2.7013e-05, 8.1056e-05], device='cuda:3') 2022-12-07 05:23:33,317 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.536e+02 3.769e+02 4.638e+02 5.846e+02 1.150e+03, threshold=9.276e+02, percent-clipped=6.0 2022-12-07 05:23:33,597 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3607.0, num_to_drop=2, layers_to_drop={1, 3} 2022-12-07 05:23:54,497 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6561, 1.7944, 2.1960, 2.3890, 2.1161, 1.8118, 1.6844, 2.0104], device='cuda:3'), covar=tensor([0.2973, 0.1608, 0.2299, 0.0658, 0.0527, 0.0991, 0.0780, 0.0671], device='cuda:3'), in_proj_covar=tensor([0.0062, 0.0043, 0.0030, 0.0030, 0.0033, 0.0028, 0.0035, 0.0035], device='cuda:3'), out_proj_covar=tensor([6.7470e-05, 4.5874e-05, 3.2404e-05, 3.1698e-05, 3.1739e-05, 3.0456e-05, 3.9399e-05, 3.5066e-05], device='cuda:3') 2022-12-07 05:24:02,791 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3641.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:24:05,399 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3644.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 05:24:33,193 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3675.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:24:34,060 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3676.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 05:24:55,837 INFO [train.py:873] (3/4) Epoch 1, batch 3700, loss[loss=0.2877, simple_loss=0.2603, pruned_loss=0.1576, over 13539.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.2644, pruned_loss=0.1617, over 2001876.23 frames. ], batch size: 100, lr: 4.48e-02, grad_scale: 8.0 2022-12-07 05:25:00,988 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.789e+02 3.561e+02 4.722e+02 6.251e+02 1.502e+03, threshold=9.443e+02, percent-clipped=7.0 2022-12-07 05:25:14,822 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3723.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:25:25,051 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8278, 1.8420, 2.4757, 2.4350, 2.2739, 2.2074, 1.6232, 2.5942], device='cuda:3'), covar=tensor([0.0882, 0.0559, 0.0164, 0.0191, 0.0173, 0.0185, 0.0458, 0.0123], device='cuda:3'), in_proj_covar=tensor([0.0054, 0.0046, 0.0028, 0.0032, 0.0033, 0.0030, 0.0030, 0.0029], device='cuda:3'), out_proj_covar=tensor([6.3637e-05, 5.2625e-05, 3.0210e-05, 3.7654e-05, 3.3566e-05, 3.1644e-05, 3.5300e-05, 2.9692e-05], device='cuda:3') 2022-12-07 05:25:26,813 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3737.0, num_to_drop=2, layers_to_drop={0, 2} 2022-12-07 05:25:33,583 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.2478, 0.9836, 1.0321, 1.0250, 1.2299, 0.8276, 1.1131, 1.0259], device='cuda:3'), covar=tensor([0.0130, 0.0205, 0.0166, 0.0220, 0.0163, 0.0234, 0.0279, 0.0153], device='cuda:3'), in_proj_covar=tensor([0.0023, 0.0028, 0.0032, 0.0024, 0.0024, 0.0032, 0.0025, 0.0026], device='cuda:3'), out_proj_covar=tensor([2.4726e-05, 2.9058e-05, 3.0824e-05, 2.5028e-05, 2.0951e-05, 3.0125e-05, 2.3333e-05, 2.4073e-05], device='cuda:3') 2022-12-07 05:26:22,262 INFO [train.py:873] (3/4) Epoch 1, batch 3800, loss[loss=0.2798, simple_loss=0.2269, pruned_loss=0.1663, over 3883.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.2643, pruned_loss=0.1622, over 1887380.35 frames. ], batch size: 100, lr: 4.46e-02, grad_scale: 8.0 2022-12-07 05:26:23,241 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.6222, 4.2922, 4.5601, 4.1255, 4.4444, 4.4391, 2.9722, 4.3846], device='cuda:3'), covar=tensor([0.0188, 0.0237, 0.0415, 0.0362, 0.0286, 0.0148, 0.1653, 0.0346], device='cuda:3'), in_proj_covar=tensor([0.0049, 0.0050, 0.0051, 0.0045, 0.0058, 0.0039, 0.0071, 0.0057], device='cuda:3'), out_proj_covar=tensor([6.3190e-05, 6.8767e-05, 7.0542e-05, 5.8299e-05, 7.8255e-05, 5.0197e-05, 9.3317e-05, 7.5702e-05], device='cuda:3') 2022-12-07 05:26:23,770 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.32 vs. limit=5.0 2022-12-07 05:26:26,271 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.19 vs. limit=5.0 2022-12-07 05:26:27,325 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.307e+02 3.207e+02 4.617e+02 6.095e+02 1.543e+03, threshold=9.233e+02, percent-clipped=5.0 2022-12-07 05:26:53,340 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2022-12-07 05:27:50,296 INFO [train.py:873] (3/4) Epoch 1, batch 3900, loss[loss=0.2684, simple_loss=0.2519, pruned_loss=0.1424, over 14642.00 frames. ], tot_loss[loss=0.2925, simple_loss=0.264, pruned_loss=0.1605, over 2031026.81 frames. ], batch size: 33, lr: 4.44e-02, grad_scale: 8.0 2022-12-07 05:27:51,215 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3902.0, num_to_drop=1, layers_to_drop={3} 2022-12-07 05:27:55,147 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.731e+02 3.646e+02 5.236e+02 6.626e+02 1.873e+03, threshold=1.047e+03, percent-clipped=8.0 2022-12-07 05:27:56,474 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.05 vs. limit=2.0 2022-12-07 05:28:11,849 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3926.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:28:27,533 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3944.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 05:29:05,227 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3987.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:29:09,493 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3992.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 05:29:17,377 INFO [train.py:873] (3/4) Epoch 1, batch 4000, loss[loss=0.2507, simple_loss=0.2422, pruned_loss=0.1296, over 13929.00 frames. ], tot_loss[loss=0.2898, simple_loss=0.2624, pruned_loss=0.1586, over 2083465.35 frames. ], batch size: 26, lr: 4.42e-02, grad_scale: 8.0 2022-12-07 05:29:22,734 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.698e+02 3.697e+02 5.328e+02 7.328e+02 1.359e+03, threshold=1.066e+03, percent-clipped=6.0 2022-12-07 05:29:43,246 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0335, 1.6346, 2.1155, 1.8502, 2.2624, 1.8692, 2.1049, 2.2011], device='cuda:3'), covar=tensor([0.0303, 0.1635, 0.0287, 0.1981, 0.0234, 0.1277, 0.0306, 0.0262], device='cuda:3'), in_proj_covar=tensor([0.0116, 0.0204, 0.0114, 0.0227, 0.0083, 0.0225, 0.0120, 0.0094], device='cuda:3'), out_proj_covar=tensor([7.9876e-05, 1.6071e-04, 8.0508e-05, 1.8288e-04, 5.8583e-05, 1.7374e-04, 8.3747e-05, 6.4875e-05], device='cuda:3') 2022-12-07 05:29:45,739 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4032.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 05:30:26,761 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9674, 3.7359, 3.7654, 3.9756, 3.8691, 2.7555, 4.1965, 3.9492], device='cuda:3'), covar=tensor([0.0749, 0.0835, 0.0760, 0.0821, 0.0801, 0.1061, 0.0651, 0.1316], device='cuda:3'), in_proj_covar=tensor([0.0069, 0.0055, 0.0077, 0.0072, 0.0076, 0.0051, 0.0070, 0.0076], device='cuda:3'), out_proj_covar=tensor([8.7887e-05, 7.2297e-05, 9.7242e-05, 9.4024e-05, 9.4979e-05, 6.5570e-05, 9.6519e-05, 1.0005e-04], device='cuda:3') 2022-12-07 05:30:47,314 INFO [train.py:873] (3/4) Epoch 1, batch 4100, loss[loss=0.2516, simple_loss=0.2438, pruned_loss=0.1297, over 14272.00 frames. ], tot_loss[loss=0.2882, simple_loss=0.2615, pruned_loss=0.1575, over 2103744.80 frames. ], batch size: 69, lr: 4.40e-02, grad_scale: 8.0 2022-12-07 05:30:52,474 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.129e+02 3.295e+02 4.605e+02 5.758e+02 1.082e+03, threshold=9.210e+02, percent-clipped=1.0 2022-12-07 05:31:13,995 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.55 vs. limit=5.0 2022-12-07 05:31:22,891 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=4141.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:32:16,951 INFO [train.py:873] (3/4) Epoch 1, batch 4200, loss[loss=0.2648, simple_loss=0.2505, pruned_loss=0.1395, over 14281.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.2611, pruned_loss=0.1573, over 2045526.74 frames. ], batch size: 25, lr: 4.38e-02, grad_scale: 8.0 2022-12-07 05:32:17,789 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4202.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 05:32:17,839 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=4202.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:32:21,867 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.012e+02 3.430e+02 4.467e+02 5.944e+02 1.498e+03, threshold=8.934e+02, percent-clipped=5.0 2022-12-07 05:33:00,181 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4250.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 05:33:28,853 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4282.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:33:34,019 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=4288.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:33:45,948 INFO [train.py:873] (3/4) Epoch 1, batch 4300, loss[loss=0.2623, simple_loss=0.2143, pruned_loss=0.1552, over 2635.00 frames. ], tot_loss[loss=0.2856, simple_loss=0.2589, pruned_loss=0.1561, over 1946820.78 frames. ], batch size: 100, lr: 4.35e-02, grad_scale: 8.0 2022-12-07 05:33:51,476 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.040e+01 3.141e+02 3.918e+02 5.765e+02 1.012e+03, threshold=7.835e+02, percent-clipped=6.0 2022-12-07 05:33:53,458 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6459, 1.6796, 1.0825, 2.0670, 1.5709, 1.5249, 1.5142, 1.1037], device='cuda:3'), covar=tensor([0.0151, 0.0144, 0.0291, 0.0102, 0.0189, 0.0128, 0.0138, 0.0283], device='cuda:3'), in_proj_covar=tensor([0.0013, 0.0015, 0.0017, 0.0014, 0.0015, 0.0015, 0.0013, 0.0014], device='cuda:3'), out_proj_covar=tensor([1.2585e-05, 1.3400e-05, 1.8076e-05, 1.3447e-05, 1.4442e-05, 1.3998e-05, 1.2528e-05, 1.5119e-05], device='cuda:3') 2022-12-07 05:34:13,550 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4332.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 05:34:21,174 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7510, 3.2563, 3.8728, 3.0578, 4.1988, 3.5844, 3.3208, 3.0094], device='cuda:3'), covar=tensor([0.0279, 0.1522, 0.0149, 0.0718, 0.0182, 0.0391, 0.0471, 0.2081], device='cuda:3'), in_proj_covar=tensor([0.0059, 0.0101, 0.0045, 0.0058, 0.0047, 0.0055, 0.0047, 0.0110], device='cuda:3'), out_proj_covar=tensor([3.7090e-05, 7.6552e-05, 2.6159e-05, 3.7857e-05, 2.7910e-05, 3.3968e-05, 3.2403e-05, 8.0254e-05], device='cuda:3') 2022-12-07 05:34:29,154 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=4349.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:34:39,384 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0037, 1.9994, 1.9255, 2.0168, 1.9465, 1.9412, 1.4917, 1.7730], device='cuda:3'), covar=tensor([0.0272, 0.0245, 0.0407, 0.0253, 0.0295, 0.0215, 0.1082, 0.0359], device='cuda:3'), in_proj_covar=tensor([0.0054, 0.0055, 0.0055, 0.0049, 0.0066, 0.0042, 0.0080, 0.0062], device='cuda:3'), out_proj_covar=tensor([7.6209e-05, 8.4240e-05, 8.1387e-05, 7.0379e-05, 9.9029e-05, 5.8753e-05, 1.1046e-04, 8.5020e-05], device='cuda:3') 2022-12-07 05:34:44,677 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1732, 1.2246, 1.0583, 1.3210, 1.2764, 1.0806, 1.3800, 1.0184], device='cuda:3'), covar=tensor([0.0122, 0.0133, 0.0198, 0.0096, 0.0166, 0.0140, 0.0150, 0.0167], device='cuda:3'), in_proj_covar=tensor([0.0014, 0.0015, 0.0017, 0.0015, 0.0015, 0.0016, 0.0014, 0.0015], device='cuda:3'), out_proj_covar=tensor([1.3343e-05, 1.3798e-05, 1.8475e-05, 1.3825e-05, 1.5540e-05, 1.4828e-05, 1.3311e-05, 1.5600e-05], device='cuda:3') 2022-12-07 05:34:56,526 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4380.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 05:35:15,297 INFO [train.py:873] (3/4) Epoch 1, batch 4400, loss[loss=0.2566, simple_loss=0.2415, pruned_loss=0.1359, over 14338.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.259, pruned_loss=0.1562, over 1963944.54 frames. ], batch size: 39, lr: 4.33e-02, grad_scale: 8.0 2022-12-07 05:35:20,267 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.063e+02 3.285e+02 4.190e+02 5.766e+02 1.366e+03, threshold=8.380e+02, percent-clipped=8.0 2022-12-07 05:35:29,352 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.84 vs. limit=5.0 2022-12-07 05:36:40,776 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4497.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:36:44,320 INFO [train.py:873] (3/4) Epoch 1, batch 4500, loss[loss=0.3267, simple_loss=0.2752, pruned_loss=0.1891, over 7771.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.2587, pruned_loss=0.1548, over 2026557.04 frames. ], batch size: 100, lr: 4.31e-02, grad_scale: 8.0 2022-12-07 05:36:49,368 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.519e+02 3.529e+02 4.865e+02 6.809e+02 1.533e+03, threshold=9.730e+02, percent-clipped=12.0 2022-12-07 05:37:06,241 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2832, 2.9535, 2.9673, 3.0020, 3.2572, 3.2359, 3.1916, 3.0601], device='cuda:3'), covar=tensor([0.0687, 0.0900, 0.0909, 0.0824, 0.0636, 0.0719, 0.0898, 0.1105], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0070, 0.0085, 0.0057, 0.0079, 0.0073, 0.0075, 0.0067], device='cuda:3'), out_proj_covar=tensor([8.8078e-05, 8.3258e-05, 9.4385e-05, 6.5285e-05, 9.1181e-05, 8.2254e-05, 9.6031e-05, 8.2947e-05], device='cuda:3') 2022-12-07 05:37:48,463 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8404, 2.0952, 3.6202, 3.3506, 3.3246, 3.3384, 2.3800, 3.6711], device='cuda:3'), covar=tensor([0.1367, 0.1039, 0.0103, 0.0128, 0.0154, 0.0170, 0.0450, 0.0109], device='cuda:3'), in_proj_covar=tensor([0.0070, 0.0067, 0.0035, 0.0043, 0.0039, 0.0041, 0.0037, 0.0036], device='cuda:3'), out_proj_covar=tensor([8.9877e-05, 8.4136e-05, 4.3135e-05, 6.1279e-05, 4.5103e-05, 5.1017e-05, 4.8716e-05, 4.2997e-05], device='cuda:3') 2022-12-07 05:37:54,679 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4582.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:37:59,890 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1048, 2.0499, 2.0124, 1.8598, 1.9892, 1.9745, 1.9724, 1.9678], device='cuda:3'), covar=tensor([0.0254, 0.0286, 0.0354, 0.0406, 0.0406, 0.0304, 0.0426, 0.0383], device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0072, 0.0085, 0.0058, 0.0081, 0.0076, 0.0078, 0.0067], device='cuda:3'), out_proj_covar=tensor([8.9494e-05, 8.5488e-05, 9.5183e-05, 6.7267e-05, 9.5299e-05, 8.5695e-05, 1.0030e-04, 8.2051e-05], device='cuda:3') 2022-12-07 05:38:00,389 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=3.27 vs. limit=2.0 2022-12-07 05:38:11,656 INFO [train.py:873] (3/4) Epoch 1, batch 4600, loss[loss=0.2771, simple_loss=0.2573, pruned_loss=0.1485, over 14179.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.2588, pruned_loss=0.1549, over 1999461.09 frames. ], batch size: 99, lr: 4.29e-02, grad_scale: 8.0 2022-12-07 05:38:16,963 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.290e+02 3.319e+02 4.792e+02 5.771e+02 3.822e+03, threshold=9.585e+02, percent-clipped=7.0 2022-12-07 05:38:37,724 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4630.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:38:37,821 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8907, 1.6078, 2.6884, 2.1547, 2.4736, 2.2054, 1.3322, 1.7805], device='cuda:3'), covar=tensor([0.0430, 0.0546, 0.0184, 0.0213, 0.0190, 0.0223, 0.1155, 0.0552], device='cuda:3'), in_proj_covar=tensor([0.0041, 0.0039, 0.0041, 0.0039, 0.0040, 0.0037, 0.0062, 0.0041], device='cuda:3'), out_proj_covar=tensor([4.0425e-05, 3.7509e-05, 4.2179e-05, 3.3270e-05, 3.6503e-05, 3.1971e-05, 6.6831e-05, 4.2819e-05], device='cuda:3') 2022-12-07 05:38:39,583 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.0169, 0.8563, 1.0327, 0.8904, 1.1741, 0.6874, 0.9420, 1.0826], device='cuda:3'), covar=tensor([0.0114, 0.0753, 0.0216, 0.0388, 0.0209, 0.0552, 0.0224, 0.0128], device='cuda:3'), in_proj_covar=tensor([0.0033, 0.0087, 0.0043, 0.0060, 0.0038, 0.0045, 0.0038, 0.0045], device='cuda:3'), out_proj_covar=tensor([3.5818e-05, 9.3937e-05, 4.7550e-05, 6.4855e-05, 4.4725e-05, 5.2923e-05, 4.9726e-05, 4.8544e-05], device='cuda:3') 2022-12-07 05:38:42,635 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.71 vs. limit=2.0 2022-12-07 05:38:49,788 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4644.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:39:09,340 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.95 vs. limit=5.0 2022-12-07 05:39:10,549 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2764, 1.4702, 2.6130, 1.9916, 2.2442, 2.1674, 1.5606, 2.3996], device='cuda:3'), covar=tensor([0.0189, 0.1548, 0.0219, 0.0794, 0.0233, 0.0301, 0.0936, 0.0249], device='cuda:3'), in_proj_covar=tensor([0.0043, 0.0075, 0.0039, 0.0062, 0.0053, 0.0044, 0.0039, 0.0042], device='cuda:3'), out_proj_covar=tensor([5.5980e-05, 9.6906e-05, 5.5534e-05, 8.3405e-05, 7.1074e-05, 5.9343e-05, 5.6278e-05, 5.7431e-05], device='cuda:3') 2022-12-07 05:39:41,708 INFO [train.py:873] (3/4) Epoch 1, batch 4700, loss[loss=0.2914, simple_loss=0.2556, pruned_loss=0.1636, over 5956.00 frames. ], tot_loss[loss=0.282, simple_loss=0.257, pruned_loss=0.1535, over 1972544.56 frames. ], batch size: 100, lr: 4.27e-02, grad_scale: 8.0 2022-12-07 05:39:46,822 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.177e+02 3.103e+02 4.134e+02 5.921e+02 2.901e+03, threshold=8.268e+02, percent-clipped=6.0 2022-12-07 05:39:55,997 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3755, 3.8611, 4.7418, 3.9551, 4.4873, 4.3847, 2.7977, 4.2590], device='cuda:3'), covar=tensor([0.0229, 0.0398, 0.0304, 0.0473, 0.0223, 0.0150, 0.1650, 0.0231], device='cuda:3'), in_proj_covar=tensor([0.0059, 0.0057, 0.0061, 0.0054, 0.0071, 0.0045, 0.0088, 0.0068], device='cuda:3'), out_proj_covar=tensor([8.5378e-05, 9.1263e-05, 9.1717e-05, 8.2265e-05, 1.0815e-04, 6.8062e-05, 1.2503e-04, 9.6898e-05], device='cuda:3') 2022-12-07 05:40:01,556 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.07 vs. limit=2.0 2022-12-07 05:40:26,867 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.51 vs. limit=5.0 2022-12-07 05:41:06,516 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4797.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:41:10,381 INFO [train.py:873] (3/4) Epoch 1, batch 4800, loss[loss=0.215, simple_loss=0.2084, pruned_loss=0.1108, over 14299.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.2563, pruned_loss=0.1527, over 1968116.39 frames. ], batch size: 18, lr: 4.25e-02, grad_scale: 16.0 2022-12-07 05:41:15,575 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.267e+02 3.055e+02 4.514e+02 6.112e+02 1.452e+03, threshold=9.028e+02, percent-clipped=8.0 2022-12-07 05:41:49,187 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4845.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:42:07,638 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.91 vs. limit=2.0 2022-12-07 05:42:39,197 INFO [train.py:873] (3/4) Epoch 1, batch 4900, loss[loss=0.2878, simple_loss=0.2502, pruned_loss=0.1627, over 3849.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.2573, pruned_loss=0.1529, over 2018970.17 frames. ], batch size: 100, lr: 4.23e-02, grad_scale: 16.0 2022-12-07 05:42:44,141 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.236e+02 3.107e+02 4.475e+02 6.071e+02 1.419e+03, threshold=8.951e+02, percent-clipped=8.0 2022-12-07 05:42:57,897 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.00 vs. limit=2.0 2022-12-07 05:43:03,730 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.18 vs. limit=2.0 2022-12-07 05:43:16,620 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4944.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:43:58,603 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4992.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:44:10,021 INFO [train.py:873] (3/4) Epoch 1, batch 5000, loss[loss=0.2166, simple_loss=0.2234, pruned_loss=0.1048, over 14271.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.2556, pruned_loss=0.1516, over 1916175.76 frames. ], batch size: 31, lr: 4.20e-02, grad_scale: 16.0 2022-12-07 05:44:12,680 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.4140, 4.8791, 4.8134, 5.2801, 5.1898, 4.9956, 5.2580, 4.9443], device='cuda:3'), covar=tensor([0.0120, 0.0353, 0.0216, 0.0268, 0.0173, 0.0264, 0.0301, 0.0235], device='cuda:3'), in_proj_covar=tensor([0.0044, 0.0068, 0.0056, 0.0052, 0.0052, 0.0061, 0.0073, 0.0068], device='cuda:3'), out_proj_covar=tensor([6.6126e-05, 9.3188e-05, 7.5732e-05, 7.7852e-05, 7.3264e-05, 9.1235e-05, 1.1679e-04, 1.0005e-04], device='cuda:3') 2022-12-07 05:44:15,206 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.545e+02 3.215e+02 4.595e+02 5.686e+02 1.097e+03, threshold=9.191e+02, percent-clipped=3.0 2022-12-07 05:44:45,480 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.46 vs. limit=2.0 2022-12-07 05:45:22,731 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.44 vs. limit=2.0 2022-12-07 05:45:39,006 INFO [train.py:873] (3/4) Epoch 1, batch 5100, loss[loss=0.2045, simple_loss=0.1702, pruned_loss=0.1194, over 1256.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.2542, pruned_loss=0.15, over 1870298.00 frames. ], batch size: 100, lr: 4.18e-02, grad_scale: 16.0 2022-12-07 05:45:43,999 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.500e+02 3.085e+02 4.030e+02 5.035e+02 8.863e+02, threshold=8.060e+02, percent-clipped=0.0 2022-12-07 05:45:51,769 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.10 vs. limit=2.0 2022-12-07 05:46:18,224 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=3.03 vs. limit=2.0 2022-12-07 05:46:30,431 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5700, 2.4325, 2.1505, 2.9319, 2.6855, 2.7871, 2.5180, 1.9219], device='cuda:3'), covar=tensor([0.0361, 0.0856, 0.2369, 0.0358, 0.0177, 0.0301, 0.0777, 0.2934], device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0106, 0.0181, 0.0088, 0.0079, 0.0092, 0.0104, 0.0206], device='cuda:3'), out_proj_covar=tensor([5.4423e-05, 6.4460e-05, 1.1232e-04, 4.9490e-05, 4.4260e-05, 5.3189e-05, 6.2075e-05, 1.2673e-04], device='cuda:3') 2022-12-07 05:47:06,725 INFO [train.py:873] (3/4) Epoch 1, batch 5200, loss[loss=0.2969, simple_loss=0.2616, pruned_loss=0.1661, over 7726.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.2549, pruned_loss=0.1506, over 1900469.39 frames. ], batch size: 100, lr: 4.16e-02, grad_scale: 16.0 2022-12-07 05:47:11,952 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.189e+02 3.667e+02 4.771e+02 6.312e+02 1.162e+03, threshold=9.542e+02, percent-clipped=12.0 2022-12-07 05:47:20,563 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.66 vs. limit=5.0 2022-12-07 05:48:26,250 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.02 vs. limit=2.0 2022-12-07 05:48:35,674 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.37 vs. limit=5.0 2022-12-07 05:48:35,988 INFO [train.py:873] (3/4) Epoch 1, batch 5300, loss[loss=0.2407, simple_loss=0.2318, pruned_loss=0.1248, over 14139.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.2546, pruned_loss=0.15, over 1930279.51 frames. ], batch size: 84, lr: 4.14e-02, grad_scale: 16.0 2022-12-07 05:48:40,962 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.435e+02 3.215e+02 4.180e+02 5.101e+02 1.112e+03, threshold=8.360e+02, percent-clipped=0.0 2022-12-07 05:48:47,418 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2891, 1.6121, 3.3527, 2.3791, 3.2578, 3.1081, 2.1232, 3.6283], device='cuda:3'), covar=tensor([0.0168, 0.1789, 0.0196, 0.0992, 0.0224, 0.0229, 0.0593, 0.0148], device='cuda:3'), in_proj_covar=tensor([0.0050, 0.0085, 0.0044, 0.0078, 0.0063, 0.0050, 0.0047, 0.0051], device='cuda:3'), out_proj_covar=tensor([7.0764e-05, 1.1652e-04, 7.1538e-05, 1.1020e-04, 9.4628e-05, 7.3420e-05, 7.1457e-05, 7.7681e-05], device='cuda:3') 2022-12-07 05:48:55,173 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4074, 1.2011, 1.9705, 1.4163, 1.8383, 1.5445, 0.8781, 1.2274], device='cuda:3'), covar=tensor([0.0359, 0.0622, 0.0149, 0.0301, 0.0182, 0.0288, 0.0816, 0.0729], device='cuda:3'), in_proj_covar=tensor([0.0044, 0.0043, 0.0046, 0.0043, 0.0043, 0.0043, 0.0070, 0.0047], device='cuda:3'), out_proj_covar=tensor([4.8489e-05, 4.4814e-05, 4.9281e-05, 3.8295e-05, 4.0589e-05, 4.0489e-05, 7.7021e-05, 5.0388e-05], device='cuda:3') 2022-12-07 05:49:32,082 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.30 vs. limit=2.0 2022-12-07 05:49:41,659 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.13 vs. limit=2.0 2022-12-07 05:50:04,681 INFO [train.py:873] (3/4) Epoch 1, batch 5400, loss[loss=0.2497, simple_loss=0.2217, pruned_loss=0.1389, over 3876.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.2526, pruned_loss=0.1483, over 1891280.65 frames. ], batch size: 100, lr: 4.12e-02, grad_scale: 16.0 2022-12-07 05:50:09,835 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.361e+02 3.355e+02 4.207e+02 5.481e+02 1.330e+03, threshold=8.415e+02, percent-clipped=4.0 2022-12-07 05:50:43,053 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9236, 1.5400, 2.8039, 2.8444, 2.7873, 2.6092, 1.8555, 2.9590], device='cuda:3'), covar=tensor([0.1216, 0.1483, 0.0221, 0.0269, 0.0170, 0.0233, 0.0558, 0.0162], device='cuda:3'), in_proj_covar=tensor([0.0081, 0.0082, 0.0042, 0.0050, 0.0043, 0.0045, 0.0042, 0.0040], device='cuda:3'), out_proj_covar=tensor([1.1300e-04, 1.1070e-04, 5.6989e-05, 7.7572e-05, 5.4908e-05, 6.3596e-05, 6.3606e-05, 5.3299e-05], device='cuda:3') 2022-12-07 05:51:33,376 INFO [train.py:873] (3/4) Epoch 1, batch 5500, loss[loss=0.271, simple_loss=0.2537, pruned_loss=0.1442, over 14356.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.2521, pruned_loss=0.1468, over 1951696.61 frames. ], batch size: 55, lr: 4.10e-02, grad_scale: 16.0 2022-12-07 05:51:35,617 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.57 vs. limit=2.0 2022-12-07 05:51:38,436 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.070e+02 3.601e+02 4.745e+02 6.042e+02 1.360e+03, threshold=9.490e+02, percent-clipped=8.0 2022-12-07 05:51:57,362 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0438, 2.0227, 2.0000, 1.8851, 1.9744, 1.9646, 1.9564, 1.9899], device='cuda:3'), covar=tensor([0.0318, 0.0398, 0.0428, 0.0373, 0.0400, 0.0330, 0.0660, 0.0561], device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0078, 0.0099, 0.0068, 0.0086, 0.0081, 0.0090, 0.0080], device='cuda:3'), out_proj_covar=tensor([1.0515e-04, 9.7276e-05, 1.2029e-04, 8.2649e-05, 1.0397e-04, 9.4773e-05, 1.2734e-04, 1.0423e-04], device='cuda:3') 2022-12-07 05:52:17,650 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2022-12-07 05:52:31,793 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.47 vs. limit=5.0 2022-12-07 05:53:01,582 INFO [train.py:873] (3/4) Epoch 1, batch 5600, loss[loss=0.262, simple_loss=0.248, pruned_loss=0.138, over 14219.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.2526, pruned_loss=0.148, over 1931915.20 frames. ], batch size: 35, lr: 4.08e-02, grad_scale: 16.0 2022-12-07 05:53:06,470 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.507e+02 3.374e+02 4.886e+02 6.824e+02 1.449e+03, threshold=9.773e+02, percent-clipped=6.0 2022-12-07 05:53:15,109 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4938, 2.4560, 2.2943, 2.2859, 2.3800, 2.3467, 2.4239, 2.4291], device='cuda:3'), covar=tensor([0.0430, 0.0414, 0.0659, 0.0584, 0.0447, 0.0348, 0.0633, 0.0488], device='cuda:3'), in_proj_covar=tensor([0.0088, 0.0081, 0.0103, 0.0071, 0.0091, 0.0084, 0.0097, 0.0081], device='cuda:3'), out_proj_covar=tensor([1.0754e-04, 1.0164e-04, 1.2587e-04, 8.6557e-05, 1.1202e-04, 9.9822e-05, 1.3735e-04, 1.0598e-04], device='cuda:3') 2022-12-07 05:53:44,245 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=5650.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 05:54:29,451 INFO [train.py:873] (3/4) Epoch 1, batch 5700, loss[loss=0.2855, simple_loss=0.2644, pruned_loss=0.1533, over 12745.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.2518, pruned_loss=0.1475, over 1917952.32 frames. ], batch size: 100, lr: 4.06e-02, grad_scale: 16.0 2022-12-07 05:54:30,438 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8790, 1.1631, 1.4012, 1.0724, 1.5546, 1.0954, 1.2565, 1.8590], device='cuda:3'), covar=tensor([0.0202, 0.2444, 0.0382, 0.0944, 0.0455, 0.0928, 0.0768, 0.0320], device='cuda:3'), in_proj_covar=tensor([0.0035, 0.0096, 0.0043, 0.0067, 0.0041, 0.0047, 0.0044, 0.0043], device='cuda:3'), out_proj_covar=tensor([4.5780e-05, 1.1818e-04, 5.4520e-05, 8.4027e-05, 5.6315e-05, 6.4903e-05, 6.5825e-05, 5.4830e-05], device='cuda:3') 2022-12-07 05:54:34,444 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.548e+02 3.498e+02 4.626e+02 6.502e+02 1.133e+03, threshold=9.251e+02, percent-clipped=2.0 2022-12-07 05:54:38,201 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=5711.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 05:55:08,935 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5513, 3.2471, 3.3424, 3.5702, 3.5563, 3.6293, 3.4445, 3.1512], device='cuda:3'), covar=tensor([0.0223, 0.0621, 0.0319, 0.0381, 0.0330, 0.0342, 0.0506, 0.0442], device='cuda:3'), in_proj_covar=tensor([0.0050, 0.0077, 0.0060, 0.0056, 0.0058, 0.0067, 0.0080, 0.0075], device='cuda:3'), out_proj_covar=tensor([8.0525e-05, 1.0888e-04, 8.5725e-05, 8.7575e-05, 8.8292e-05, 1.0509e-04, 1.3284e-04, 1.1497e-04], device='cuda:3') 2022-12-07 05:55:16,795 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=5755.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:55:58,101 INFO [train.py:873] (3/4) Epoch 1, batch 5800, loss[loss=0.2671, simple_loss=0.2452, pruned_loss=0.1446, over 6906.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.2514, pruned_loss=0.1472, over 1894855.98 frames. ], batch size: 100, lr: 4.04e-02, grad_scale: 16.0 2022-12-07 05:56:03,252 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.919e+02 3.658e+02 4.622e+02 6.453e+02 1.135e+03, threshold=9.244e+02, percent-clipped=6.0 2022-12-07 05:56:11,990 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=5816.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:56:34,598 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9831, 3.1140, 2.3653, 3.4728, 2.9665, 3.3245, 3.1176, 2.3284], device='cuda:3'), covar=tensor([0.0241, 0.0509, 0.2613, 0.0230, 0.0205, 0.0394, 0.0404, 0.2724], device='cuda:3'), in_proj_covar=tensor([0.0108, 0.0124, 0.0204, 0.0096, 0.0091, 0.0100, 0.0113, 0.0234], device='cuda:3'), out_proj_covar=tensor([6.2827e-05, 7.8044e-05, 1.2617e-04, 5.6347e-05, 5.4447e-05, 6.2053e-05, 7.0529e-05, 1.4370e-04], device='cuda:3') 2022-12-07 05:56:48,682 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=5858.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:56:58,118 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5162, 3.9689, 4.2775, 3.3105, 4.1420, 2.6703, 3.6714, 2.4341], device='cuda:3'), covar=tensor([0.0300, 0.0245, 0.0092, 0.0386, 0.0130, 0.0975, 0.0079, 0.1024], device='cuda:3'), in_proj_covar=tensor([0.0054, 0.0059, 0.0041, 0.0056, 0.0058, 0.0094, 0.0040, 0.0089], device='cuda:3'), out_proj_covar=tensor([4.6460e-05, 5.3433e-05, 3.6911e-05, 5.0263e-05, 4.9622e-05, 8.3670e-05, 3.2193e-05, 7.9632e-05], device='cuda:3') 2022-12-07 05:57:02,380 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3133, 3.3421, 3.5595, 3.0085, 3.6906, 2.4577, 3.0311, 2.3344], device='cuda:3'), covar=tensor([0.0319, 0.0365, 0.0171, 0.0445, 0.0204, 0.1180, 0.0129, 0.1208], device='cuda:3'), in_proj_covar=tensor([0.0054, 0.0059, 0.0041, 0.0056, 0.0058, 0.0095, 0.0041, 0.0090], device='cuda:3'), out_proj_covar=tensor([4.6747e-05, 5.3810e-05, 3.7099e-05, 5.0681e-05, 4.9775e-05, 8.4655e-05, 3.2401e-05, 8.0294e-05], device='cuda:3') 2022-12-07 05:57:03,363 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=16.49 vs. limit=5.0 2022-12-07 05:57:05,815 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5849, 1.5709, 3.1845, 2.1276, 3.5745, 3.4538, 2.3676, 3.7128], device='cuda:3'), covar=tensor([0.0229, 0.2318, 0.0358, 0.1555, 0.0272, 0.0294, 0.0805, 0.0258], device='cuda:3'), in_proj_covar=tensor([0.0053, 0.0091, 0.0048, 0.0085, 0.0065, 0.0052, 0.0051, 0.0053], device='cuda:3'), out_proj_covar=tensor([8.0544e-05, 1.3214e-04, 8.3886e-05, 1.2584e-04, 1.0327e-04, 8.4253e-05, 8.3212e-05, 8.4741e-05], device='cuda:3') 2022-12-07 05:57:26,653 INFO [train.py:873] (3/4) Epoch 1, batch 5900, loss[loss=0.2459, simple_loss=0.2373, pruned_loss=0.1273, over 14211.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.2503, pruned_loss=0.1453, over 1886396.59 frames. ], batch size: 35, lr: 4.02e-02, grad_scale: 16.0 2022-12-07 05:57:31,833 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.800e+02 3.450e+02 4.399e+02 5.936e+02 1.198e+03, threshold=8.798e+02, percent-clipped=6.0 2022-12-07 05:57:34,353 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=7.19 vs. limit=5.0 2022-12-07 05:57:42,963 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=5919.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:58:18,463 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2022-12-07 05:58:20,863 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.93 vs. limit=2.0 2022-12-07 05:58:44,821 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=5990.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:58:54,697 INFO [train.py:873] (3/4) Epoch 1, batch 6000, loss[loss=0.2831, simple_loss=0.2651, pruned_loss=0.1506, over 14296.00 frames. ], tot_loss[loss=0.271, simple_loss=0.2508, pruned_loss=0.1456, over 1933794.92 frames. ], batch size: 25, lr: 4.00e-02, grad_scale: 16.0 2022-12-07 05:58:54,697 INFO [train.py:896] (3/4) Computing validation loss 2022-12-07 05:59:02,167 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.0859, 5.2547, 4.6389, 5.3198, 5.0225, 4.8852, 5.5514, 5.1595], device='cuda:3'), covar=tensor([0.0531, 0.0288, 0.1054, 0.0972, 0.0977, 0.0396, 0.0513, 0.1892], device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0063, 0.0087, 0.0082, 0.0085, 0.0055, 0.0081, 0.0081], device='cuda:3'), out_proj_covar=tensor([1.1197e-04, 9.4482e-05, 1.2475e-04, 1.1654e-04, 1.1896e-04, 7.9187e-05, 1.2346e-04, 1.1779e-04], device='cuda:3') 2022-12-07 05:59:02,830 INFO [train.py:905] (3/4) Epoch 1, validation: loss=0.159, simple_loss=0.1938, pruned_loss=0.06211, over 857387.00 frames. 2022-12-07 05:59:02,830 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17680MB 2022-12-07 05:59:05,100 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.10 vs. limit=2.0 2022-12-07 05:59:07,243 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6006.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 05:59:07,938 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.676e+02 3.896e+02 5.286e+02 6.584e+02 1.638e+03, threshold=1.057e+03, percent-clipped=9.0 2022-12-07 05:59:27,196 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.7354, 4.4135, 4.5621, 4.8410, 4.7077, 3.5978, 5.0856, 5.0427], device='cuda:3'), covar=tensor([0.0604, 0.0503, 0.0512, 0.0632, 0.0502, 0.0711, 0.0549, 0.0576], device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0065, 0.0088, 0.0086, 0.0088, 0.0057, 0.0083, 0.0084], device='cuda:3'), out_proj_covar=tensor([1.1669e-04, 9.8142e-05, 1.2573e-04, 1.2230e-04, 1.2415e-04, 8.1177e-05, 1.2660e-04, 1.2181e-04], device='cuda:3') 2022-12-07 05:59:38,749 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.2592, 4.0098, 3.8338, 4.1954, 4.3108, 4.2317, 4.2186, 3.9460], device='cuda:3'), covar=tensor([0.0289, 0.0471, 0.0329, 0.0394, 0.0289, 0.0315, 0.0351, 0.0390], device='cuda:3'), in_proj_covar=tensor([0.0049, 0.0073, 0.0061, 0.0058, 0.0056, 0.0065, 0.0082, 0.0077], device='cuda:3'), out_proj_covar=tensor([7.9493e-05, 1.0713e-04, 8.8700e-05, 9.2647e-05, 8.7131e-05, 1.0361e-04, 1.3682e-04, 1.2118e-04], device='cuda:3') 2022-12-07 05:59:39,714 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6043.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:59:46,745 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6051.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 05:59:56,735 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.95 vs. limit=5.0 2022-12-07 06:00:30,194 INFO [train.py:873] (3/4) Epoch 1, batch 6100, loss[loss=0.2939, simple_loss=0.2578, pruned_loss=0.165, over 4999.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.2503, pruned_loss=0.145, over 1949933.41 frames. ], batch size: 100, lr: 3.98e-02, grad_scale: 16.0 2022-12-07 06:00:33,333 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6104.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:00:35,714 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.366e+02 3.351e+02 5.105e+02 6.275e+02 1.538e+03, threshold=1.021e+03, percent-clipped=3.0 2022-12-07 06:00:39,079 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6111.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:00:44,555 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.98 vs. limit=5.0 2022-12-07 06:01:00,467 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6135.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:01:03,083 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8145, 2.2951, 4.5026, 3.0013, 4.4624, 4.4901, 3.9637, 5.2018], device='cuda:3'), covar=tensor([0.0106, 0.1655, 0.0234, 0.1270, 0.0143, 0.0190, 0.0201, 0.0097], device='cuda:3'), in_proj_covar=tensor([0.0054, 0.0092, 0.0050, 0.0092, 0.0067, 0.0053, 0.0053, 0.0052], device='cuda:3'), out_proj_covar=tensor([8.4792e-05, 1.3735e-04, 8.7470e-05, 1.3836e-04, 1.0918e-04, 8.7382e-05, 8.9142e-05, 8.6060e-05], device='cuda:3') 2022-12-07 06:01:30,349 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.7114, 0.8738, 1.0313, 1.2142, 1.3744, 1.3345, 0.6826, 0.9848], device='cuda:3'), covar=tensor([0.0298, 0.0423, 0.0582, 0.0131, 0.0130, 0.0107, 0.0181, 0.0329], device='cuda:3'), in_proj_covar=tensor([0.0016, 0.0017, 0.0018, 0.0016, 0.0016, 0.0018, 0.0017, 0.0018], device='cuda:3'), out_proj_covar=tensor([1.9941e-05, 1.9529e-05, 2.4413e-05, 1.7893e-05, 1.8914e-05, 2.0224e-05, 2.2512e-05, 2.1271e-05], device='cuda:3') 2022-12-07 06:01:31,135 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6170.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:01:38,207 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.72 vs. limit=5.0 2022-12-07 06:01:53,781 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6196.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:01:58,122 INFO [train.py:873] (3/4) Epoch 1, batch 6200, loss[loss=0.3072, simple_loss=0.2731, pruned_loss=0.1706, over 14268.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.2501, pruned_loss=0.1446, over 1976806.92 frames. ], batch size: 37, lr: 3.96e-02, grad_scale: 16.0 2022-12-07 06:02:00,052 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6203.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:02:03,437 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.672e+01 3.177e+02 4.421e+02 6.121e+02 1.475e+03, threshold=8.841e+02, percent-clipped=2.0 2022-12-07 06:02:07,165 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6211.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:02:08,039 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3812, 2.6892, 2.0170, 2.8525, 2.5421, 2.6655, 2.4441, 2.0567], device='cuda:3'), covar=tensor([0.0217, 0.0455, 0.1749, 0.0180, 0.0199, 0.0295, 0.0460, 0.2254], device='cuda:3'), in_proj_covar=tensor([0.0113, 0.0127, 0.0208, 0.0098, 0.0093, 0.0106, 0.0119, 0.0240], device='cuda:3'), out_proj_covar=tensor([6.6139e-05, 7.9968e-05, 1.2911e-04, 5.8860e-05, 5.9515e-05, 6.6914e-05, 7.5424e-05, 1.4905e-04], device='cuda:3') 2022-12-07 06:02:09,382 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.05 vs. limit=2.0 2022-12-07 06:02:09,671 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6214.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:02:24,759 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6231.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:02:27,504 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.02 vs. limit=2.0 2022-12-07 06:02:36,067 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8479, 0.8563, 1.1506, 1.2329, 0.9412, 1.3361, 0.7926, 0.8605], device='cuda:3'), covar=tensor([0.1331, 0.0683, 0.0233, 0.0216, 0.0400, 0.0239, 0.0784, 0.0404], device='cuda:3'), in_proj_covar=tensor([0.0094, 0.0053, 0.0033, 0.0036, 0.0044, 0.0037, 0.0046, 0.0042], device='cuda:3'), out_proj_covar=tensor([1.3881e-04, 8.2399e-05, 5.9092e-05, 6.0044e-05, 6.5617e-05, 6.2857e-05, 7.3730e-05, 6.3558e-05], device='cuda:3') 2022-12-07 06:02:54,090 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6264.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:02:55,636 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6266.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:03:00,869 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6272.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:03:03,386 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9096, 1.9115, 3.8614, 2.9927, 3.5730, 2.4686, 3.6311, 3.5383], device='cuda:3'), covar=tensor([0.0299, 0.4238, 0.0310, 0.5462, 0.0143, 0.2409, 0.0479, 0.0143], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0245, 0.0134, 0.0342, 0.0103, 0.0271, 0.0164, 0.0117], device='cuda:3'), out_proj_covar=tensor([1.1981e-04, 2.1041e-04, 1.0868e-04, 2.8365e-04, 8.2945e-05, 2.2090e-04, 1.2889e-04, 9.1810e-05], device='cuda:3') 2022-12-07 06:03:26,923 INFO [train.py:873] (3/4) Epoch 1, batch 6300, loss[loss=0.2972, simple_loss=0.2794, pruned_loss=0.1575, over 14184.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.2496, pruned_loss=0.1438, over 1972439.86 frames. ], batch size: 25, lr: 3.94e-02, grad_scale: 16.0 2022-12-07 06:03:31,256 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6306.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 06:03:31,914 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.538e+02 3.956e+02 5.182e+02 6.845e+02 1.642e+03, threshold=1.036e+03, percent-clipped=11.0 2022-12-07 06:03:44,917 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5090, 2.4478, 2.3720, 2.1679, 2.4230, 2.3840, 2.4237, 2.4853], device='cuda:3'), covar=tensor([0.0317, 0.0412, 0.0571, 0.0726, 0.0489, 0.0332, 0.0650, 0.0466], device='cuda:3'), in_proj_covar=tensor([0.0088, 0.0088, 0.0108, 0.0073, 0.0093, 0.0089, 0.0097, 0.0081], device='cuda:3'), out_proj_covar=tensor([1.1030e-04, 1.1441e-04, 1.3620e-04, 9.2904e-05, 1.1452e-04, 1.0956e-04, 1.4196e-04, 1.0834e-04], device='cuda:3') 2022-12-07 06:03:49,239 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6327.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:04:05,594 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6346.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:04:12,444 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6354.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 06:04:29,827 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0659, 1.9472, 2.2587, 2.4896, 2.2071, 1.7912, 2.2889, 2.1338], device='cuda:3'), covar=tensor([0.0232, 0.0285, 0.0179, 0.0253, 0.0211, 0.0747, 0.0100, 0.0572], device='cuda:3'), in_proj_covar=tensor([0.0061, 0.0067, 0.0049, 0.0067, 0.0069, 0.0111, 0.0045, 0.0107], device='cuda:3'), out_proj_covar=tensor([5.5462e-05, 6.3275e-05, 4.6155e-05, 6.5329e-05, 6.3338e-05, 1.0381e-04, 3.8101e-05, 9.9772e-05], device='cuda:3') 2022-12-07 06:04:51,686 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6399.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:04:53,610 INFO [train.py:873] (3/4) Epoch 1, batch 6400, loss[loss=0.2101, simple_loss=0.2153, pruned_loss=0.1025, over 14324.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.249, pruned_loss=0.1439, over 1946579.87 frames. ], batch size: 28, lr: 3.92e-02, grad_scale: 8.0 2022-12-07 06:05:00,167 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.419e+02 3.375e+02 4.095e+02 5.340e+02 1.039e+03, threshold=8.189e+02, percent-clipped=1.0 2022-12-07 06:05:03,114 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6411.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:05:44,997 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6459.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:05:47,788 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6462.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:05:52,021 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6467.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:06:13,010 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6491.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:06:21,244 INFO [train.py:873] (3/4) Epoch 1, batch 6500, loss[loss=0.2532, simple_loss=0.2378, pruned_loss=0.1343, over 14252.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.2493, pruned_loss=0.1442, over 1977685.19 frames. ], batch size: 63, lr: 3.90e-02, grad_scale: 8.0 2022-12-07 06:06:27,500 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.203e+01 3.294e+02 4.343e+02 5.711e+02 1.098e+03, threshold=8.686e+02, percent-clipped=4.0 2022-12-07 06:06:32,760 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6514.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:06:40,837 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6523.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 06:06:43,802 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6526.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:06:45,597 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6528.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:07:12,795 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6559.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:07:14,692 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.7327, 0.6261, 0.8255, 0.7185, 0.8344, 0.6871, 0.7982, 0.7788], device='cuda:3'), covar=tensor([0.0725, 0.0926, 0.0403, 0.1106, 0.0428, 0.0640, 0.0524, 0.0493], device='cuda:3'), in_proj_covar=tensor([0.0026, 0.0028, 0.0029, 0.0025, 0.0025, 0.0032, 0.0022, 0.0024], device='cuda:3'), out_proj_covar=tensor([3.6705e-05, 4.2716e-05, 3.8748e-05, 3.6507e-05, 3.2869e-05, 4.4177e-05, 3.1989e-05, 3.1937e-05], device='cuda:3') 2022-12-07 06:07:15,379 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6562.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:07:20,359 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6567.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:07:50,328 INFO [train.py:873] (3/4) Epoch 1, batch 6600, loss[loss=0.2686, simple_loss=0.2497, pruned_loss=0.1438, over 14186.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.2477, pruned_loss=0.1423, over 1994548.91 frames. ], batch size: 84, lr: 3.89e-02, grad_scale: 8.0 2022-12-07 06:07:56,666 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.510e+02 3.480e+02 4.603e+02 5.913e+02 1.121e+03, threshold=9.206e+02, percent-clipped=7.0 2022-12-07 06:08:08,500 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6622.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:08:14,817 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6629.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:08:19,245 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7645, 2.9948, 2.3511, 3.1429, 2.4958, 3.0780, 2.6471, 2.3251], device='cuda:3'), covar=tensor([0.0209, 0.0444, 0.2151, 0.0215, 0.0291, 0.0207, 0.0624, 0.2591], device='cuda:3'), in_proj_covar=tensor([0.0113, 0.0128, 0.0215, 0.0105, 0.0100, 0.0108, 0.0129, 0.0249], device='cuda:3'), out_proj_covar=tensor([6.7494e-05, 8.3564e-05, 1.3504e-04, 6.3849e-05, 6.4973e-05, 6.8034e-05, 8.3254e-05, 1.5343e-04], device='cuda:3') 2022-12-07 06:08:30,341 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6646.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:08:36,378 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9001, 1.2632, 2.0358, 1.5048, 1.8865, 1.9198, 1.1611, 1.9007], device='cuda:3'), covar=tensor([0.0233, 0.0989, 0.0227, 0.0957, 0.0265, 0.0288, 0.0693, 0.0316], device='cuda:3'), in_proj_covar=tensor([0.0058, 0.0099, 0.0054, 0.0099, 0.0073, 0.0057, 0.0056, 0.0056], device='cuda:3'), out_proj_covar=tensor([9.7478e-05, 1.5399e-04, 1.0039e-04, 1.5534e-04, 1.2676e-04, 9.9429e-05, 9.8226e-05, 9.6933e-05], device='cuda:3') 2022-12-07 06:08:58,759 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.33 vs. limit=5.0 2022-12-07 06:09:09,370 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6690.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:09:12,777 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6694.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:09:17,199 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6699.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:09:18,766 INFO [train.py:873] (3/4) Epoch 1, batch 6700, loss[loss=0.2607, simple_loss=0.2488, pruned_loss=0.1363, over 14275.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.2472, pruned_loss=0.1419, over 1993593.58 frames. ], batch size: 44, lr: 3.87e-02, grad_scale: 8.0 2022-12-07 06:09:24,613 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.051e+02 3.291e+02 4.382e+02 6.181e+02 1.245e+03, threshold=8.764e+02, percent-clipped=7.0 2022-12-07 06:09:30,254 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.26 vs. limit=2.0 2022-12-07 06:09:58,746 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6747.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:10:21,129 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5622, 2.0888, 2.2444, 2.3600, 2.3186, 2.4101, 2.3076, 2.2171], device='cuda:3'), covar=tensor([0.0250, 0.1534, 0.0121, 0.0510, 0.0144, 0.0296, 0.0301, 0.1529], device='cuda:3'), in_proj_covar=tensor([0.0106, 0.0195, 0.0080, 0.0117, 0.0083, 0.0102, 0.0083, 0.0226], device='cuda:3'), out_proj_covar=tensor([8.1426e-05, 1.5294e-04, 5.7894e-05, 9.3632e-05, 6.0755e-05, 7.7001e-05, 7.4669e-05, 1.7190e-04], device='cuda:3') 2022-12-07 06:10:26,745 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5385, 3.3741, 3.3297, 3.0512, 3.4252, 3.3435, 3.4649, 3.4531], device='cuda:3'), covar=tensor([0.0440, 0.0517, 0.0533, 0.0743, 0.0381, 0.0418, 0.0691, 0.0674], device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0091, 0.0116, 0.0079, 0.0097, 0.0094, 0.0103, 0.0088], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001], device='cuda:3') 2022-12-07 06:10:27,055 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.26 vs. limit=5.0 2022-12-07 06:10:37,364 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6791.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:10:46,420 INFO [train.py:873] (3/4) Epoch 1, batch 6800, loss[loss=0.2387, simple_loss=0.2323, pruned_loss=0.1226, over 13949.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.2454, pruned_loss=0.1406, over 1963021.41 frames. ], batch size: 23, lr: 3.85e-02, grad_scale: 8.0 2022-12-07 06:10:52,799 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.483e+01 3.173e+02 4.494e+02 5.972e+02 9.591e+02, threshold=8.989e+02, percent-clipped=5.0 2022-12-07 06:11:01,514 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6818.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 06:11:05,730 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6823.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:11:08,603 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6826.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:11:12,148 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8966, 1.9349, 4.1416, 3.1743, 3.7695, 2.4743, 4.0380, 3.5892], device='cuda:3'), covar=tensor([0.0255, 0.5630, 0.0309, 0.6498, 0.0161, 0.3560, 0.0341, 0.0194], device='cuda:3'), in_proj_covar=tensor([0.0159, 0.0264, 0.0141, 0.0366, 0.0111, 0.0290, 0.0174, 0.0125], device='cuda:3'), out_proj_covar=tensor([1.2870e-04, 2.3130e-04, 1.1797e-04, 3.0478e-04, 9.5641e-05, 2.4048e-04, 1.3991e-04, 1.0252e-04], device='cuda:3') 2022-12-07 06:11:19,536 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6839.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:11:37,259 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6859.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:11:44,231 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6867.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:11:49,844 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6874.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:11:53,608 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6878.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 06:12:13,455 INFO [train.py:873] (3/4) Epoch 1, batch 6900, loss[loss=0.2865, simple_loss=0.2601, pruned_loss=0.1565, over 14403.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.2452, pruned_loss=0.1403, over 1937439.91 frames. ], batch size: 53, lr: 3.83e-02, grad_scale: 8.0 2022-12-07 06:12:18,831 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6907.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:12:19,586 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.182e+02 3.106e+02 4.451e+02 6.203e+02 1.044e+03, threshold=8.902e+02, percent-clipped=8.0 2022-12-07 06:12:20,697 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6909.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:12:25,627 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6915.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:12:32,011 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6922.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:12:46,750 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6939.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 06:13:09,177 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9427, 1.4623, 1.3242, 1.1172, 1.6742, 1.1343, 1.2130, 1.7106], device='cuda:3'), covar=tensor([0.0291, 0.3058, 0.0610, 0.1183, 0.0776, 0.0628, 0.2108, 0.0484], device='cuda:3'), in_proj_covar=tensor([0.0043, 0.0115, 0.0053, 0.0076, 0.0048, 0.0051, 0.0049, 0.0046], device='cuda:3'), out_proj_covar=tensor([6.3792e-05, 1.5984e-04, 7.5750e-05, 1.1034e-04, 7.7342e-05, 7.9527e-05, 8.1441e-05, 6.6925e-05], device='cuda:3') 2022-12-07 06:13:14,001 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6970.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:13:14,141 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6970.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:13:22,140 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.98 vs. limit=2.0 2022-12-07 06:13:27,137 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6985.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:13:33,384 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.21 vs. limit=5.0 2022-12-07 06:13:41,155 INFO [train.py:873] (3/4) Epoch 1, batch 7000, loss[loss=0.2148, simple_loss=0.2253, pruned_loss=0.1022, over 14045.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.244, pruned_loss=0.1388, over 1968160.76 frames. ], batch size: 19, lr: 3.81e-02, grad_scale: 8.0 2022-12-07 06:13:47,945 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.205e+02 2.871e+02 4.067e+02 4.755e+02 1.195e+03, threshold=8.134e+02, percent-clipped=1.0 2022-12-07 06:14:23,497 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7048.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 06:15:09,089 INFO [train.py:873] (3/4) Epoch 1, batch 7100, loss[loss=0.2201, simple_loss=0.1885, pruned_loss=0.1259, over 1306.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.2448, pruned_loss=0.1395, over 1956731.13 frames. ], batch size: 100, lr: 3.79e-02, grad_scale: 8.0 2022-12-07 06:15:11,332 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 2022-12-07 06:15:15,338 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.451e+02 3.113e+02 4.115e+02 5.668e+02 1.201e+03, threshold=8.229e+02, percent-clipped=4.0 2022-12-07 06:15:16,480 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7109.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 06:15:19,442 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2563, 2.3468, 3.3754, 2.8031, 2.8802, 2.4768, 1.3019, 2.4592], device='cuda:3'), covar=tensor([0.0574, 0.0486, 0.0304, 0.0324, 0.0255, 0.0468, 0.1439, 0.0537], device='cuda:3'), in_proj_covar=tensor([0.0049, 0.0051, 0.0049, 0.0050, 0.0052, 0.0050, 0.0088, 0.0053], device='cuda:3'), out_proj_covar=tensor([6.1508e-05, 6.0959e-05, 6.1003e-05, 5.2875e-05, 5.5923e-05, 5.6178e-05, 1.0663e-04, 6.4267e-05], device='cuda:3') 2022-12-07 06:15:23,832 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7118.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 06:15:28,206 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7123.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:15:57,603 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.5948, 0.3961, 0.4296, 0.6274, 0.5602, 0.3099, 0.6069, 0.4390], device='cuda:3'), covar=tensor([0.0167, 0.0181, 0.0129, 0.0282, 0.0164, 0.0251, 0.0132, 0.0177], device='cuda:3'), in_proj_covar=tensor([0.0024, 0.0026, 0.0028, 0.0023, 0.0025, 0.0029, 0.0020, 0.0023], device='cuda:3'), out_proj_covar=tensor([3.5956e-05, 4.1514e-05, 3.9753e-05, 3.7824e-05, 3.5252e-05, 4.3819e-05, 3.1409e-05, 3.2944e-05], device='cuda:3') 2022-12-07 06:16:03,219 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7163.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:16:05,595 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7166.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:16:09,766 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7171.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:16:35,875 INFO [train.py:873] (3/4) Epoch 1, batch 7200, loss[loss=0.253, simple_loss=0.2024, pruned_loss=0.1517, over 1239.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.2459, pruned_loss=0.1401, over 2016894.46 frames. ], batch size: 100, lr: 3.78e-02, grad_scale: 8.0 2022-12-07 06:16:42,299 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.199e+02 3.143e+02 4.183e+02 5.672e+02 1.523e+03, threshold=8.365e+02, percent-clipped=9.0 2022-12-07 06:16:56,245 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7224.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 06:17:04,795 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7234.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 06:17:20,209 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.0064, 4.6088, 4.5933, 4.7739, 4.7559, 4.7867, 5.0233, 4.9157], device='cuda:3'), covar=tensor([0.0420, 0.0546, 0.0569, 0.0305, 0.0273, 0.0314, 0.0375, 0.0667], device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0096, 0.0117, 0.0084, 0.0097, 0.0099, 0.0107, 0.0091], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001], device='cuda:3') 2022-12-07 06:17:28,907 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0453, 1.6768, 1.3194, 0.9969, 1.4775, 1.2386, 1.3312, 1.7717], device='cuda:3'), covar=tensor([0.0266, 0.2207, 0.0711, 0.1221, 0.0991, 0.0730, 0.1138, 0.0349], device='cuda:3'), in_proj_covar=tensor([0.0040, 0.0115, 0.0054, 0.0078, 0.0049, 0.0052, 0.0049, 0.0046], device='cuda:3'), out_proj_covar=tensor([6.0949e-05, 1.6375e-04, 8.0331e-05, 1.1478e-04, 8.0314e-05, 8.2326e-05, 8.4765e-05, 6.8814e-05], device='cuda:3') 2022-12-07 06:17:32,302 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7265.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:17:41,158 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7275.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:17:48,380 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.66 vs. limit=2.0 2022-12-07 06:17:49,094 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1558, 3.5659, 4.0509, 3.4300, 3.9937, 4.0399, 1.8805, 4.0377], device='cuda:3'), covar=tensor([0.0190, 0.0417, 0.0490, 0.0486, 0.0320, 0.0280, 0.2770, 0.0258], device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0066, 0.0072, 0.0061, 0.0090, 0.0062, 0.0109, 0.0083], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2022-12-07 06:17:50,864 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7285.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:18:04,544 INFO [train.py:873] (3/4) Epoch 1, batch 7300, loss[loss=0.2742, simple_loss=0.2589, pruned_loss=0.1448, over 14355.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.2438, pruned_loss=0.1377, over 2055473.66 frames. ], batch size: 55, lr: 3.76e-02, grad_scale: 8.0 2022-12-07 06:18:10,449 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.504e+02 3.489e+02 4.612e+02 5.728e+02 1.039e+03, threshold=9.225e+02, percent-clipped=2.0 2022-12-07 06:18:32,906 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7333.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:18:35,672 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7336.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:18:50,804 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8499, 0.6006, 0.4033, 0.7275, 0.7908, 0.4619, 0.5946, 0.4950], device='cuda:3'), covar=tensor([0.0130, 0.0271, 0.0125, 0.0534, 0.0119, 0.0332, 0.0195, 0.0201], device='cuda:3'), in_proj_covar=tensor([0.0022, 0.0025, 0.0027, 0.0022, 0.0023, 0.0028, 0.0020, 0.0023], device='cuda:3'), out_proj_covar=tensor([3.4793e-05, 4.0669e-05, 3.9279e-05, 3.6438e-05, 3.3453e-05, 4.3905e-05, 3.1870e-05, 3.2604e-05], device='cuda:3') 2022-12-07 06:19:03,883 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5129, 3.5669, 2.6705, 3.5481, 3.0267, 3.5770, 3.3757, 2.4547], device='cuda:3'), covar=tensor([0.0196, 0.0381, 0.2527, 0.0154, 0.0255, 0.0333, 0.0463, 0.2820], device='cuda:3'), in_proj_covar=tensor([0.0117, 0.0137, 0.0221, 0.0104, 0.0096, 0.0106, 0.0134, 0.0252], device='cuda:3'), out_proj_covar=tensor([7.2285e-05, 9.2209e-05, 1.4036e-04, 6.5617e-05, 6.6553e-05, 7.0711e-05, 8.9843e-05, 1.5777e-04], device='cuda:3') 2022-12-07 06:19:33,640 INFO [train.py:873] (3/4) Epoch 1, batch 7400, loss[loss=0.2613, simple_loss=0.2274, pruned_loss=0.1475, over 3830.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.2432, pruned_loss=0.1369, over 2037585.46 frames. ], batch size: 100, lr: 3.74e-02, grad_scale: 8.0 2022-12-07 06:19:36,723 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7404.0, num_to_drop=1, layers_to_drop={3} 2022-12-07 06:19:39,384 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7407.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:19:39,942 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.745e+02 3.232e+02 4.380e+02 5.803e+02 1.577e+03, threshold=8.760e+02, percent-clipped=3.0 2022-12-07 06:20:32,154 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7468.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:21:00,784 INFO [train.py:873] (3/4) Epoch 1, batch 7500, loss[loss=0.2398, simple_loss=0.2349, pruned_loss=0.1224, over 14286.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.2441, pruned_loss=0.1377, over 2011425.12 frames. ], batch size: 31, lr: 3.72e-02, grad_scale: 8.0 2022-12-07 06:21:06,494 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.457e+02 3.575e+02 4.286e+02 5.078e+02 8.997e+02, threshold=8.573e+02, percent-clipped=1.0 2022-12-07 06:21:11,956 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.7821, 5.0456, 5.2091, 5.6500, 5.5202, 5.0690, 5.6978, 5.1927], device='cuda:3'), covar=tensor([0.0147, 0.0500, 0.0221, 0.0288, 0.0270, 0.0294, 0.0440, 0.0266], device='cuda:3'), in_proj_covar=tensor([0.0055, 0.0094, 0.0068, 0.0063, 0.0067, 0.0077, 0.0092, 0.0084], device='cuda:3'), out_proj_covar=tensor([9.1695e-05, 1.4854e-04, 1.0489e-04, 1.0385e-04, 1.1266e-04, 1.2849e-04, 1.5966e-04, 1.3509e-04], device='cuda:3') 2022-12-07 06:21:16,674 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7519.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 06:21:29,235 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7534.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 06:22:29,585 INFO [train.py:873] (3/4) Epoch 2, batch 0, loss[loss=0.3751, simple_loss=0.328, pruned_loss=0.2111, over 6006.00 frames. ], tot_loss[loss=0.3751, simple_loss=0.328, pruned_loss=0.2111, over 6006.00 frames. ], batch size: 100, lr: 3.64e-02, grad_scale: 8.0 2022-12-07 06:22:29,586 INFO [train.py:896] (3/4) Computing validation loss 2022-12-07 06:22:36,878 INFO [train.py:905] (3/4) Epoch 2, validation: loss=0.201, simple_loss=0.225, pruned_loss=0.08852, over 857387.00 frames. 2022-12-07 06:22:36,879 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17680MB 2022-12-07 06:22:38,760 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7565.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:22:54,114 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7582.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 06:23:16,974 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.396e+01 2.991e+02 4.594e+02 6.425e+02 1.765e+03, threshold=9.187e+02, percent-clipped=13.0 2022-12-07 06:23:21,543 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7613.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:23:22,384 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5355, 4.1498, 4.3355, 4.7232, 4.4127, 3.3387, 4.8001, 4.7651], device='cuda:3'), covar=tensor([0.0794, 0.0594, 0.0717, 0.0734, 0.0628, 0.0683, 0.0688, 0.0778], device='cuda:3'), in_proj_covar=tensor([0.0088, 0.0067, 0.0091, 0.0083, 0.0090, 0.0060, 0.0081, 0.0088], device='cuda:3'), out_proj_covar=tensor([1.3262e-04, 1.0640e-04, 1.3824e-04, 1.2904e-04, 1.3517e-04, 9.1702e-05, 1.2923e-04, 1.3562e-04], device='cuda:3') 2022-12-07 06:23:31,713 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7624.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:23:35,905 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6909, 1.2384, 1.0201, 0.7029, 1.3828, 1.1875, 1.2689, 1.3989], device='cuda:3'), covar=tensor([0.0270, 0.2567, 0.0759, 0.1512, 0.0562, 0.0560, 0.0926, 0.0464], device='cuda:3'), in_proj_covar=tensor([0.0042, 0.0111, 0.0055, 0.0079, 0.0048, 0.0052, 0.0051, 0.0048], device='cuda:3'), out_proj_covar=tensor([6.6293e-05, 1.6496e-04, 8.7118e-05, 1.1861e-04, 8.1994e-05, 8.5987e-05, 8.9606e-05, 7.6615e-05], device='cuda:3') 2022-12-07 06:23:37,429 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7631.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:23:50,388 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0086, 1.9243, 2.0290, 1.9620, 2.0445, 1.7953, 1.1613, 1.9047], device='cuda:3'), covar=tensor([0.0268, 0.0272, 0.0338, 0.0187, 0.0247, 0.0403, 0.1280, 0.0326], device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0069, 0.0072, 0.0063, 0.0094, 0.0066, 0.0113, 0.0085], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2022-12-07 06:24:05,652 INFO [train.py:873] (3/4) Epoch 2, batch 100, loss[loss=0.2769, simple_loss=0.2651, pruned_loss=0.1443, over 14302.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.2466, pruned_loss=0.1392, over 893026.87 frames. ], batch size: 60, lr: 3.62e-02, grad_scale: 8.0 2022-12-07 06:24:24,617 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7685.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:24:37,199 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.73 vs. limit=2.0 2022-12-07 06:24:41,584 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7704.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 06:24:44,979 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.630e+02 3.684e+02 4.699e+02 5.885e+02 1.470e+03, threshold=9.397e+02, percent-clipped=3.0 2022-12-07 06:25:16,178 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7743.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:25:23,888 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7752.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 06:25:32,821 INFO [train.py:873] (3/4) Epoch 2, batch 200, loss[loss=0.312, simple_loss=0.2562, pruned_loss=0.1838, over 2628.00 frames. ], tot_loss[loss=0.262, simple_loss=0.2457, pruned_loss=0.1391, over 1343497.89 frames. ], batch size: 100, lr: 3.61e-02, grad_scale: 8.0 2022-12-07 06:25:33,219 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7763.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:25:38,658 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.31 vs. limit=5.0 2022-12-07 06:26:00,367 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.59 vs. limit=2.0 2022-12-07 06:26:07,040 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7802.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:26:08,746 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7804.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:26:12,299 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.334e+02 3.123e+02 4.049e+02 5.757e+02 1.056e+03, threshold=8.099e+02, percent-clipped=3.0 2022-12-07 06:26:22,346 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7819.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 06:26:38,124 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.7981, 0.7013, 0.7855, 0.7894, 0.7584, 0.7153, 1.0168, 0.8866], device='cuda:3'), covar=tensor([0.0692, 0.0969, 0.0531, 0.0591, 0.0770, 0.0456, 0.0391, 0.0352], device='cuda:3'), in_proj_covar=tensor([0.0021, 0.0024, 0.0025, 0.0023, 0.0023, 0.0025, 0.0019, 0.0021], device='cuda:3'), out_proj_covar=tensor([3.4304e-05, 4.1215e-05, 3.9449e-05, 3.9797e-05, 3.4437e-05, 4.0478e-05, 3.2278e-05, 3.2553e-05], device='cuda:3') 2022-12-07 06:26:59,712 INFO [train.py:873] (3/4) Epoch 2, batch 300, loss[loss=0.2731, simple_loss=0.2541, pruned_loss=0.1461, over 11979.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.2441, pruned_loss=0.1385, over 1591756.20 frames. ], batch size: 100, lr: 3.59e-02, grad_scale: 8.0 2022-12-07 06:26:59,916 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7863.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:27:03,175 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7867.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:27:38,763 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.219e+02 3.418e+02 4.416e+02 5.390e+02 1.169e+03, threshold=8.833e+02, percent-clipped=6.0 2022-12-07 06:27:58,878 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7931.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:28:00,857 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6888, 1.4548, 1.8249, 1.6225, 1.9364, 1.6216, 1.7590, 1.6767], device='cuda:3'), covar=tensor([0.0195, 0.0512, 0.0078, 0.0126, 0.0084, 0.0196, 0.0110, 0.0241], device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0233, 0.0092, 0.0143, 0.0096, 0.0123, 0.0103, 0.0264], device='cuda:3'), out_proj_covar=tensor([1.0022e-04, 1.8985e-04, 6.9670e-05, 1.1673e-04, 7.7141e-05, 9.5792e-05, 9.6315e-05, 2.0870e-04], device='cuda:3') 2022-12-07 06:28:26,420 INFO [train.py:873] (3/4) Epoch 2, batch 400, loss[loss=0.2353, simple_loss=0.2327, pruned_loss=0.119, over 14206.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.2425, pruned_loss=0.1372, over 1638093.14 frames. ], batch size: 60, lr: 3.58e-02, grad_scale: 8.0 2022-12-07 06:28:40,958 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7979.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:28:41,824 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7980.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:28:46,147 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.23 vs. limit=5.0 2022-12-07 06:28:53,955 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.99 vs. limit=2.0 2022-12-07 06:28:55,218 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6872, 1.5907, 2.0124, 2.1041, 2.1984, 1.7991, 1.9796, 2.0054], device='cuda:3'), covar=tensor([0.0325, 0.1060, 0.0125, 0.0283, 0.0104, 0.0338, 0.0209, 0.0574], device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0233, 0.0089, 0.0144, 0.0094, 0.0124, 0.0103, 0.0265], device='cuda:3'), out_proj_covar=tensor([1.0239e-04, 1.9110e-04, 6.6898e-05, 1.1791e-04, 7.5224e-05, 9.7719e-05, 9.6519e-05, 2.0965e-04], device='cuda:3') 2022-12-07 06:29:05,272 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.294e+01 3.140e+02 4.082e+02 5.332e+02 1.723e+03, threshold=8.164e+02, percent-clipped=4.0 2022-12-07 06:29:53,231 INFO [train.py:873] (3/4) Epoch 2, batch 500, loss[loss=0.2456, simple_loss=0.2344, pruned_loss=0.1284, over 14461.00 frames. ], tot_loss[loss=0.258, simple_loss=0.2423, pruned_loss=0.1369, over 1787209.56 frames. ], batch size: 27, lr: 3.56e-02, grad_scale: 8.0 2022-12-07 06:29:53,391 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8063.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:30:25,018 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8099.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:30:32,478 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.590e+01 3.381e+02 4.581e+02 6.129e+02 1.327e+03, threshold=9.162e+02, percent-clipped=13.0 2022-12-07 06:30:35,069 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=8111.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:31:15,334 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8158.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:31:19,468 INFO [train.py:873] (3/4) Epoch 2, batch 600, loss[loss=0.2319, simple_loss=0.2091, pruned_loss=0.1273, over 3922.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.2422, pruned_loss=0.1366, over 1845751.16 frames. ], batch size: 100, lr: 3.54e-02, grad_scale: 8.0 2022-12-07 06:31:59,046 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.447e+02 3.066e+02 3.782e+02 5.192e+02 1.492e+03, threshold=7.564e+02, percent-clipped=6.0 2022-12-07 06:32:47,259 INFO [train.py:873] (3/4) Epoch 2, batch 700, loss[loss=0.2686, simple_loss=0.2267, pruned_loss=0.1552, over 3871.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.2428, pruned_loss=0.1363, over 1945249.12 frames. ], batch size: 100, lr: 3.53e-02, grad_scale: 8.0 2022-12-07 06:32:48,650 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=19.88 vs. limit=5.0 2022-12-07 06:32:49,135 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8265.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:33:00,477 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.99 vs. limit=5.0 2022-12-07 06:33:01,907 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8280.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:33:26,031 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.287e+02 3.101e+02 3.944e+02 5.331e+02 1.080e+03, threshold=7.888e+02, percent-clipped=5.0 2022-12-07 06:33:41,599 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8326.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:33:43,081 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=8328.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:33:48,650 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4810, 1.3923, 1.9373, 2.0351, 1.3580, 1.3393, 1.8198, 1.5263], device='cuda:3'), covar=tensor([0.0132, 0.0126, 0.0042, 0.0060, 0.0116, 0.0269, 0.0066, 0.0145], device='cuda:3'), in_proj_covar=tensor([0.0080, 0.0090, 0.0065, 0.0090, 0.0093, 0.0137, 0.0057, 0.0137], device='cuda:3'), out_proj_covar=tensor([8.2596e-05, 9.9043e-05, 7.3772e-05, 1.0195e-04, 1.0588e-04, 1.5235e-04, 5.9579e-05, 1.4687e-04], device='cuda:3') 2022-12-07 06:34:00,015 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.99 vs. limit=5.0 2022-12-07 06:34:10,288 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.89 vs. limit=2.0 2022-12-07 06:34:13,679 INFO [train.py:873] (3/4) Epoch 2, batch 800, loss[loss=0.2396, simple_loss=0.225, pruned_loss=0.1271, over 14257.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.2424, pruned_loss=0.1362, over 1956686.27 frames. ], batch size: 76, lr: 3.51e-02, grad_scale: 16.0 2022-12-07 06:34:45,240 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8399.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:34:53,071 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.100e+02 2.975e+02 4.226e+02 5.564e+02 1.451e+03, threshold=8.453e+02, percent-clipped=7.0 2022-12-07 06:35:26,085 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=8447.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:35:28,518 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.91 vs. limit=2.0 2022-12-07 06:35:36,614 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8458.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:35:40,663 INFO [train.py:873] (3/4) Epoch 2, batch 900, loss[loss=0.2551, simple_loss=0.2357, pruned_loss=0.1373, over 14369.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.2415, pruned_loss=0.1349, over 2014678.27 frames. ], batch size: 73, lr: 3.50e-02, grad_scale: 16.0 2022-12-07 06:35:43,623 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.88 vs. limit=2.0 2022-12-07 06:35:50,679 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0453, 3.7844, 3.7598, 3.7870, 3.8167, 3.8222, 3.9396, 4.0266], device='cuda:3'), covar=tensor([0.0377, 0.0386, 0.0407, 0.0345, 0.0315, 0.0272, 0.0661, 0.0431], device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0099, 0.0116, 0.0096, 0.0108, 0.0108, 0.0124, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001], device='cuda:3') 2022-12-07 06:35:59,262 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=15.30 vs. limit=5.0 2022-12-07 06:36:11,705 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7542, 1.3676, 1.8728, 1.4799, 1.9337, 1.6513, 1.8077, 1.6670], device='cuda:3'), covar=tensor([0.0139, 0.0529, 0.0043, 0.0133, 0.0059, 0.0131, 0.0090, 0.0208], device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0240, 0.0094, 0.0160, 0.0101, 0.0127, 0.0116, 0.0274], device='cuda:3'), out_proj_covar=tensor([1.0512e-04, 1.9672e-04, 7.3248e-05, 1.3038e-04, 8.3246e-05, 1.0077e-04, 1.1058e-04, 2.2023e-04], device='cuda:3') 2022-12-07 06:36:17,476 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=8506.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:36:19,007 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.248e+02 3.358e+02 4.277e+02 5.886e+02 1.045e+03, threshold=8.554e+02, percent-clipped=3.0 2022-12-07 06:37:06,427 INFO [train.py:873] (3/4) Epoch 2, batch 1000, loss[loss=0.1973, simple_loss=0.1745, pruned_loss=0.1101, over 2581.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.2401, pruned_loss=0.1336, over 1975448.03 frames. ], batch size: 100, lr: 3.48e-02, grad_scale: 16.0 2022-12-07 06:37:27,774 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6270, 1.2579, 1.1492, 0.9716, 1.2682, 0.9324, 1.2166, 1.2799], device='cuda:3'), covar=tensor([0.0129, 0.0679, 0.0284, 0.0607, 0.0279, 0.0504, 0.0306, 0.0200], device='cuda:3'), in_proj_covar=tensor([0.0043, 0.0118, 0.0056, 0.0083, 0.0054, 0.0053, 0.0052, 0.0049], device='cuda:3'), out_proj_covar=tensor([7.2716e-05, 1.8509e-04, 9.4928e-05, 1.3625e-04, 9.8397e-05, 9.5403e-05, 9.7984e-05, 8.5947e-05], device='cuda:3') 2022-12-07 06:37:32,710 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9302, 1.5495, 2.9872, 2.8282, 3.0166, 2.8805, 2.2322, 3.1368], device='cuda:3'), covar=tensor([0.1353, 0.1353, 0.0137, 0.0177, 0.0125, 0.0137, 0.0315, 0.0104], device='cuda:3'), in_proj_covar=tensor([0.0108, 0.0115, 0.0056, 0.0075, 0.0062, 0.0062, 0.0052, 0.0052], device='cuda:3'), out_proj_covar=tensor([1.8810e-04, 1.9017e-04, 1.0105e-04, 1.4385e-04, 1.0799e-04, 1.1134e-04, 9.9897e-05, 9.1596e-05], device='cuda:3') 2022-12-07 06:37:46,391 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.423e+02 3.331e+02 4.272e+02 5.638e+02 1.026e+03, threshold=8.545e+02, percent-clipped=1.0 2022-12-07 06:37:57,598 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8621.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:38:05,282 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.7837, 0.6820, 0.6743, 0.8474, 0.8215, 0.7064, 0.8563, 0.7614], device='cuda:3'), covar=tensor([0.0388, 0.0495, 0.0232, 0.0589, 0.0532, 0.0435, 0.0413, 0.0337], device='cuda:3'), in_proj_covar=tensor([0.0022, 0.0024, 0.0025, 0.0022, 0.0022, 0.0024, 0.0021, 0.0021], device='cuda:3'), out_proj_covar=tensor([3.8660e-05, 4.3537e-05, 4.1353e-05, 4.0319e-05, 3.6076e-05, 4.2104e-05, 3.8308e-05, 3.5109e-05], device='cuda:3') 2022-12-07 06:38:34,123 INFO [train.py:873] (3/4) Epoch 2, batch 1100, loss[loss=0.2516, simple_loss=0.2447, pruned_loss=0.1292, over 14270.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.2394, pruned_loss=0.1335, over 1949282.93 frames. ], batch size: 35, lr: 3.47e-02, grad_scale: 8.0 2022-12-07 06:38:57,329 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.69 vs. limit=5.0 2022-12-07 06:39:13,894 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.477e+02 3.265e+02 4.282e+02 5.954e+02 1.002e+03, threshold=8.564e+02, percent-clipped=5.0 2022-12-07 06:39:31,246 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.26 vs. limit=2.0 2022-12-07 06:39:33,387 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9864, 4.7439, 4.5126, 4.9228, 4.6844, 4.8801, 5.0780, 4.9256], device='cuda:3'), covar=tensor([0.0470, 0.0437, 0.0421, 0.0270, 0.0336, 0.0277, 0.0412, 0.0656], device='cuda:3'), in_proj_covar=tensor([0.0107, 0.0100, 0.0124, 0.0097, 0.0111, 0.0113, 0.0125, 0.0103], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001], device='cuda:3') 2022-12-07 06:39:42,757 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9572, 1.4339, 3.4459, 1.9292, 3.7781, 3.5285, 2.2168, 4.0121], device='cuda:3'), covar=tensor([0.0155, 0.2840, 0.0450, 0.2321, 0.0255, 0.0323, 0.1049, 0.0165], device='cuda:3'), in_proj_covar=tensor([0.0072, 0.0113, 0.0069, 0.0122, 0.0086, 0.0072, 0.0070, 0.0070], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], device='cuda:3') 2022-12-07 06:39:43,592 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3053, 3.0194, 2.8996, 3.2878, 2.9687, 2.3877, 3.2988, 3.2243], device='cuda:3'), covar=tensor([0.0858, 0.0880, 0.1045, 0.0905, 0.0866, 0.1127, 0.0936, 0.1150], device='cuda:3'), in_proj_covar=tensor([0.0085, 0.0071, 0.0095, 0.0088, 0.0091, 0.0061, 0.0086, 0.0088], device='cuda:3'), out_proj_covar=tensor([1.3412e-04, 1.1591e-04, 1.5077e-04, 1.4052e-04, 1.4567e-04, 9.6520e-05, 1.3992e-04, 1.3974e-04], device='cuda:3') 2022-12-07 06:39:58,173 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8760.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:40:00,143 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=8.86 vs. limit=5.0 2022-12-07 06:40:00,477 INFO [train.py:873] (3/4) Epoch 2, batch 1200, loss[loss=0.2826, simple_loss=0.2412, pruned_loss=0.1619, over 3862.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.2399, pruned_loss=0.1337, over 1959241.00 frames. ], batch size: 100, lr: 3.45e-02, grad_scale: 8.0 2022-12-07 06:40:20,262 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.23 vs. limit=2.0 2022-12-07 06:40:40,233 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.965e+02 3.180e+02 4.401e+02 5.751e+02 1.287e+03, threshold=8.803e+02, percent-clipped=4.0 2022-12-07 06:40:40,380 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.0945, 0.8208, 1.4279, 0.7994, 1.2696, 1.2618, 0.7154, 1.0631], device='cuda:3'), covar=tensor([0.0371, 0.0889, 0.0191, 0.0420, 0.0197, 0.0236, 0.0646, 0.0265], device='cuda:3'), in_proj_covar=tensor([0.0052, 0.0054, 0.0050, 0.0048, 0.0054, 0.0050, 0.0093, 0.0057], device='cuda:3'), out_proj_covar=tensor([7.0511e-05, 7.2963e-05, 6.7273e-05, 6.0898e-05, 6.5170e-05, 6.3321e-05, 1.2235e-04, 7.5885e-05], device='cuda:3') 2022-12-07 06:40:51,374 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8821.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:41:12,048 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.95 vs. limit=2.0 2022-12-07 06:41:27,153 INFO [train.py:873] (3/4) Epoch 2, batch 1300, loss[loss=0.2471, simple_loss=0.2341, pruned_loss=0.13, over 14167.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.239, pruned_loss=0.1328, over 1923820.57 frames. ], batch size: 29, lr: 3.44e-02, grad_scale: 8.0 2022-12-07 06:42:06,753 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.360e+02 2.973e+02 3.684e+02 4.915e+02 1.042e+03, threshold=7.368e+02, percent-clipped=2.0 2022-12-07 06:42:06,940 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7865, 1.9224, 1.5644, 2.0643, 2.0422, 1.9766, 1.7622, 1.5363], device='cuda:3'), covar=tensor([0.0177, 0.0235, 0.0513, 0.0073, 0.0111, 0.0094, 0.0231, 0.0424], device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0158, 0.0256, 0.0121, 0.0116, 0.0118, 0.0151, 0.0293], device='cuda:3'), out_proj_covar=tensor([8.9343e-05, 1.0914e-04, 1.6795e-04, 8.1360e-05, 8.5373e-05, 8.3940e-05, 1.0481e-04, 1.8835e-04], device='cuda:3') 2022-12-07 06:42:10,692 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.91 vs. limit=2.0 2022-12-07 06:42:17,232 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8921.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:42:22,333 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8927.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:42:53,350 INFO [train.py:873] (3/4) Epoch 2, batch 1400, loss[loss=0.2308, simple_loss=0.2193, pruned_loss=0.1212, over 6933.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.2408, pruned_loss=0.1341, over 1976059.97 frames. ], batch size: 100, lr: 3.42e-02, grad_scale: 8.0 2022-12-07 06:42:58,583 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=8969.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:43:13,817 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8986.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:43:15,510 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8988.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:43:30,928 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1613, 1.9504, 2.1139, 2.1949, 2.1747, 2.2124, 2.2576, 1.9681], device='cuda:3'), covar=tensor([0.0334, 0.0965, 0.0431, 0.0484, 0.0599, 0.0507, 0.0571, 0.0647], device='cuda:3'), in_proj_covar=tensor([0.0061, 0.0109, 0.0074, 0.0072, 0.0080, 0.0083, 0.0108, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:3') 2022-12-07 06:43:33,021 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.228e+02 3.417e+02 4.711e+02 5.630e+02 1.118e+03, threshold=9.423e+02, percent-clipped=7.0 2022-12-07 06:43:36,713 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9013.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:44:06,546 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9047.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:44:20,923 INFO [train.py:873] (3/4) Epoch 2, batch 1500, loss[loss=0.2187, simple_loss=0.2264, pruned_loss=0.1055, over 14511.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.239, pruned_loss=0.1324, over 1988345.96 frames. ], batch size: 49, lr: 3.41e-02, grad_scale: 8.0 2022-12-07 06:44:25,278 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3200, 1.7765, 3.7079, 2.3487, 3.9487, 3.7988, 3.2440, 4.5155], device='cuda:3'), covar=tensor([0.0132, 0.2642, 0.0372, 0.1856, 0.0214, 0.0303, 0.0413, 0.0132], device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0115, 0.0069, 0.0122, 0.0086, 0.0073, 0.0068, 0.0070], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], device='cuda:3') 2022-12-07 06:44:30,409 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9074.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:44:59,853 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.468e+02 3.213e+02 4.122e+02 5.427e+02 9.769e+02, threshold=8.244e+02, percent-clipped=1.0 2022-12-07 06:45:05,745 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9116.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:45:44,167 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.48 vs. limit=2.0 2022-12-07 06:45:46,185 INFO [train.py:873] (3/4) Epoch 2, batch 1600, loss[loss=0.3152, simple_loss=0.2757, pruned_loss=0.1774, over 13925.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.2399, pruned_loss=0.1342, over 1962751.93 frames. ], batch size: 23, lr: 3.39e-02, grad_scale: 8.0 2022-12-07 06:46:26,066 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.627e+02 3.267e+02 4.651e+02 6.782e+02 1.677e+03, threshold=9.302e+02, percent-clipped=10.0 2022-12-07 06:46:46,542 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.7384, 0.7762, 0.6303, 0.8544, 1.0725, 0.4814, 0.9986, 0.7300], device='cuda:3'), covar=tensor([0.0345, 0.0409, 0.0206, 0.0346, 0.0215, 0.0232, 0.0273, 0.0221], device='cuda:3'), in_proj_covar=tensor([0.0021, 0.0022, 0.0024, 0.0022, 0.0019, 0.0023, 0.0018, 0.0019], device='cuda:3'), out_proj_covar=tensor([4.0218e-05, 4.4082e-05, 4.2358e-05, 4.2075e-05, 3.4268e-05, 4.1884e-05, 3.4819e-05, 3.4430e-05], device='cuda:3') 2022-12-07 06:46:49,933 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1249, 1.8773, 2.0855, 2.2906, 2.3940, 2.2109, 2.2351, 2.1187], device='cuda:3'), covar=tensor([0.0257, 0.1410, 0.0098, 0.0414, 0.0119, 0.0264, 0.0308, 0.1302], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0243, 0.0097, 0.0166, 0.0106, 0.0137, 0.0118, 0.0284], device='cuda:3'), out_proj_covar=tensor([1.1033e-04, 2.0187e-04, 7.8509e-05, 1.3716e-04, 9.1166e-05, 1.1148e-04, 1.1805e-04, 2.3113e-04], device='cuda:3') 2022-12-07 06:47:13,307 INFO [train.py:873] (3/4) Epoch 2, batch 1700, loss[loss=0.2309, simple_loss=0.1917, pruned_loss=0.135, over 2607.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.239, pruned_loss=0.1324, over 1971699.99 frames. ], batch size: 100, lr: 3.38e-02, grad_scale: 8.0 2022-12-07 06:47:30,407 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9283.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:47:53,043 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=13.77 vs. limit=5.0 2022-12-07 06:47:53,364 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.693e+02 2.995e+02 4.223e+02 5.322e+02 8.026e+02, threshold=8.446e+02, percent-clipped=0.0 2022-12-07 06:47:54,485 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1543, 1.2813, 0.4515, 0.8522, 1.3294, 1.1968, 0.9467, 1.3912], device='cuda:3'), covar=tensor([0.0549, 0.1149, 0.0713, 0.0528, 0.0516, 0.0604, 0.0385, 0.0302], device='cuda:3'), in_proj_covar=tensor([0.0014, 0.0016, 0.0017, 0.0017, 0.0016, 0.0017, 0.0017, 0.0018], device='cuda:3'), out_proj_covar=tensor([2.4258e-05, 2.6135e-05, 2.9541e-05, 2.5643e-05, 2.7075e-05, 2.5466e-05, 3.0190e-05, 2.8023e-05], device='cuda:3') 2022-12-07 06:48:12,956 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.7681, 0.6419, 0.1816, 0.5777, 0.7633, 0.1610, 0.7707, 0.6711], device='cuda:3'), covar=tensor([0.0071, 0.0071, 0.0052, 0.0159, 0.0072, 0.0053, 0.0056, 0.0072], device='cuda:3'), in_proj_covar=tensor([0.0022, 0.0023, 0.0024, 0.0022, 0.0021, 0.0023, 0.0019, 0.0020], device='cuda:3'), out_proj_covar=tensor([4.1670e-05, 4.6793e-05, 4.3253e-05, 4.3418e-05, 3.6813e-05, 4.2808e-05, 3.7044e-05, 3.6073e-05], device='cuda:3') 2022-12-07 06:48:20,826 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8639, 0.8372, 0.6245, 0.6053, 0.9721, 0.5391, 0.9988, 0.9237], device='cuda:3'), covar=tensor([0.0855, 0.0723, 0.0357, 0.1377, 0.0784, 0.0328, 0.0343, 0.0299], device='cuda:3'), in_proj_covar=tensor([0.0021, 0.0023, 0.0024, 0.0021, 0.0020, 0.0023, 0.0018, 0.0019], device='cuda:3'), out_proj_covar=tensor([4.0815e-05, 4.5789e-05, 4.2482e-05, 4.2317e-05, 3.5845e-05, 4.1897e-05, 3.5993e-05, 3.5293e-05], device='cuda:3') 2022-12-07 06:48:21,906 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9342.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:48:23,262 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.95 vs. limit=2.0 2022-12-07 06:48:25,773 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.76 vs. limit=5.0 2022-12-07 06:48:39,299 INFO [train.py:873] (3/4) Epoch 2, batch 1800, loss[loss=0.2298, simple_loss=0.2262, pruned_loss=0.1167, over 14293.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.2384, pruned_loss=0.1313, over 1979044.72 frames. ], batch size: 39, lr: 3.37e-02, grad_scale: 8.0 2022-12-07 06:48:44,737 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9369.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:49:19,721 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.290e+02 2.949e+02 3.801e+02 5.110e+02 1.024e+03, threshold=7.602e+02, percent-clipped=3.0 2022-12-07 06:49:24,151 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9414.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:49:25,782 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9416.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:50:07,242 INFO [train.py:873] (3/4) Epoch 2, batch 1900, loss[loss=0.2388, simple_loss=0.2344, pruned_loss=0.1216, over 14138.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.2376, pruned_loss=0.1313, over 1966572.88 frames. ], batch size: 84, lr: 3.35e-02, grad_scale: 8.0 2022-12-07 06:50:07,740 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4200, 1.4681, 2.6381, 2.1517, 2.5617, 1.6185, 2.1337, 2.3336], device='cuda:3'), covar=tensor([0.0346, 0.3979, 0.0237, 0.3832, 0.0218, 0.2928, 0.0764, 0.0191], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0280, 0.0145, 0.0393, 0.0122, 0.0305, 0.0202, 0.0128], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0001, 0.0003, 0.0001, 0.0003, 0.0002, 0.0001], device='cuda:3') 2022-12-07 06:50:08,451 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9464.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:50:17,807 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9475.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:50:19,063 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2022-12-07 06:50:46,648 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.598e+02 2.987e+02 3.780e+02 5.318e+02 1.239e+03, threshold=7.560e+02, percent-clipped=8.0 2022-12-07 06:50:53,533 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.92 vs. limit=2.0 2022-12-07 06:51:07,340 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.0315, 0.9855, 1.1115, 0.8859, 0.7378, 1.1984, 0.9339, 1.0090], device='cuda:3'), covar=tensor([0.0761, 0.0856, 0.0352, 0.0816, 0.0973, 0.0352, 0.0298, 0.0471], device='cuda:3'), in_proj_covar=tensor([0.0014, 0.0017, 0.0016, 0.0018, 0.0018, 0.0017, 0.0016, 0.0017], device='cuda:3'), out_proj_covar=tensor([2.3763e-05, 2.6102e-05, 2.8779e-05, 2.6728e-05, 2.9786e-05, 2.6902e-05, 2.9614e-05, 2.7901e-05], device='cuda:3') 2022-12-07 06:51:09,138 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5628, 3.5380, 2.4636, 3.7170, 2.9796, 3.3704, 3.0825, 2.6510], device='cuda:3'), covar=tensor([0.0258, 0.0350, 0.2474, 0.0193, 0.0199, 0.0183, 0.0462, 0.2266], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0166, 0.0269, 0.0123, 0.0119, 0.0123, 0.0165, 0.0303], device='cuda:3'), out_proj_covar=tensor([9.7421e-05, 1.1608e-04, 1.7866e-04, 8.3458e-05, 8.8566e-05, 8.7490e-05, 1.1788e-04, 1.9720e-04], device='cuda:3') 2022-12-07 06:51:33,030 INFO [train.py:873] (3/4) Epoch 2, batch 2000, loss[loss=0.2488, simple_loss=0.237, pruned_loss=0.1303, over 12760.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.2381, pruned_loss=0.1315, over 2021215.35 frames. ], batch size: 100, lr: 3.34e-02, grad_scale: 8.0 2022-12-07 06:51:50,810 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9583.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:51:51,636 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9584.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:52:12,819 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.740e+02 3.149e+02 4.158e+02 5.178e+02 1.006e+03, threshold=8.316e+02, percent-clipped=10.0 2022-12-07 06:52:13,795 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1266, 0.7634, 0.2544, 1.0170, 1.1406, 1.0372, 1.2288, 0.8637], device='cuda:3'), covar=tensor([0.0564, 0.0960, 0.0342, 0.0659, 0.0451, 0.0280, 0.0429, 0.0745], device='cuda:3'), in_proj_covar=tensor([0.0021, 0.0024, 0.0025, 0.0021, 0.0019, 0.0023, 0.0018, 0.0021], device='cuda:3'), out_proj_covar=tensor([4.2463e-05, 4.8610e-05, 4.3493e-05, 4.1180e-05, 3.4585e-05, 4.3149e-05, 3.5986e-05, 3.8917e-05], device='cuda:3') 2022-12-07 06:52:22,320 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.92 vs. limit=2.0 2022-12-07 06:52:32,382 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9631.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:52:41,417 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9642.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:52:43,972 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9645.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:52:54,908 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9658.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:52:59,449 INFO [train.py:873] (3/4) Epoch 2, batch 2100, loss[loss=0.2263, simple_loss=0.1889, pruned_loss=0.1319, over 2678.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.2365, pruned_loss=0.1304, over 1973319.30 frames. ], batch size: 100, lr: 3.32e-02, grad_scale: 8.0 2022-12-07 06:53:05,107 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9669.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:53:22,855 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9690.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:53:39,633 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.662e+02 3.245e+02 4.428e+02 6.131e+02 1.111e+03, threshold=8.856e+02, percent-clipped=8.0 2022-12-07 06:53:46,125 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9717.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:53:48,064 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9719.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:54:09,275 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9743.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:54:26,279 INFO [train.py:873] (3/4) Epoch 2, batch 2200, loss[loss=0.2528, simple_loss=0.2238, pruned_loss=0.1409, over 4963.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.2367, pruned_loss=0.1305, over 2047018.10 frames. ], batch size: 100, lr: 3.31e-02, grad_scale: 8.0 2022-12-07 06:54:32,303 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9770.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:54:42,845 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7880, 2.8632, 4.2660, 3.5478, 3.7900, 2.9390, 3.8259, 2.8162], device='cuda:3'), covar=tensor([0.0141, 0.0313, 0.0135, 0.0239, 0.0160, 0.0693, 0.0054, 0.0609], device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0105, 0.0080, 0.0097, 0.0097, 0.0153, 0.0062, 0.0145], device='cuda:3'), out_proj_covar=tensor([1.0004e-04, 1.2734e-04, 1.0186e-04, 1.2168e-04, 1.1744e-04, 1.8339e-04, 7.2128e-05, 1.6614e-04], device='cuda:3') 2022-12-07 06:54:53,338 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 2022-12-07 06:55:01,599 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9804.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:55:05,584 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.644e+01 3.042e+02 4.179e+02 5.699e+02 1.039e+03, threshold=8.358e+02, percent-clipped=2.0 2022-12-07 06:55:07,073 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.16 vs. limit=5.0 2022-12-07 06:55:41,278 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.03 vs. limit=2.0 2022-12-07 06:55:53,425 INFO [train.py:873] (3/4) Epoch 2, batch 2300, loss[loss=0.2411, simple_loss=0.2371, pruned_loss=0.1225, over 14287.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.2361, pruned_loss=0.1294, over 2028131.47 frames. ], batch size: 25, lr: 3.30e-02, grad_scale: 8.0 2022-12-07 06:56:33,304 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.330e+02 3.202e+02 3.964e+02 5.237e+02 8.345e+02, threshold=7.929e+02, percent-clipped=0.0 2022-12-07 06:57:00,526 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9940.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:57:05,187 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.99 vs. limit=2.0 2022-12-07 06:57:20,361 INFO [train.py:873] (3/4) Epoch 2, batch 2400, loss[loss=0.232, simple_loss=0.2298, pruned_loss=0.1171, over 13971.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.2366, pruned_loss=0.1298, over 2040397.70 frames. ], batch size: 19, lr: 3.28e-02, grad_scale: 8.0 2022-12-07 06:58:03,556 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.570e+02 3.087e+02 4.289e+02 5.805e+02 1.264e+03, threshold=8.579e+02, percent-clipped=4.0 2022-12-07 06:58:07,923 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10014.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:58:18,618 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6307, 1.5934, 2.6919, 1.3440, 1.6813, 1.5006, 0.8744, 1.5618], device='cuda:3'), covar=tensor([0.1168, 0.0946, 0.0414, 0.0807, 0.0562, 0.0829, 0.1922, 0.1094], device='cuda:3'), in_proj_covar=tensor([0.0054, 0.0053, 0.0050, 0.0055, 0.0053, 0.0049, 0.0095, 0.0058], device='cuda:3'), out_proj_covar=tensor([7.9201e-05, 7.5677e-05, 7.2698e-05, 7.5067e-05, 7.0324e-05, 7.0701e-05, 1.3269e-04, 8.2002e-05], device='cuda:3') 2022-12-07 06:58:30,078 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10039.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:58:41,354 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2022-12-07 06:58:50,301 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10062.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 06:58:50,992 INFO [train.py:873] (3/4) Epoch 2, batch 2500, loss[loss=0.2473, simple_loss=0.2338, pruned_loss=0.1304, over 6009.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.2355, pruned_loss=0.1284, over 2045491.89 frames. ], batch size: 100, lr: 3.27e-02, grad_scale: 8.0 2022-12-07 06:58:57,319 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10070.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:59:21,283 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10099.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:59:22,476 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10100.0, num_to_drop=1, layers_to_drop={3} 2022-12-07 06:59:30,092 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.651e+02 3.039e+02 4.037e+02 4.970e+02 1.749e+03, threshold=8.074e+02, percent-clipped=6.0 2022-12-07 06:59:35,487 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6343, 1.2846, 1.4494, 1.2574, 1.1177, 1.6150, 1.5339, 1.2232], device='cuda:3'), covar=tensor([0.4944, 0.1786, 0.1637, 0.1658, 0.0777, 0.0573, 0.0896, 0.0960], device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0055, 0.0042, 0.0044, 0.0054, 0.0040, 0.0055, 0.0056], device='cuda:3'), out_proj_covar=tensor([2.5237e-04, 1.2745e-04, 1.0748e-04, 1.0185e-04, 1.1304e-04, 9.4586e-05, 1.2977e-04, 1.1953e-04], device='cuda:3') 2022-12-07 06:59:37,901 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10118.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 06:59:39,241 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.48 vs. limit=5.0 2022-12-07 06:59:42,197 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10123.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 06:59:49,827 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6907, 1.4330, 3.3583, 1.6964, 3.4561, 3.3159, 2.4363, 3.9093], device='cuda:3'), covar=tensor([0.0150, 0.2544, 0.0362, 0.2120, 0.0251, 0.0304, 0.0762, 0.0117], device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0123, 0.0076, 0.0134, 0.0094, 0.0081, 0.0073, 0.0076], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 07:00:16,956 INFO [train.py:873] (3/4) Epoch 2, batch 2600, loss[loss=0.2354, simple_loss=0.2087, pruned_loss=0.131, over 3839.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.2352, pruned_loss=0.1282, over 2032964.35 frames. ], batch size: 100, lr: 3.26e-02, grad_scale: 8.0 2022-12-07 07:00:30,505 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.09 vs. limit=2.0 2022-12-07 07:00:42,080 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8785, 2.5329, 4.7506, 3.6102, 4.6543, 2.3197, 3.9771, 4.5492], device='cuda:3'), covar=tensor([0.0104, 0.3704, 0.0089, 0.4807, 0.0060, 0.2780, 0.0393, 0.0109], device='cuda:3'), in_proj_covar=tensor([0.0183, 0.0286, 0.0147, 0.0395, 0.0123, 0.0309, 0.0209, 0.0135], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0001, 0.0004, 0.0001, 0.0003, 0.0002, 0.0001], device='cuda:3') 2022-12-07 07:00:55,758 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.05 vs. limit=2.0 2022-12-07 07:00:56,780 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.626e+02 3.080e+02 4.222e+02 5.388e+02 1.025e+03, threshold=8.444e+02, percent-clipped=5.0 2022-12-07 07:01:24,250 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10240.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:01:44,394 INFO [train.py:873] (3/4) Epoch 2, batch 2700, loss[loss=0.2091, simple_loss=0.2105, pruned_loss=0.1039, over 14232.00 frames. ], tot_loss[loss=0.2453, simple_loss=0.2347, pruned_loss=0.128, over 2021676.28 frames. ], batch size: 37, lr: 3.24e-02, grad_scale: 8.0 2022-12-07 07:02:06,317 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10288.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:02:21,345 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=16.50 vs. limit=5.0 2022-12-07 07:02:24,380 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.544e+02 3.215e+02 3.934e+02 5.144e+02 9.049e+02, threshold=7.869e+02, percent-clipped=4.0 2022-12-07 07:02:28,983 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10314.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:03:09,546 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10362.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:03:10,326 INFO [train.py:873] (3/4) Epoch 2, batch 2800, loss[loss=0.2515, simple_loss=0.2415, pruned_loss=0.1307, over 5979.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.2347, pruned_loss=0.1284, over 1938033.33 frames. ], batch size: 100, lr: 3.23e-02, grad_scale: 8.0 2022-12-07 07:03:38,848 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10395.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 07:03:42,177 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10399.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:03:47,367 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.5918, 1.1798, 1.2462, 1.2461, 1.1654, 1.3829, 1.3215, 1.1747], device='cuda:3'), covar=tensor([0.2534, 0.1019, 0.0850, 0.0558, 0.0683, 0.0416, 0.0763, 0.0676], device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0055, 0.0043, 0.0042, 0.0054, 0.0039, 0.0050, 0.0057], device='cuda:3'), out_proj_covar=tensor([2.5561e-04, 1.2858e-04, 1.1155e-04, 1.0170e-04, 1.1731e-04, 9.5906e-05, 1.2232e-04, 1.2291e-04], device='cuda:3') 2022-12-07 07:03:50,604 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.391e+02 3.431e+02 4.642e+02 5.590e+02 1.030e+03, threshold=9.284e+02, percent-clipped=5.0 2022-12-07 07:03:50,817 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10409.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:03:56,062 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7369, 3.0757, 4.2081, 3.7115, 3.9349, 2.8402, 3.8186, 2.7171], device='cuda:3'), covar=tensor([0.0123, 0.0245, 0.0130, 0.0190, 0.0082, 0.0573, 0.0044, 0.0483], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0112, 0.0086, 0.0106, 0.0100, 0.0158, 0.0065, 0.0150], device='cuda:3'), out_proj_covar=tensor([1.0596e-04, 1.3852e-04, 1.1537e-04, 1.3369e-04, 1.2443e-04, 1.9634e-04, 7.7340e-05, 1.7716e-04], device='cuda:3') 2022-12-07 07:03:58,244 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10418.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 07:04:08,991 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1630, 2.8791, 2.8552, 2.8090, 3.0080, 2.9025, 3.1152, 3.0306], device='cuda:3'), covar=tensor([0.0573, 0.0615, 0.0841, 0.1003, 0.0477, 0.0516, 0.0768, 0.0820], device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0110, 0.0138, 0.0122, 0.0114, 0.0125, 0.0147, 0.0116], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 07:04:21,444 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8882, 1.8296, 1.5202, 1.9732, 1.9192, 2.0354, 1.7200, 1.4636], device='cuda:3'), covar=tensor([0.0157, 0.0200, 0.0542, 0.0082, 0.0156, 0.0082, 0.0227, 0.0417], device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0177, 0.0283, 0.0126, 0.0125, 0.0127, 0.0169, 0.0305], device='cuda:3'), out_proj_covar=tensor([9.9956e-05, 1.2370e-04, 1.9063e-04, 8.4994e-05, 9.3254e-05, 9.2364e-05, 1.2269e-04, 2.0288e-04], device='cuda:3') 2022-12-07 07:04:23,704 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10447.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:04:29,888 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10454.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:04:37,271 INFO [train.py:873] (3/4) Epoch 2, batch 2900, loss[loss=0.2325, simple_loss=0.2227, pruned_loss=0.1212, over 6947.00 frames. ], tot_loss[loss=0.245, simple_loss=0.2342, pruned_loss=0.1279, over 1977845.03 frames. ], batch size: 100, lr: 3.22e-02, grad_scale: 8.0 2022-12-07 07:04:43,806 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10470.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:05:01,260 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1339, 2.2268, 4.9664, 4.3713, 4.8010, 5.2047, 4.6361, 5.2165], device='cuda:3'), covar=tensor([0.1288, 0.1283, 0.0068, 0.0083, 0.0066, 0.0045, 0.0105, 0.0045], device='cuda:3'), in_proj_covar=tensor([0.0109, 0.0117, 0.0062, 0.0079, 0.0068, 0.0069, 0.0058, 0.0058], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2022-12-07 07:05:18,327 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.484e+02 2.981e+02 3.992e+02 4.851e+02 1.210e+03, threshold=7.983e+02, percent-clipped=1.0 2022-12-07 07:05:23,818 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10515.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 07:06:05,152 INFO [train.py:873] (3/4) Epoch 2, batch 3000, loss[loss=0.2324, simple_loss=0.2178, pruned_loss=0.1235, over 5944.00 frames. ], tot_loss[loss=0.2426, simple_loss=0.2328, pruned_loss=0.1261, over 1903567.48 frames. ], batch size: 100, lr: 3.21e-02, grad_scale: 8.0 2022-12-07 07:06:05,152 INFO [train.py:896] (3/4) Computing validation loss 2022-12-07 07:06:13,212 INFO [train.py:905] (3/4) Epoch 2, validation: loss=0.1433, simple_loss=0.1828, pruned_loss=0.05186, over 857387.00 frames. 2022-12-07 07:06:13,213 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17680MB 2022-12-07 07:06:18,686 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.2972, 1.1390, 0.9227, 0.8989, 0.6976, 1.9081, 1.0129, 0.8498], device='cuda:3'), covar=tensor([0.0296, 0.0594, 0.0668, 0.0410, 0.1240, 0.0171, 0.0324, 0.0607], device='cuda:3'), in_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0017, 0.0016, 0.0017, 0.0016, 0.0018], device='cuda:3'), out_proj_covar=tensor([2.6680e-05, 2.7816e-05, 3.0315e-05, 2.7344e-05, 3.0099e-05, 2.6003e-05, 3.3670e-05, 3.0919e-05], device='cuda:3') 2022-12-07 07:06:54,389 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.643e+02 3.162e+02 4.298e+02 5.900e+02 1.176e+03, threshold=8.596e+02, percent-clipped=5.0 2022-12-07 07:07:36,678 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.69 vs. limit=2.0 2022-12-07 07:07:41,331 INFO [train.py:873] (3/4) Epoch 2, batch 3100, loss[loss=0.3131, simple_loss=0.2514, pruned_loss=0.1874, over 1256.00 frames. ], tot_loss[loss=0.2424, simple_loss=0.2326, pruned_loss=0.1261, over 1907363.58 frames. ], batch size: 100, lr: 3.19e-02, grad_scale: 8.0 2022-12-07 07:08:07,961 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.77 vs. limit=2.0 2022-12-07 07:08:09,624 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10695.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:08:19,838 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9291, 2.1272, 4.4913, 2.8923, 4.1044, 4.3431, 4.0921, 5.1233], device='cuda:3'), covar=tensor([0.0096, 0.2582, 0.0196, 0.1682, 0.0276, 0.0243, 0.0209, 0.0079], device='cuda:3'), in_proj_covar=tensor([0.0084, 0.0126, 0.0077, 0.0133, 0.0096, 0.0084, 0.0075, 0.0077], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 07:08:22,249 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.335e+02 3.394e+02 4.427e+02 5.401e+02 1.921e+03, threshold=8.855e+02, percent-clipped=7.0 2022-12-07 07:08:29,215 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10718.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 07:08:51,478 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10743.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:08:52,402 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10744.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:09:08,153 INFO [train.py:873] (3/4) Epoch 2, batch 3200, loss[loss=0.2392, simple_loss=0.2364, pruned_loss=0.121, over 14215.00 frames. ], tot_loss[loss=0.2413, simple_loss=0.2324, pruned_loss=0.1251, over 1993027.13 frames. ], batch size: 35, lr: 3.18e-02, grad_scale: 8.0 2022-12-07 07:09:10,181 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10765.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:09:10,986 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10766.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 07:09:18,367 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10774.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:09:36,609 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.85 vs. limit=2.0 2022-12-07 07:09:43,285 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1638, 2.1961, 1.4834, 1.5222, 1.6811, 1.7426, 1.9456, 1.6902], device='cuda:3'), covar=tensor([0.0276, 0.2541, 0.0681, 0.0993, 0.0647, 0.0460, 0.0783, 0.0609], device='cuda:3'), in_proj_covar=tensor([0.0048, 0.0127, 0.0060, 0.0086, 0.0052, 0.0056, 0.0050, 0.0051], device='cuda:3'), out_proj_covar=tensor([9.4198e-05, 2.2003e-04, 1.1416e-04, 1.6171e-04, 1.1089e-04, 1.1250e-04, 1.0796e-04, 1.0206e-04], device='cuda:3') 2022-12-07 07:09:45,278 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10805.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:09:49,654 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.759e+01 3.135e+02 4.113e+02 5.521e+02 8.692e+02, threshold=8.227e+02, percent-clipped=0.0 2022-12-07 07:09:50,114 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10810.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 07:10:11,348 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10835.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:10:34,933 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6319, 3.5296, 2.3208, 3.7969, 3.5190, 3.5586, 3.4588, 2.5537], device='cuda:3'), covar=tensor([0.0196, 0.0400, 0.4307, 0.0223, 0.0205, 0.0252, 0.0503, 0.3805], device='cuda:3'), in_proj_covar=tensor([0.0158, 0.0186, 0.0288, 0.0126, 0.0132, 0.0130, 0.0179, 0.0322], device='cuda:3'), out_proj_covar=tensor([1.0734e-04, 1.3187e-04, 1.9655e-04, 8.6301e-05, 9.7432e-05, 9.5075e-05, 1.3127e-04, 2.1488e-04], device='cuda:3') 2022-12-07 07:10:35,551 INFO [train.py:873] (3/4) Epoch 2, batch 3300, loss[loss=0.2483, simple_loss=0.2413, pruned_loss=0.1276, over 14363.00 frames. ], tot_loss[loss=0.243, simple_loss=0.2335, pruned_loss=0.1263, over 2001178.95 frames. ], batch size: 73, lr: 3.17e-02, grad_scale: 8.0 2022-12-07 07:10:37,256 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8185, 2.2473, 2.9804, 2.0390, 2.2864, 1.9153, 0.9284, 2.2515], device='cuda:3'), covar=tensor([0.1294, 0.0659, 0.0674, 0.0762, 0.0407, 0.0924, 0.1873, 0.0780], device='cuda:3'), in_proj_covar=tensor([0.0053, 0.0057, 0.0054, 0.0059, 0.0057, 0.0054, 0.0100, 0.0059], device='cuda:3'), out_proj_covar=tensor([8.1974e-05, 8.5641e-05, 8.0847e-05, 8.5524e-05, 7.9804e-05, 8.1474e-05, 1.4308e-04, 9.0131e-05], device='cuda:3') 2022-12-07 07:10:49,428 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1849, 1.8658, 2.0684, 2.1912, 2.1370, 2.1047, 2.2010, 1.8753], device='cuda:3'), covar=tensor([0.0351, 0.1140, 0.0415, 0.0458, 0.0547, 0.0514, 0.0612, 0.0617], device='cuda:3'), in_proj_covar=tensor([0.0073, 0.0136, 0.0086, 0.0082, 0.0090, 0.0092, 0.0128, 0.0103], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 07:10:54,186 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10885.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 07:11:15,637 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.702e+02 2.971e+02 4.122e+02 5.420e+02 1.001e+03, threshold=8.244e+02, percent-clipped=4.0 2022-12-07 07:11:29,888 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1135, 1.8777, 5.0708, 4.5981, 4.5461, 4.7583, 4.5718, 5.3128], device='cuda:3'), covar=tensor([0.1496, 0.1541, 0.0064, 0.0096, 0.0091, 0.0111, 0.0107, 0.0043], device='cuda:3'), in_proj_covar=tensor([0.0111, 0.0118, 0.0062, 0.0079, 0.0069, 0.0070, 0.0059, 0.0059], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2022-12-07 07:11:38,334 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.06 vs. limit=2.0 2022-12-07 07:11:46,491 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10946.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 07:11:48,879 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.0832, 4.7922, 5.2176, 4.2504, 4.9835, 5.3215, 2.4513, 4.9150], device='cuda:3'), covar=tensor([0.0182, 0.0260, 0.0390, 0.0360, 0.0205, 0.0088, 0.2569, 0.0213], device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0091, 0.0090, 0.0074, 0.0116, 0.0081, 0.0130, 0.0109], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 07:12:00,256 INFO [train.py:873] (3/4) Epoch 2, batch 3400, loss[loss=0.2237, simple_loss=0.1845, pruned_loss=0.1315, over 1285.00 frames. ], tot_loss[loss=0.2403, simple_loss=0.2319, pruned_loss=0.1243, over 2007814.87 frames. ], batch size: 100, lr: 3.16e-02, grad_scale: 8.0 2022-12-07 07:12:42,275 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.436e+02 2.877e+02 3.778e+02 5.417e+02 1.188e+03, threshold=7.556e+02, percent-clipped=7.0 2022-12-07 07:13:28,831 INFO [train.py:873] (3/4) Epoch 2, batch 3500, loss[loss=0.22, simple_loss=0.2277, pruned_loss=0.1061, over 14316.00 frames. ], tot_loss[loss=0.2402, simple_loss=0.2316, pruned_loss=0.1244, over 1969390.60 frames. ], batch size: 31, lr: 3.15e-02, grad_scale: 8.0 2022-12-07 07:13:30,351 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11065.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:13:40,640 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.5539, 1.1762, 1.9970, 1.8974, 2.0028, 1.8995, 1.4599, 2.0987], device='cuda:3'), covar=tensor([0.0498, 0.0996, 0.0117, 0.0213, 0.0131, 0.0107, 0.0393, 0.0080], device='cuda:3'), in_proj_covar=tensor([0.0113, 0.0122, 0.0065, 0.0082, 0.0070, 0.0071, 0.0060, 0.0061], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2022-12-07 07:14:00,518 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11100.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:14:08,559 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.123e+02 3.159e+02 4.112e+02 5.380e+02 8.623e+02, threshold=8.224e+02, percent-clipped=2.0 2022-12-07 07:14:08,749 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11110.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 07:14:11,089 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11113.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:14:24,224 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4552, 1.5906, 2.6530, 2.1332, 2.4102, 1.6913, 1.9924, 2.6063], device='cuda:3'), covar=tensor([0.0398, 0.3033, 0.0204, 0.3432, 0.0196, 0.2281, 0.0839, 0.0158], device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0297, 0.0149, 0.0401, 0.0129, 0.0309, 0.0226, 0.0136], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0001, 0.0004, 0.0001, 0.0003, 0.0002, 0.0001], device='cuda:3') 2022-12-07 07:14:26,063 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11130.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:14:50,093 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11158.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:14:53,254 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.15 vs. limit=2.0 2022-12-07 07:14:54,228 INFO [train.py:873] (3/4) Epoch 2, batch 3600, loss[loss=0.2579, simple_loss=0.2433, pruned_loss=0.1362, over 14458.00 frames. ], tot_loss[loss=0.2403, simple_loss=0.2316, pruned_loss=0.1245, over 1931670.10 frames. ], batch size: 51, lr: 3.13e-02, grad_scale: 8.0 2022-12-07 07:15:04,176 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11174.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:15:33,347 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.10 vs. limit=2.0 2022-12-07 07:15:35,179 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.191e+02 2.817e+02 3.568e+02 4.695e+02 1.140e+03, threshold=7.135e+02, percent-clipped=4.0 2022-12-07 07:15:56,856 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11235.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:16:02,101 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11241.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 07:16:21,364 INFO [train.py:873] (3/4) Epoch 2, batch 3700, loss[loss=0.1827, simple_loss=0.1631, pruned_loss=0.1011, over 2614.00 frames. ], tot_loss[loss=0.2393, simple_loss=0.2307, pruned_loss=0.124, over 1919569.66 frames. ], batch size: 100, lr: 3.12e-02, grad_scale: 8.0 2022-12-07 07:16:44,933 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8023, 1.4214, 1.7414, 1.1322, 1.5221, 1.6706, 1.3980, 1.5780], device='cuda:3'), covar=tensor([0.0379, 0.4085, 0.0639, 0.1450, 0.0691, 0.0502, 0.1530, 0.0528], device='cuda:3'), in_proj_covar=tensor([0.0051, 0.0132, 0.0062, 0.0091, 0.0056, 0.0059, 0.0052, 0.0053], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2022-12-07 07:16:51,365 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2022-12-07 07:17:02,054 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.068e+02 3.008e+02 3.803e+02 5.016e+02 1.141e+03, threshold=7.605e+02, percent-clipped=8.0 2022-12-07 07:17:20,713 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.78 vs. limit=5.0 2022-12-07 07:17:47,319 INFO [train.py:873] (3/4) Epoch 2, batch 3800, loss[loss=0.1775, simple_loss=0.1526, pruned_loss=0.1012, over 1260.00 frames. ], tot_loss[loss=0.2386, simple_loss=0.2302, pruned_loss=0.1235, over 1917811.09 frames. ], batch size: 100, lr: 3.11e-02, grad_scale: 8.0 2022-12-07 07:18:06,038 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.0439, 0.8042, 0.9078, 0.7618, 1.1422, 1.1881, 1.5986, 0.9025], device='cuda:3'), covar=tensor([0.0470, 0.0460, 0.0216, 0.0910, 0.0220, 0.0250, 0.0180, 0.0464], device='cuda:3'), in_proj_covar=tensor([0.0018, 0.0020, 0.0021, 0.0020, 0.0019, 0.0022, 0.0016, 0.0019], device='cuda:3'), out_proj_covar=tensor([4.3034e-05, 4.7151e-05, 4.3319e-05, 4.4823e-05, 3.8629e-05, 4.5529e-05, 3.7770e-05, 4.0003e-05], device='cuda:3') 2022-12-07 07:18:20,007 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11400.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:18:28,499 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.452e+02 3.176e+02 4.502e+02 5.921e+02 1.352e+03, threshold=9.005e+02, percent-clipped=11.0 2022-12-07 07:18:37,012 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6936, 2.3334, 4.0543, 3.8740, 3.5314, 2.6468, 3.8178, 2.9459], device='cuda:3'), covar=tensor([0.0125, 0.0326, 0.0140, 0.0162, 0.0149, 0.0636, 0.0044, 0.0468], device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0125, 0.0105, 0.0118, 0.0107, 0.0168, 0.0072, 0.0161], device='cuda:3'), out_proj_covar=tensor([1.1842e-04, 1.6139e-04, 1.4603e-04, 1.5525e-04, 1.4001e-04, 2.2260e-04, 9.1139e-05, 1.9808e-04], device='cuda:3') 2022-12-07 07:18:37,315 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.94 vs. limit=2.0 2022-12-07 07:18:43,172 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=9.63 vs. limit=5.0 2022-12-07 07:18:46,222 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11430.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:19:02,208 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11448.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:19:14,043 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2022-12-07 07:19:15,325 INFO [train.py:873] (3/4) Epoch 2, batch 3900, loss[loss=0.2573, simple_loss=0.2306, pruned_loss=0.142, over 7795.00 frames. ], tot_loss[loss=0.2382, simple_loss=0.2298, pruned_loss=0.1233, over 1890996.37 frames. ], batch size: 100, lr: 3.10e-02, grad_scale: 8.0 2022-12-07 07:19:16,649 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.09 vs. limit=2.0 2022-12-07 07:19:18,252 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.03 vs. limit=2.0 2022-12-07 07:19:27,735 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11478.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:19:55,407 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.433e+02 2.890e+02 4.018e+02 5.176e+02 1.364e+03, threshold=8.037e+02, percent-clipped=4.0 2022-12-07 07:20:12,869 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11530.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:20:14,721 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11532.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:20:22,706 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11541.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 07:20:40,966 INFO [train.py:873] (3/4) Epoch 2, batch 4000, loss[loss=0.2509, simple_loss=0.2436, pruned_loss=0.1291, over 14151.00 frames. ], tot_loss[loss=0.2385, simple_loss=0.2302, pruned_loss=0.1234, over 1914842.86 frames. ], batch size: 84, lr: 3.09e-02, grad_scale: 8.0 2022-12-07 07:21:03,659 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11589.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 07:21:07,077 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11593.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:21:07,915 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8757, 1.5005, 1.4225, 1.3487, 1.0458, 1.4241, 1.2832, 1.0397], device='cuda:3'), covar=tensor([0.2629, 0.0993, 0.1368, 0.1634, 0.0951, 0.0520, 0.1117, 0.1046], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0057, 0.0044, 0.0047, 0.0059, 0.0044, 0.0057, 0.0061], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2022-12-07 07:21:22,285 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.261e+02 3.113e+02 4.062e+02 5.460e+02 1.002e+03, threshold=8.123e+02, percent-clipped=4.0 2022-12-07 07:21:23,354 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8310, 3.5154, 3.4740, 3.4082, 3.6629, 3.5577, 3.7577, 3.7736], device='cuda:3'), covar=tensor([0.0478, 0.0475, 0.0565, 0.0973, 0.0308, 0.0400, 0.0636, 0.0613], device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0122, 0.0152, 0.0158, 0.0123, 0.0138, 0.0153, 0.0126], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 07:21:44,134 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11635.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:22:08,542 INFO [train.py:873] (3/4) Epoch 2, batch 4100, loss[loss=0.2437, simple_loss=0.2097, pruned_loss=0.1389, over 2611.00 frames. ], tot_loss[loss=0.2398, simple_loss=0.2312, pruned_loss=0.1242, over 1958643.48 frames. ], batch size: 100, lr: 3.08e-02, grad_scale: 8.0 2022-12-07 07:22:09,149 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.07 vs. limit=2.0 2022-12-07 07:22:10,012 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.66 vs. limit=5.0 2022-12-07 07:22:37,398 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11696.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:22:39,992 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11699.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:22:49,365 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.791e+01 3.003e+02 4.031e+02 5.512e+02 8.434e+02, threshold=8.062e+02, percent-clipped=2.0 2022-12-07 07:22:55,674 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.59 vs. limit=2.0 2022-12-07 07:23:32,931 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11760.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:23:35,181 INFO [train.py:873] (3/4) Epoch 2, batch 4200, loss[loss=0.2818, simple_loss=0.2288, pruned_loss=0.1674, over 1211.00 frames. ], tot_loss[loss=0.2384, simple_loss=0.2307, pruned_loss=0.123, over 1979290.26 frames. ], batch size: 100, lr: 3.07e-02, grad_scale: 8.0 2022-12-07 07:24:16,691 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.779e+02 3.300e+02 4.114e+02 5.198e+02 1.405e+03, threshold=8.228e+02, percent-clipped=5.0 2022-12-07 07:24:22,330 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11816.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:24:34,008 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11830.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:24:47,687 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11846.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:25:02,802 INFO [train.py:873] (3/4) Epoch 2, batch 4300, loss[loss=0.2465, simple_loss=0.235, pruned_loss=0.129, over 14278.00 frames. ], tot_loss[loss=0.2368, simple_loss=0.2296, pruned_loss=0.122, over 1951046.11 frames. ], batch size: 44, lr: 3.06e-02, grad_scale: 8.0 2022-12-07 07:25:15,080 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11877.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:25:15,834 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11878.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:25:24,626 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11888.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:25:41,729 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11907.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:25:44,083 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.554e+02 3.023e+02 3.927e+02 4.940e+02 9.062e+02, threshold=7.854e+02, percent-clipped=2.0 2022-12-07 07:25:52,712 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11920.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:26:17,488 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11948.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:26:27,731 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4927, 2.1352, 2.6815, 2.4761, 2.6651, 2.5057, 2.7552, 2.2853], device='cuda:3'), covar=tensor([0.0134, 0.0907, 0.0053, 0.0375, 0.0113, 0.0165, 0.0211, 0.0901], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0282, 0.0118, 0.0213, 0.0144, 0.0172, 0.0160, 0.0325], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0003, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0003], device='cuda:3') 2022-12-07 07:26:30,062 INFO [train.py:873] (3/4) Epoch 2, batch 4400, loss[loss=0.1864, simple_loss=0.1662, pruned_loss=0.1033, over 2715.00 frames. ], tot_loss[loss=0.2361, simple_loss=0.2291, pruned_loss=0.1216, over 1975151.46 frames. ], batch size: 100, lr: 3.04e-02, grad_scale: 8.0 2022-12-07 07:26:46,388 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11981.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:26:48,654 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11984.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:26:54,445 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11991.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:27:10,730 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12009.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:27:11,348 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.252e+02 2.977e+02 4.090e+02 5.341e+02 9.805e+02, threshold=8.180e+02, percent-clipped=2.0 2022-12-07 07:27:41,933 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12045.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:27:51,278 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12055.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:27:57,858 INFO [train.py:873] (3/4) Epoch 2, batch 4500, loss[loss=0.2256, simple_loss=0.2237, pruned_loss=0.1137, over 13888.00 frames. ], tot_loss[loss=0.2357, simple_loss=0.2285, pruned_loss=0.1214, over 1939010.04 frames. ], batch size: 20, lr: 3.03e-02, grad_scale: 8.0 2022-12-07 07:28:09,030 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12076.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:28:13,871 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9160, 3.4550, 3.6370, 3.9859, 3.6139, 2.7363, 3.9724, 3.9442], device='cuda:3'), covar=tensor([0.0541, 0.0778, 0.0544, 0.0476, 0.0719, 0.0705, 0.0487, 0.0528], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0071, 0.0091, 0.0087, 0.0097, 0.0064, 0.0084, 0.0087], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001], device='cuda:3') 2022-12-07 07:28:14,393 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.98 vs. limit=2.0 2022-12-07 07:28:15,629 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7696, 1.1859, 1.5536, 1.5123, 1.0509, 1.4539, 1.1140, 1.0757], device='cuda:3'), covar=tensor([0.3295, 0.1518, 0.1338, 0.1052, 0.1060, 0.0766, 0.1199, 0.1099], device='cuda:3'), in_proj_covar=tensor([0.0137, 0.0058, 0.0047, 0.0045, 0.0058, 0.0045, 0.0056, 0.0063], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2022-12-07 07:28:25,911 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.05 vs. limit=2.0 2022-12-07 07:28:38,201 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.235e+02 3.534e+02 4.689e+02 6.101e+02 2.207e+03, threshold=9.377e+02, percent-clipped=9.0 2022-12-07 07:29:01,458 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12137.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:29:23,549 INFO [train.py:873] (3/4) Epoch 2, batch 4600, loss[loss=0.2007, simple_loss=0.1957, pruned_loss=0.1028, over 4986.00 frames. ], tot_loss[loss=0.2388, simple_loss=0.2303, pruned_loss=0.1236, over 1925780.14 frames. ], batch size: 100, lr: 3.02e-02, grad_scale: 8.0 2022-12-07 07:29:31,615 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12172.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:29:37,402 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=7.69 vs. limit=5.0 2022-12-07 07:29:45,281 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12188.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:29:57,598 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12202.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:30:04,544 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.122e+02 2.710e+02 3.814e+02 4.807e+02 1.336e+03, threshold=7.627e+02, percent-clipped=4.0 2022-12-07 07:30:07,791 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.18 vs. limit=2.0 2022-12-07 07:30:27,144 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12236.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:30:50,578 INFO [train.py:873] (3/4) Epoch 2, batch 4700, loss[loss=0.2168, simple_loss=0.1932, pruned_loss=0.1202, over 3813.00 frames. ], tot_loss[loss=0.2377, simple_loss=0.2299, pruned_loss=0.1227, over 1939191.41 frames. ], batch size: 100, lr: 3.01e-02, grad_scale: 8.0 2022-12-07 07:30:50,947 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=3.27 vs. limit=2.0 2022-12-07 07:30:53,986 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6458, 1.4748, 2.0108, 2.0291, 1.4999, 1.5922, 1.8689, 1.8023], device='cuda:3'), covar=tensor([0.0057, 0.0048, 0.0030, 0.0027, 0.0040, 0.0073, 0.0022, 0.0031], device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0131, 0.0120, 0.0125, 0.0112, 0.0174, 0.0077, 0.0164], device='cuda:3'), out_proj_covar=tensor([1.3240e-04, 1.7418e-04, 1.6652e-04, 1.6749e-04, 1.4960e-04, 2.3754e-04, 9.6744e-05, 2.0492e-04], device='cuda:3') 2022-12-07 07:31:01,590 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12276.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:31:08,430 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2887, 3.0604, 2.2204, 3.4961, 2.9371, 3.4531, 3.0367, 2.5194], device='cuda:3'), covar=tensor([0.0162, 0.0428, 0.2470, 0.0125, 0.0186, 0.0159, 0.0488, 0.1985], device='cuda:3'), in_proj_covar=tensor([0.0161, 0.0198, 0.0307, 0.0140, 0.0146, 0.0142, 0.0195, 0.0322], device='cuda:3'), out_proj_covar=tensor([1.1233e-04, 1.4303e-04, 2.1355e-04, 9.7725e-05, 1.1052e-04, 1.0909e-04, 1.4930e-04, 2.1943e-04], device='cuda:3') 2022-12-07 07:31:14,705 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12291.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:31:25,444 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12304.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:31:30,461 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.134e+02 2.849e+02 3.785e+02 5.259e+02 1.272e+03, threshold=7.569e+02, percent-clipped=8.0 2022-12-07 07:31:55,998 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12339.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:31:56,787 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12340.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:31:56,847 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6342, 1.3122, 1.5273, 0.8694, 1.0943, 1.3334, 1.3484, 1.2946], device='cuda:3'), covar=tensor([0.0092, 0.0507, 0.0197, 0.0327, 0.0381, 0.0247, 0.0175, 0.0299], device='cuda:3'), in_proj_covar=tensor([0.0053, 0.0138, 0.0069, 0.0093, 0.0062, 0.0060, 0.0057, 0.0060], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0003, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2022-12-07 07:32:09,874 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12355.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:32:16,668 INFO [train.py:873] (3/4) Epoch 2, batch 4800, loss[loss=0.2472, simple_loss=0.2277, pruned_loss=0.1334, over 5998.00 frames. ], tot_loss[loss=0.2368, simple_loss=0.2296, pruned_loss=0.1221, over 1946202.28 frames. ], batch size: 100, lr: 3.00e-02, grad_scale: 8.0 2022-12-07 07:32:23,864 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.94 vs. limit=2.0 2022-12-07 07:32:28,915 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0202, 1.9387, 1.9893, 2.0227, 2.0348, 1.8533, 0.9278, 1.8243], device='cuda:3'), covar=tensor([0.0290, 0.0320, 0.0426, 0.0163, 0.0270, 0.0491, 0.1559, 0.0330], device='cuda:3'), in_proj_covar=tensor([0.0094, 0.0096, 0.0095, 0.0072, 0.0122, 0.0083, 0.0133, 0.0111], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 07:32:33,131 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.7953, 0.9395, 1.0164, 1.0129, 0.8747, 1.0167, 0.8396, 0.6786], device='cuda:3'), covar=tensor([0.1565, 0.0276, 0.0271, 0.0256, 0.0461, 0.0168, 0.0763, 0.0884], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0058, 0.0046, 0.0047, 0.0057, 0.0044, 0.0055, 0.0065], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002], device='cuda:3') 2022-12-07 07:32:51,091 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12403.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:32:57,419 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.788e+02 3.011e+02 4.103e+02 5.225e+02 9.734e+02, threshold=8.206e+02, percent-clipped=3.0 2022-12-07 07:33:10,259 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.98 vs. limit=2.0 2022-12-07 07:33:15,516 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12432.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:33:20,242 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.58 vs. limit=5.0 2022-12-07 07:33:27,380 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12446.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:33:28,153 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6430, 1.1152, 1.9165, 1.9080, 1.9380, 1.8193, 1.2713, 1.9968], device='cuda:3'), covar=tensor([0.0474, 0.0922, 0.0184, 0.0248, 0.0177, 0.0189, 0.0371, 0.0139], device='cuda:3'), in_proj_covar=tensor([0.0119, 0.0131, 0.0068, 0.0092, 0.0076, 0.0079, 0.0064, 0.0065], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001], device='cuda:3') 2022-12-07 07:33:42,230 INFO [train.py:873] (3/4) Epoch 2, batch 4900, loss[loss=0.2227, simple_loss=0.2287, pruned_loss=0.1083, over 14289.00 frames. ], tot_loss[loss=0.2358, simple_loss=0.2293, pruned_loss=0.1212, over 1947857.39 frames. ], batch size: 25, lr: 2.99e-02, grad_scale: 8.0 2022-12-07 07:33:49,687 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12472.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:33:57,655 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7508, 1.8521, 1.0851, 1.7817, 1.2955, 1.4149, 1.4989, 1.5079], device='cuda:3'), covar=tensor([0.0715, 0.1473, 0.0711, 0.0106, 0.0395, 0.0194, 0.0212, 0.0253], device='cuda:3'), in_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0014, 0.0013, 0.0016, 0.0014, 0.0015], device='cuda:3'), out_proj_covar=tensor([2.5183e-05, 2.7071e-05, 2.9737e-05, 2.6996e-05, 2.7521e-05, 2.8513e-05, 3.2309e-05, 2.9990e-05], device='cuda:3') 2022-12-07 07:34:15,564 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12502.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:34:19,794 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12507.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:34:21,912 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.645e+02 3.534e+02 4.435e+02 5.863e+02 9.969e+02, threshold=8.870e+02, percent-clipped=5.0 2022-12-07 07:34:30,739 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12520.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:34:46,869 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4227, 3.0890, 2.1131, 3.6598, 3.1547, 3.5592, 3.0340, 2.4335], device='cuda:3'), covar=tensor([0.0183, 0.0365, 0.2199, 0.0121, 0.0157, 0.0220, 0.0411, 0.2093], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0209, 0.0318, 0.0144, 0.0154, 0.0148, 0.0200, 0.0342], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:3') 2022-12-07 07:34:56,809 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12550.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:35:08,167 INFO [train.py:873] (3/4) Epoch 2, batch 5000, loss[loss=0.2496, simple_loss=0.2319, pruned_loss=0.1337, over 14010.00 frames. ], tot_loss[loss=0.2379, simple_loss=0.2303, pruned_loss=0.1228, over 1991224.97 frames. ], batch size: 22, lr: 2.98e-02, grad_scale: 8.0 2022-12-07 07:35:19,767 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12576.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:35:21,566 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 2022-12-07 07:35:25,030 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.89 vs. limit=2.0 2022-12-07 07:35:44,148 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12604.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:35:49,096 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.602e+02 3.073e+02 4.044e+02 5.211e+02 8.552e+02, threshold=8.087e+02, percent-clipped=0.0 2022-12-07 07:36:01,969 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12624.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:36:15,484 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12640.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:36:18,013 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.6483, 2.6414, 3.6597, 2.0985, 2.5458, 2.4328, 1.4664, 2.5522], device='cuda:3'), covar=tensor([0.1028, 0.1148, 0.0553, 0.1313, 0.0545, 0.1579, 0.2737, 0.1144], device='cuda:3'), in_proj_covar=tensor([0.0057, 0.0060, 0.0057, 0.0064, 0.0061, 0.0055, 0.0112, 0.0069], device='cuda:3'), out_proj_covar=tensor([9.8321e-05, 9.9637e-05, 9.4792e-05, 1.0183e-04, 9.5301e-05, 9.3240e-05, 1.7384e-04, 1.1568e-04], device='cuda:3') 2022-12-07 07:36:25,609 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12652.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:36:35,313 INFO [train.py:873] (3/4) Epoch 2, batch 5100, loss[loss=0.2527, simple_loss=0.2416, pruned_loss=0.1319, over 14278.00 frames. ], tot_loss[loss=0.236, simple_loss=0.2293, pruned_loss=0.1214, over 1968298.11 frames. ], batch size: 31, lr: 2.97e-02, grad_scale: 8.0 2022-12-07 07:36:56,657 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12688.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:36:58,251 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.1487, 4.5991, 4.5673, 5.1323, 4.8464, 4.4021, 5.1063, 4.4921], device='cuda:3'), covar=tensor([0.0356, 0.1158, 0.0314, 0.0423, 0.0607, 0.0455, 0.0543, 0.0445], device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0148, 0.0094, 0.0087, 0.0096, 0.0092, 0.0134, 0.0107], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 07:37:16,599 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.397e+02 3.113e+02 4.447e+02 5.356e+02 1.001e+03, threshold=8.894e+02, percent-clipped=4.0 2022-12-07 07:37:20,171 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9660, 2.5853, 3.9441, 2.8265, 3.6431, 3.5618, 3.3735, 2.9597], device='cuda:3'), covar=tensor([0.0086, 0.1183, 0.0040, 0.0629, 0.0157, 0.0206, 0.0619, 0.1072], device='cuda:3'), in_proj_covar=tensor([0.0179, 0.0304, 0.0120, 0.0231, 0.0165, 0.0192, 0.0177, 0.0338], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:3') 2022-12-07 07:37:34,960 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12732.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:37:46,785 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.80 vs. limit=5.0 2022-12-07 07:37:55,184 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.14 vs. limit=2.0 2022-12-07 07:38:01,732 INFO [train.py:873] (3/4) Epoch 2, batch 5200, loss[loss=0.2363, simple_loss=0.2278, pruned_loss=0.1224, over 14027.00 frames. ], tot_loss[loss=0.2338, simple_loss=0.2276, pruned_loss=0.12, over 1940516.79 frames. ], batch size: 22, lr: 2.96e-02, grad_scale: 8.0 2022-12-07 07:38:01,860 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12763.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:38:16,883 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12780.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:38:35,781 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12802.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:38:43,164 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.389e+02 3.055e+02 3.810e+02 5.193e+02 9.032e+02, threshold=7.619e+02, percent-clipped=1.0 2022-12-07 07:38:54,871 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12824.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 07:39:03,074 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.0790, 1.0763, 1.2682, 1.3151, 1.0795, 1.4350, 0.7077, 1.1909], device='cuda:3'), covar=tensor([0.0604, 0.2002, 0.0517, 0.0319, 0.0460, 0.0214, 0.0234, 0.1128], device='cuda:3'), in_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0016, 0.0014, 0.0016, 0.0016, 0.0016], device='cuda:3'), out_proj_covar=tensor([2.8429e-05, 3.0333e-05, 3.2316e-05, 3.0516e-05, 3.0251e-05, 2.9896e-05, 3.8055e-05, 3.3413e-05], device='cuda:3') 2022-12-07 07:39:28,089 INFO [train.py:873] (3/4) Epoch 2, batch 5300, loss[loss=0.2133, simple_loss=0.2182, pruned_loss=0.1042, over 14139.00 frames. ], tot_loss[loss=0.2348, simple_loss=0.2283, pruned_loss=0.1207, over 1975526.25 frames. ], batch size: 84, lr: 2.95e-02, grad_scale: 8.0 2022-12-07 07:39:30,771 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1332, 0.9295, 0.7556, 0.8383, 0.7860, 0.9832, 0.7999, 0.7619], device='cuda:3'), covar=tensor([0.0576, 0.1434, 0.1309, 0.0853, 0.0588, 0.0421, 0.0316, 0.0797], device='cuda:3'), in_proj_covar=tensor([0.0014, 0.0016, 0.0015, 0.0016, 0.0014, 0.0017, 0.0015, 0.0017], device='cuda:3'), out_proj_covar=tensor([2.8542e-05, 3.0758e-05, 3.2734e-05, 3.0770e-05, 3.0126e-05, 2.9954e-05, 3.8138e-05, 3.3591e-05], device='cuda:3') 2022-12-07 07:40:09,055 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.807e+01 2.959e+02 3.875e+02 4.793e+02 1.192e+03, threshold=7.749e+02, percent-clipped=2.0 2022-12-07 07:40:53,786 INFO [train.py:873] (3/4) Epoch 2, batch 5400, loss[loss=0.2077, simple_loss=0.213, pruned_loss=0.1012, over 14148.00 frames. ], tot_loss[loss=0.2341, simple_loss=0.2278, pruned_loss=0.1202, over 1967667.16 frames. ], batch size: 84, lr: 2.94e-02, grad_scale: 8.0 2022-12-07 07:41:03,625 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.3376, 2.4294, 3.8313, 1.9920, 2.6822, 2.1519, 1.0402, 2.7309], device='cuda:3'), covar=tensor([0.1766, 0.0573, 0.0384, 0.0952, 0.0407, 0.1275, 0.2260, 0.0964], device='cuda:3'), in_proj_covar=tensor([0.0059, 0.0058, 0.0051, 0.0060, 0.0060, 0.0053, 0.0109, 0.0066], device='cuda:3'), out_proj_covar=tensor([1.0243e-04, 9.6928e-05, 8.8303e-05, 1.0016e-04, 9.5356e-05, 9.2636e-05, 1.7263e-04, 1.1262e-04], device='cuda:3') 2022-12-07 07:41:35,175 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.219e+02 3.363e+02 4.433e+02 5.944e+02 1.364e+03, threshold=8.866e+02, percent-clipped=4.0 2022-12-07 07:41:45,631 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0698, 2.2281, 4.2566, 3.6920, 3.9463, 4.1919, 3.5452, 4.3136], device='cuda:3'), covar=tensor([0.1401, 0.1194, 0.0071, 0.0144, 0.0115, 0.0107, 0.0272, 0.0080], device='cuda:3'), in_proj_covar=tensor([0.0118, 0.0131, 0.0068, 0.0094, 0.0078, 0.0081, 0.0066, 0.0067], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001], device='cuda:3') 2022-12-07 07:41:58,348 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9423, 1.7705, 4.4997, 2.1763, 4.3053, 4.5586, 4.3812, 5.0268], device='cuda:3'), covar=tensor([0.0102, 0.2311, 0.0190, 0.1990, 0.0192, 0.0150, 0.0143, 0.0101], device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0133, 0.0088, 0.0145, 0.0108, 0.0092, 0.0081, 0.0085], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 07:42:21,076 INFO [train.py:873] (3/4) Epoch 2, batch 5500, loss[loss=0.2349, simple_loss=0.2322, pruned_loss=0.1189, over 13938.00 frames. ], tot_loss[loss=0.2325, simple_loss=0.227, pruned_loss=0.119, over 2026791.54 frames. ], batch size: 20, lr: 2.93e-02, grad_scale: 8.0 2022-12-07 07:42:54,585 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=13102.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:42:58,195 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 2022-12-07 07:43:01,929 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.454e+02 3.268e+02 4.131e+02 4.754e+02 9.036e+02, threshold=8.263e+02, percent-clipped=2.0 2022-12-07 07:43:08,787 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=13119.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 07:43:08,813 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3951, 1.2640, 3.2742, 1.6689, 3.5502, 3.2079, 2.2845, 3.6200], device='cuda:3'), covar=tensor([0.0290, 0.3140, 0.0405, 0.2510, 0.0345, 0.0432, 0.0997, 0.0239], device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0135, 0.0088, 0.0143, 0.0109, 0.0093, 0.0083, 0.0085], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 07:43:20,407 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.12 vs. limit=2.0 2022-12-07 07:43:31,967 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0277, 4.1534, 3.7656, 3.5785, 3.9191, 4.1369, 4.2261, 4.1497], device='cuda:3'), covar=tensor([0.1010, 0.0538, 0.1123, 0.1742, 0.0655, 0.0603, 0.0903, 0.1002], device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0137, 0.0178, 0.0203, 0.0144, 0.0163, 0.0175, 0.0137], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 07:43:35,168 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=13150.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:43:40,624 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 2022-12-07 07:43:44,127 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.02 vs. limit=2.0 2022-12-07 07:43:46,160 INFO [train.py:873] (3/4) Epoch 2, batch 5600, loss[loss=0.2134, simple_loss=0.2187, pruned_loss=0.104, over 14307.00 frames. ], tot_loss[loss=0.232, simple_loss=0.2269, pruned_loss=0.1185, over 1979785.45 frames. ], batch size: 39, lr: 2.92e-02, grad_scale: 8.0 2022-12-07 07:44:28,273 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.372e+02 3.284e+02 4.567e+02 6.071e+02 1.505e+03, threshold=9.133e+02, percent-clipped=13.0 2022-12-07 07:44:28,484 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1933, 1.8711, 1.8675, 1.7398, 1.3030, 1.8166, 1.6619, 1.0467], device='cuda:3'), covar=tensor([0.4104, 0.1199, 0.2111, 0.2898, 0.1266, 0.0526, 0.1396, 0.1989], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0061, 0.0050, 0.0052, 0.0064, 0.0047, 0.0060, 0.0076], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2022-12-07 07:44:37,824 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2022-12-07 07:44:44,766 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.55 vs. limit=5.0 2022-12-07 07:45:01,470 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2022-12-07 07:45:03,598 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7289, 2.6124, 2.5049, 2.8269, 2.4929, 2.1837, 2.7673, 2.8058], device='cuda:3'), covar=tensor([0.0786, 0.0728, 0.0846, 0.0757, 0.1035, 0.0930, 0.0833, 0.0845], device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0072, 0.0092, 0.0089, 0.0101, 0.0065, 0.0090, 0.0088], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2022-12-07 07:45:14,446 INFO [train.py:873] (3/4) Epoch 2, batch 5700, loss[loss=0.2517, simple_loss=0.2401, pruned_loss=0.1317, over 12745.00 frames. ], tot_loss[loss=0.2322, simple_loss=0.2269, pruned_loss=0.1188, over 2010400.14 frames. ], batch size: 100, lr: 2.91e-02, grad_scale: 8.0 2022-12-07 07:45:55,408 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.305e+02 2.840e+02 3.959e+02 4.964e+02 1.183e+03, threshold=7.919e+02, percent-clipped=6.0 2022-12-07 07:45:55,670 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9256, 2.4690, 2.3365, 1.8052, 2.2949, 2.3854, 2.4549, 1.9838], device='cuda:3'), covar=tensor([0.0257, 0.2524, 0.0540, 0.1081, 0.0651, 0.0317, 0.1481, 0.0692], device='cuda:3'), in_proj_covar=tensor([0.0059, 0.0154, 0.0077, 0.0102, 0.0070, 0.0066, 0.0060, 0.0068], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0003, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], device='cuda:3') 2022-12-07 07:46:23,512 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.46 vs. limit=5.0 2022-12-07 07:46:24,891 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.6494, 4.5139, 4.2166, 4.2420, 4.3443, 4.6704, 4.7859, 4.8131], device='cuda:3'), covar=tensor([0.0850, 0.0456, 0.0899, 0.1521, 0.0430, 0.0279, 0.0600, 0.0654], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0145, 0.0185, 0.0214, 0.0150, 0.0165, 0.0183, 0.0142], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 07:46:40,288 INFO [train.py:873] (3/4) Epoch 2, batch 5800, loss[loss=0.2405, simple_loss=0.2244, pruned_loss=0.1283, over 13987.00 frames. ], tot_loss[loss=0.2325, simple_loss=0.2272, pruned_loss=0.1189, over 2019510.14 frames. ], batch size: 19, lr: 2.90e-02, grad_scale: 8.0 2022-12-07 07:46:46,953 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.03 vs. limit=2.0 2022-12-07 07:47:20,398 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.58 vs. limit=2.0 2022-12-07 07:47:21,675 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.414e+02 3.070e+02 4.104e+02 5.303e+02 9.108e+02, threshold=8.208e+02, percent-clipped=3.0 2022-12-07 07:47:29,364 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=13419.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:48:07,861 INFO [train.py:873] (3/4) Epoch 2, batch 5900, loss[loss=0.2071, simple_loss=0.1931, pruned_loss=0.1105, over 5008.00 frames. ], tot_loss[loss=0.231, simple_loss=0.226, pruned_loss=0.118, over 1981458.14 frames. ], batch size: 100, lr: 2.89e-02, grad_scale: 8.0 2022-12-07 07:48:11,332 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=13467.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:48:49,279 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.385e+02 3.527e+02 4.364e+02 5.819e+02 1.292e+03, threshold=8.728e+02, percent-clipped=7.0 2022-12-07 07:48:50,633 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.95 vs. limit=2.0 2022-12-07 07:48:59,795 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.7310, 0.8156, 0.7498, 0.5657, 0.6028, 0.4499, 0.3071, 0.7462], device='cuda:3'), covar=tensor([0.0058, 0.0086, 0.0072, 0.0142, 0.0256, 0.0107, 0.0079, 0.0250], device='cuda:3'), in_proj_covar=tensor([0.0012, 0.0012, 0.0013, 0.0014, 0.0014, 0.0016, 0.0014, 0.0015], device='cuda:3'), out_proj_covar=tensor([2.7467e-05, 2.6757e-05, 3.1051e-05, 2.9173e-05, 2.9652e-05, 3.1891e-05, 3.5375e-05, 3.3147e-05], device='cuda:3') 2022-12-07 07:49:33,858 INFO [train.py:873] (3/4) Epoch 2, batch 6000, loss[loss=0.1893, simple_loss=0.204, pruned_loss=0.08729, over 13944.00 frames. ], tot_loss[loss=0.2311, simple_loss=0.2255, pruned_loss=0.1184, over 1943367.20 frames. ], batch size: 20, lr: 2.88e-02, grad_scale: 8.0 2022-12-07 07:49:33,858 INFO [train.py:896] (3/4) Computing validation loss 2022-12-07 07:49:40,446 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3075, 3.2183, 3.3570, 3.1092, 3.2357, 3.3442, 1.6471, 3.2349], device='cuda:3'), covar=tensor([0.0209, 0.0324, 0.0340, 0.0246, 0.0365, 0.0232, 0.3006, 0.0252], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0102, 0.0096, 0.0077, 0.0130, 0.0087, 0.0137, 0.0120], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003], device='cuda:3') 2022-12-07 07:49:42,178 INFO [train.py:905] (3/4) Epoch 2, validation: loss=0.139, simple_loss=0.1792, pruned_loss=0.04941, over 857387.00 frames. 2022-12-07 07:49:42,179 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17690MB 2022-12-07 07:49:49,220 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3617, 1.8647, 2.8487, 2.3129, 2.7044, 2.2268, 2.3963, 2.1892], device='cuda:3'), covar=tensor([0.0118, 0.0796, 0.0058, 0.0349, 0.0095, 0.0172, 0.0133, 0.0509], device='cuda:3'), in_proj_covar=tensor([0.0182, 0.0315, 0.0130, 0.0240, 0.0171, 0.0194, 0.0186, 0.0348], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:3') 2022-12-07 07:50:03,717 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4274, 1.4955, 0.8799, 1.0982, 0.8197, 1.0267, 1.4525, 1.0408], device='cuda:3'), covar=tensor([0.0656, 0.0726, 0.0727, 0.0378, 0.1048, 0.0291, 0.0213, 0.0421], device='cuda:3'), in_proj_covar=tensor([0.0012, 0.0013, 0.0014, 0.0015, 0.0014, 0.0017, 0.0013, 0.0016], device='cuda:3'), out_proj_covar=tensor([2.7916e-05, 2.7373e-05, 3.2438e-05, 2.9869e-05, 3.1411e-05, 3.2901e-05, 3.4805e-05, 3.3513e-05], device='cuda:3') 2022-12-07 07:50:09,651 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5598, 2.3931, 2.2199, 2.1518, 2.5151, 2.3308, 2.4547, 2.4187], device='cuda:3'), covar=tensor([0.0462, 0.1026, 0.1332, 0.1975, 0.0448, 0.0601, 0.0968, 0.0879], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0145, 0.0182, 0.0211, 0.0151, 0.0168, 0.0175, 0.0147], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 07:50:20,362 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0928, 2.1011, 1.7316, 2.3052, 2.1320, 2.3884, 1.9312, 1.7945], device='cuda:3'), covar=tensor([0.0199, 0.0378, 0.1407, 0.0115, 0.0227, 0.0177, 0.0458, 0.1101], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0225, 0.0328, 0.0150, 0.0162, 0.0158, 0.0211, 0.0335], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:3') 2022-12-07 07:50:24,253 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.694e+02 3.403e+02 4.162e+02 5.697e+02 1.072e+03, threshold=8.324e+02, percent-clipped=3.0 2022-12-07 07:50:40,331 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.09 vs. limit=2.0 2022-12-07 07:51:06,872 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8087, 2.6879, 2.0288, 2.9646, 2.6017, 2.9972, 2.4744, 2.0765], device='cuda:3'), covar=tensor([0.0172, 0.0300, 0.1274, 0.0115, 0.0169, 0.0155, 0.0389, 0.1470], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0226, 0.0327, 0.0153, 0.0163, 0.0158, 0.0214, 0.0335], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:3') 2022-12-07 07:51:09,282 INFO [train.py:873] (3/4) Epoch 2, batch 6100, loss[loss=0.2156, simple_loss=0.2184, pruned_loss=0.1064, over 14659.00 frames. ], tot_loss[loss=0.2304, simple_loss=0.225, pruned_loss=0.1179, over 1883511.42 frames. ], batch size: 33, lr: 2.87e-02, grad_scale: 8.0 2022-12-07 07:51:50,708 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.181e+02 3.186e+02 4.052e+02 4.953e+02 1.165e+03, threshold=8.105e+02, percent-clipped=5.0 2022-12-07 07:52:13,569 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=13737.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:52:17,799 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9026, 4.7207, 4.3813, 4.4460, 4.4980, 4.6050, 4.8447, 4.8556], device='cuda:3'), covar=tensor([0.0524, 0.0482, 0.1185, 0.1380, 0.0462, 0.0419, 0.0840, 0.0754], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0144, 0.0183, 0.0211, 0.0157, 0.0168, 0.0177, 0.0150], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 07:52:35,525 INFO [train.py:873] (3/4) Epoch 2, batch 6200, loss[loss=0.1788, simple_loss=0.1584, pruned_loss=0.09961, over 1228.00 frames. ], tot_loss[loss=0.2301, simple_loss=0.2245, pruned_loss=0.1178, over 1881949.89 frames. ], batch size: 100, lr: 2.86e-02, grad_scale: 8.0 2022-12-07 07:53:05,585 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=13798.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 07:53:17,360 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.626e+02 2.778e+02 3.775e+02 5.188e+02 1.088e+03, threshold=7.550e+02, percent-clipped=5.0 2022-12-07 07:53:54,633 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8708, 0.7833, 0.4001, 0.8954, 0.9786, 0.3531, 0.5578, 0.7976], device='cuda:3'), covar=tensor([0.0192, 0.0124, 0.0049, 0.0282, 0.0092, 0.0086, 0.0385, 0.0093], device='cuda:3'), in_proj_covar=tensor([0.0017, 0.0019, 0.0020, 0.0016, 0.0018, 0.0021, 0.0016, 0.0017], device='cuda:3'), out_proj_covar=tensor([4.5896e-05, 4.6902e-05, 4.3740e-05, 4.1772e-05, 4.3513e-05, 4.7595e-05, 4.1746e-05, 4.1040e-05], device='cuda:3') 2022-12-07 07:54:01,843 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.95 vs. limit=2.0 2022-12-07 07:54:02,989 INFO [train.py:873] (3/4) Epoch 2, batch 6300, loss[loss=0.2341, simple_loss=0.2297, pruned_loss=0.1193, over 13524.00 frames. ], tot_loss[loss=0.2282, simple_loss=0.224, pruned_loss=0.1163, over 1918758.46 frames. ], batch size: 100, lr: 2.86e-02, grad_scale: 8.0 2022-12-07 07:54:32,293 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.0224, 1.0059, 1.1094, 1.1208, 0.9405, 1.1603, 0.7998, 0.8789], device='cuda:3'), covar=tensor([0.5236, 0.2170, 0.0921, 0.1068, 0.1177, 0.0763, 0.1969, 0.2085], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0065, 0.0049, 0.0053, 0.0063, 0.0050, 0.0066, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2022-12-07 07:54:43,892 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.964e+01 3.256e+02 4.500e+02 5.561e+02 1.186e+03, threshold=8.999e+02, percent-clipped=4.0 2022-12-07 07:54:58,951 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2022-12-07 07:55:29,726 INFO [train.py:873] (3/4) Epoch 2, batch 6400, loss[loss=0.2121, simple_loss=0.1789, pruned_loss=0.1227, over 1330.00 frames. ], tot_loss[loss=0.2297, simple_loss=0.2249, pruned_loss=0.1172, over 1890778.22 frames. ], batch size: 100, lr: 2.85e-02, grad_scale: 8.0 2022-12-07 07:55:45,392 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0969, 3.7204, 4.1723, 3.6002, 3.9138, 4.1530, 1.7100, 3.9222], device='cuda:3'), covar=tensor([0.0142, 0.0295, 0.0338, 0.0299, 0.0283, 0.0177, 0.2626, 0.0194], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0103, 0.0099, 0.0078, 0.0134, 0.0088, 0.0138, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003], device='cuda:3') 2022-12-07 07:56:11,802 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.544e+02 3.193e+02 4.162e+02 5.343e+02 1.659e+03, threshold=8.325e+02, percent-clipped=5.0 2022-12-07 07:56:41,436 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.95 vs. limit=2.0 2022-12-07 07:56:49,223 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.19 vs. limit=2.0 2022-12-07 07:56:57,035 INFO [train.py:873] (3/4) Epoch 2, batch 6500, loss[loss=0.2122, simple_loss=0.2152, pruned_loss=0.1046, over 14234.00 frames. ], tot_loss[loss=0.2315, simple_loss=0.2258, pruned_loss=0.1186, over 1856819.71 frames. ], batch size: 35, lr: 2.84e-02, grad_scale: 8.0 2022-12-07 07:57:04,660 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14072.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:57:17,824 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.31 vs. limit=2.0 2022-12-07 07:57:19,167 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1745, 0.8623, 0.2290, 1.3481, 1.2353, 1.3345, 0.9352, 1.1804], device='cuda:3'), covar=tensor([0.1051, 0.1510, 0.2067, 0.0811, 0.1069, 0.0191, 0.0268, 0.1011], device='cuda:3'), in_proj_covar=tensor([0.0011, 0.0012, 0.0013, 0.0014, 0.0013, 0.0015, 0.0013, 0.0015], device='cuda:3'), out_proj_covar=tensor([2.5691e-05, 2.7361e-05, 3.1903e-05, 2.9136e-05, 2.9279e-05, 3.1838e-05, 3.4094e-05, 3.2787e-05], device='cuda:3') 2022-12-07 07:57:23,634 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14093.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 07:57:31,917 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14103.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:57:38,491 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.039e+02 3.036e+02 3.911e+02 5.054e+02 1.167e+03, threshold=7.822e+02, percent-clipped=3.0 2022-12-07 07:57:46,655 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14120.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 07:57:58,307 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14133.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:58:12,887 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.02 vs. limit=2.0 2022-12-07 07:58:23,892 INFO [train.py:873] (3/4) Epoch 2, batch 6600, loss[loss=0.2445, simple_loss=0.2373, pruned_loss=0.1259, over 14291.00 frames. ], tot_loss[loss=0.2308, simple_loss=0.2257, pruned_loss=0.1179, over 1919003.37 frames. ], batch size: 76, lr: 2.83e-02, grad_scale: 8.0 2022-12-07 07:58:24,886 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14164.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 07:58:39,633 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14181.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 07:58:45,334 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8437, 3.3865, 2.2445, 3.9148, 3.3587, 3.8549, 3.4874, 2.3281], device='cuda:3'), covar=tensor([0.0125, 0.0404, 0.2621, 0.0155, 0.0155, 0.0309, 0.0417, 0.2307], device='cuda:3'), in_proj_covar=tensor([0.0185, 0.0241, 0.0346, 0.0159, 0.0171, 0.0166, 0.0224, 0.0355], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0003], device='cuda:3') 2022-12-07 07:59:01,256 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.46 vs. limit=5.0 2022-12-07 07:59:06,278 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.319e+02 2.900e+02 3.866e+02 4.857e+02 9.306e+02, threshold=7.732e+02, percent-clipped=4.0 2022-12-07 07:59:30,826 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.85 vs. limit=2.0 2022-12-07 07:59:34,808 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2022-12-07 07:59:51,416 INFO [train.py:873] (3/4) Epoch 2, batch 6700, loss[loss=0.2097, simple_loss=0.2183, pruned_loss=0.1005, over 14275.00 frames. ], tot_loss[loss=0.2286, simple_loss=0.2244, pruned_loss=0.1164, over 1955972.75 frames. ], batch size: 44, lr: 2.82e-02, grad_scale: 8.0 2022-12-07 08:00:32,373 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.313e+02 3.050e+02 4.021e+02 5.480e+02 1.335e+03, threshold=8.042e+02, percent-clipped=5.0 2022-12-07 08:00:47,681 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2022-12-07 08:00:55,967 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7907, 1.6091, 2.1336, 1.7859, 2.0857, 1.6910, 1.8802, 1.8603], device='cuda:3'), covar=tensor([0.0065, 0.0259, 0.0026, 0.0063, 0.0031, 0.0092, 0.0049, 0.0103], device='cuda:3'), in_proj_covar=tensor([0.0189, 0.0323, 0.0138, 0.0258, 0.0187, 0.0200, 0.0210, 0.0354], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:3') 2022-12-07 08:01:00,073 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3677, 1.9710, 2.6043, 2.2277, 2.6398, 2.2982, 2.3617, 2.1670], device='cuda:3'), covar=tensor([0.0090, 0.0678, 0.0054, 0.0256, 0.0087, 0.0163, 0.0127, 0.0483], device='cuda:3'), in_proj_covar=tensor([0.0189, 0.0321, 0.0137, 0.0256, 0.0186, 0.0199, 0.0208, 0.0353], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:3') 2022-12-07 08:01:18,008 INFO [train.py:873] (3/4) Epoch 2, batch 6800, loss[loss=0.2087, simple_loss=0.1752, pruned_loss=0.1211, over 1200.00 frames. ], tot_loss[loss=0.2293, simple_loss=0.2244, pruned_loss=0.1171, over 1945097.14 frames. ], batch size: 100, lr: 2.81e-02, grad_scale: 8.0 2022-12-07 08:01:18,180 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6652, 1.4904, 1.9356, 2.1254, 1.3732, 1.6697, 1.8975, 1.8476], device='cuda:3'), covar=tensor([0.0024, 0.0040, 0.0027, 0.0012, 0.0041, 0.0050, 0.0014, 0.0022], device='cuda:3'), in_proj_covar=tensor([0.0109, 0.0140, 0.0146, 0.0134, 0.0125, 0.0181, 0.0086, 0.0172], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0001, 0.0002], device='cuda:3') 2022-12-07 08:01:25,774 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 2022-12-07 08:01:39,177 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.88 vs. limit=2.0 2022-12-07 08:01:43,684 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14393.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:01:55,302 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9474, 4.4749, 4.5066, 5.0891, 4.8123, 4.0205, 5.1544, 5.0156], device='cuda:3'), covar=tensor([0.0603, 0.0601, 0.0551, 0.0494, 0.0544, 0.0431, 0.0512, 0.0607], device='cuda:3'), in_proj_covar=tensor([0.0094, 0.0076, 0.0094, 0.0090, 0.0100, 0.0067, 0.0094, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2022-12-07 08:01:59,490 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.256e+02 3.284e+02 4.234e+02 5.689e+02 1.698e+03, threshold=8.469e+02, percent-clipped=9.0 2022-12-07 08:02:01,474 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14413.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:02:14,270 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14428.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:02:25,491 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14441.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:02:41,233 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14459.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:02:44,541 INFO [train.py:873] (3/4) Epoch 2, batch 6900, loss[loss=0.2487, simple_loss=0.2285, pruned_loss=0.1345, over 6953.00 frames. ], tot_loss[loss=0.2307, simple_loss=0.2255, pruned_loss=0.1179, over 1990110.57 frames. ], batch size: 100, lr: 2.80e-02, grad_scale: 8.0 2022-12-07 08:02:53,943 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14474.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:02:55,529 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14476.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 08:03:05,488 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14487.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:03:25,507 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.414e+02 3.007e+02 4.167e+02 5.148e+02 1.275e+03, threshold=8.334e+02, percent-clipped=3.0 2022-12-07 08:03:57,821 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14548.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:04:04,651 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6245, 1.3393, 3.3193, 1.4963, 3.5440, 3.5101, 2.4954, 3.8317], device='cuda:3'), covar=tensor([0.0162, 0.2623, 0.0369, 0.2224, 0.0257, 0.0281, 0.0619, 0.0142], device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0139, 0.0094, 0.0150, 0.0112, 0.0099, 0.0089, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 08:04:10,427 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2813, 3.0971, 3.3893, 3.1449, 3.1640, 2.8117, 1.4229, 3.1319], device='cuda:3'), covar=tensor([0.0197, 0.0273, 0.0325, 0.0242, 0.0254, 0.0500, 0.2551, 0.0198], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0102, 0.0098, 0.0079, 0.0132, 0.0089, 0.0139, 0.0121], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003], device='cuda:3') 2022-12-07 08:04:11,094 INFO [train.py:873] (3/4) Epoch 2, batch 7000, loss[loss=0.2603, simple_loss=0.246, pruned_loss=0.1373, over 9497.00 frames. ], tot_loss[loss=0.2296, simple_loss=0.2246, pruned_loss=0.1173, over 1965367.70 frames. ], batch size: 100, lr: 2.79e-02, grad_scale: 8.0 2022-12-07 08:04:41,753 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=19.42 vs. limit=5.0 2022-12-07 08:04:50,596 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2022-12-07 08:04:53,422 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.467e+02 3.149e+02 3.940e+02 5.259e+02 1.451e+03, threshold=7.881e+02, percent-clipped=6.0 2022-12-07 08:05:04,499 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.89 vs. limit=5.0 2022-12-07 08:05:37,945 INFO [train.py:873] (3/4) Epoch 2, batch 7100, loss[loss=0.1936, simple_loss=0.208, pruned_loss=0.08956, over 14137.00 frames. ], tot_loss[loss=0.2282, simple_loss=0.2237, pruned_loss=0.1163, over 2005435.45 frames. ], batch size: 84, lr: 2.79e-02, grad_scale: 16.0 2022-12-07 08:06:07,912 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.11 vs. limit=2.0 2022-12-07 08:06:19,126 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.591e+01 3.151e+02 3.961e+02 5.034e+02 1.065e+03, threshold=7.923e+02, percent-clipped=3.0 2022-12-07 08:06:34,584 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14728.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:06:44,682 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14740.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:06:53,179 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.6155, 1.6620, 2.8443, 2.1424, 2.9382, 1.6070, 2.1597, 2.7315], device='cuda:3'), covar=tensor([0.0317, 0.3179, 0.0165, 0.4091, 0.0169, 0.2496, 0.0729, 0.0267], device='cuda:3'), in_proj_covar=tensor([0.0212, 0.0322, 0.0164, 0.0429, 0.0154, 0.0327, 0.0270, 0.0167], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0004, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-07 08:07:00,760 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14759.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:07:04,457 INFO [train.py:873] (3/4) Epoch 2, batch 7200, loss[loss=0.1931, simple_loss=0.2108, pruned_loss=0.0877, over 14274.00 frames. ], tot_loss[loss=0.2292, simple_loss=0.224, pruned_loss=0.1172, over 1965658.37 frames. ], batch size: 18, lr: 2.78e-02, grad_scale: 16.0 2022-12-07 08:07:09,550 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14769.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:07:13,134 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.2110, 5.1855, 4.8596, 5.4286, 5.1616, 4.3903, 5.4644, 5.3462], device='cuda:3'), covar=tensor([0.0602, 0.0450, 0.0549, 0.0576, 0.0509, 0.0316, 0.0467, 0.0627], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0075, 0.0092, 0.0087, 0.0098, 0.0065, 0.0090, 0.0085], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2022-12-07 08:07:15,476 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14776.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:07:15,580 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14776.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 08:07:24,009 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5267, 1.6814, 2.7809, 1.9982, 2.8596, 1.6466, 2.1739, 2.7360], device='cuda:3'), covar=tensor([0.0306, 0.3060, 0.0181, 0.4167, 0.0169, 0.2364, 0.0698, 0.0217], device='cuda:3'), in_proj_covar=tensor([0.0209, 0.0315, 0.0163, 0.0423, 0.0151, 0.0321, 0.0266, 0.0163], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0004, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-07 08:07:37,093 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14801.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:07:42,104 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14807.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:07:45,741 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.424e+02 2.988e+02 3.907e+02 5.109e+02 1.521e+03, threshold=7.813e+02, percent-clipped=5.0 2022-12-07 08:07:56,618 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14824.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 08:08:12,981 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14843.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:08:29,763 INFO [train.py:873] (3/4) Epoch 2, batch 7300, loss[loss=0.2437, simple_loss=0.2329, pruned_loss=0.1272, over 14296.00 frames. ], tot_loss[loss=0.2276, simple_loss=0.2231, pruned_loss=0.1161, over 2014683.19 frames. ], batch size: 69, lr: 2.77e-02, grad_scale: 16.0 2022-12-07 08:09:10,700 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.284e+02 2.806e+02 3.909e+02 5.179e+02 1.856e+03, threshold=7.818e+02, percent-clipped=6.0 2022-12-07 08:09:55,278 INFO [train.py:873] (3/4) Epoch 2, batch 7400, loss[loss=0.2412, simple_loss=0.2131, pruned_loss=0.1346, over 3830.00 frames. ], tot_loss[loss=0.227, simple_loss=0.2227, pruned_loss=0.1156, over 1966416.93 frames. ], batch size: 100, lr: 2.76e-02, grad_scale: 16.0 2022-12-07 08:10:25,544 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2022-12-07 08:10:39,742 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.247e+01 3.005e+02 3.840e+02 4.787e+02 1.067e+03, threshold=7.681e+02, percent-clipped=4.0 2022-12-07 08:11:24,371 INFO [train.py:873] (3/4) Epoch 2, batch 7500, loss[loss=0.2178, simple_loss=0.2218, pruned_loss=0.1069, over 14385.00 frames. ], tot_loss[loss=0.2277, simple_loss=0.2234, pruned_loss=0.116, over 1991816.70 frames. ], batch size: 53, lr: 2.75e-02, grad_scale: 8.0 2022-12-07 08:11:29,634 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15069.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:11:41,747 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7339, 1.5892, 1.7651, 1.2445, 1.3002, 1.1994, 1.9032, 1.4126], device='cuda:3'), covar=tensor([0.0339, 0.2317, 0.0783, 0.1712, 0.1101, 0.0629, 0.0559, 0.1134], device='cuda:3'), in_proj_covar=tensor([0.0065, 0.0177, 0.0087, 0.0115, 0.0071, 0.0073, 0.0066, 0.0088], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 08:11:46,875 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9839, 1.8562, 1.6773, 2.0307, 2.0203, 2.0811, 1.8449, 1.6651], device='cuda:3'), covar=tensor([0.0092, 0.0233, 0.0291, 0.0078, 0.0148, 0.0061, 0.0242, 0.0245], device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0252, 0.0343, 0.0157, 0.0179, 0.0179, 0.0233, 0.0355], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0003, 0.0001, 0.0001, 0.0001, 0.0002, 0.0003], device='cuda:3') 2022-12-07 08:11:52,695 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=15096.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:11:53,474 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.2351, 3.9292, 3.7883, 3.7402, 4.0100, 4.0934, 4.2554, 4.1481], device='cuda:3'), covar=tensor([0.0829, 0.0580, 0.1295, 0.1910, 0.0582, 0.0471, 0.0823, 0.0872], device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0153, 0.0203, 0.0242, 0.0166, 0.0186, 0.0192, 0.0168], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 08:11:57,859 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.65 vs. limit=2.0 2022-12-07 08:12:03,099 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.166e+01 2.953e+02 4.080e+02 5.440e+02 1.285e+03, threshold=8.159e+02, percent-clipped=6.0 2022-12-07 08:12:05,497 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=15117.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:12:48,570 INFO [train.py:873] (3/4) Epoch 3, batch 0, loss[loss=0.3464, simple_loss=0.301, pruned_loss=0.1959, over 14266.00 frames. ], tot_loss[loss=0.3464, simple_loss=0.301, pruned_loss=0.1959, over 14266.00 frames. ], batch size: 63, lr: 2.61e-02, grad_scale: 8.0 2022-12-07 08:12:48,571 INFO [train.py:896] (3/4) Computing validation loss 2022-12-07 08:12:55,593 INFO [train.py:905] (3/4) Epoch 3, validation: loss=0.159, simple_loss=0.1997, pruned_loss=0.05911, over 857387.00 frames. 2022-12-07 08:12:55,594 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17841MB 2022-12-07 08:12:59,155 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.3293, 4.8039, 4.8578, 5.3101, 5.1847, 4.3918, 5.2931, 4.5162], device='cuda:3'), covar=tensor([0.0253, 0.0829, 0.0201, 0.0266, 0.0389, 0.0354, 0.0362, 0.0384], device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0156, 0.0101, 0.0093, 0.0097, 0.0093, 0.0142, 0.0116], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 08:13:12,554 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15143.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:13:51,907 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.3101, 1.3554, 1.3398, 1.3998, 1.1141, 1.4294, 1.0688, 0.8690], device='cuda:3'), covar=tensor([0.2950, 0.1716, 0.0651, 0.0607, 0.1095, 0.0397, 0.1505, 0.2474], device='cuda:3'), in_proj_covar=tensor([0.0157, 0.0066, 0.0052, 0.0058, 0.0066, 0.0051, 0.0065, 0.0081], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 08:13:53,455 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=15191.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:13:57,064 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=15195.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:14:04,975 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6504, 1.3559, 2.0418, 1.1087, 1.4114, 1.7106, 1.0881, 1.7443], device='cuda:3'), covar=tensor([0.0706, 0.2136, 0.0526, 0.2399, 0.1633, 0.0881, 0.3428, 0.0754], device='cuda:3'), in_proj_covar=tensor([0.0060, 0.0063, 0.0056, 0.0069, 0.0070, 0.0057, 0.0118, 0.0066], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001], device='cuda:3') 2022-12-07 08:14:11,335 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.526e+01 3.048e+02 3.683e+02 5.299e+02 1.167e+03, threshold=7.365e+02, percent-clipped=4.0 2022-12-07 08:14:23,258 INFO [train.py:873] (3/4) Epoch 3, batch 100, loss[loss=0.2588, simple_loss=0.2419, pruned_loss=0.1379, over 9469.00 frames. ], tot_loss[loss=0.2267, simple_loss=0.2245, pruned_loss=0.1145, over 839110.59 frames. ], batch size: 100, lr: 2.60e-02, grad_scale: 8.0 2022-12-07 08:14:49,209 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=15256.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 08:15:09,224 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=3.27 vs. limit=2.0 2022-12-07 08:15:38,765 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.471e+02 3.082e+02 4.015e+02 5.606e+02 9.270e+02, threshold=8.031e+02, percent-clipped=6.0 2022-12-07 08:15:49,797 INFO [train.py:873] (3/4) Epoch 3, batch 200, loss[loss=0.2385, simple_loss=0.221, pruned_loss=0.128, over 7811.00 frames. ], tot_loss[loss=0.2263, simple_loss=0.223, pruned_loss=0.1148, over 1311686.15 frames. ], batch size: 100, lr: 2.59e-02, grad_scale: 8.0 2022-12-07 08:15:53,952 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.30 vs. limit=5.0 2022-12-07 08:16:32,733 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3506, 1.7398, 1.5033, 1.6909, 1.1953, 1.7086, 1.6838, 0.9722], device='cuda:3'), covar=tensor([0.3831, 0.1446, 0.3285, 0.1706, 0.1700, 0.0986, 0.1311, 0.2817], device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0062, 0.0053, 0.0058, 0.0065, 0.0051, 0.0064, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 08:16:49,281 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3912, 2.7253, 3.3082, 2.1504, 2.6025, 2.7548, 1.2469, 2.5306], device='cuda:3'), covar=tensor([0.1145, 0.0495, 0.0432, 0.1656, 0.0728, 0.0724, 0.3039, 0.0921], device='cuda:3'), in_proj_covar=tensor([0.0062, 0.0063, 0.0056, 0.0068, 0.0072, 0.0057, 0.0120, 0.0069], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001], device='cuda:3') 2022-12-07 08:16:50,919 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15396.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:16:56,234 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8748, 2.0787, 3.0485, 3.1177, 2.6615, 1.9459, 2.8052, 2.3432], device='cuda:3'), covar=tensor([0.0048, 0.0124, 0.0099, 0.0076, 0.0060, 0.0261, 0.0029, 0.0182], device='cuda:3'), in_proj_covar=tensor([0.0114, 0.0144, 0.0156, 0.0145, 0.0125, 0.0189, 0.0087, 0.0178], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0001, 0.0002], device='cuda:3') 2022-12-07 08:17:04,617 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.006e+02 3.006e+02 3.776e+02 5.007e+02 1.251e+03, threshold=7.551e+02, percent-clipped=5.0 2022-12-07 08:17:16,243 INFO [train.py:873] (3/4) Epoch 3, batch 300, loss[loss=0.2263, simple_loss=0.2208, pruned_loss=0.1159, over 14244.00 frames. ], tot_loss[loss=0.2246, simple_loss=0.2215, pruned_loss=0.1138, over 1521654.92 frames. ], batch size: 80, lr: 2.59e-02, grad_scale: 8.0 2022-12-07 08:17:32,204 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=15444.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:18:23,159 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.06 vs. limit=2.0 2022-12-07 08:18:31,036 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.142e+02 2.693e+02 3.512e+02 4.941e+02 9.484e+02, threshold=7.024e+02, percent-clipped=2.0 2022-12-07 08:18:41,946 INFO [train.py:873] (3/4) Epoch 3, batch 400, loss[loss=0.2068, simple_loss=0.1991, pruned_loss=0.1072, over 3876.00 frames. ], tot_loss[loss=0.223, simple_loss=0.2206, pruned_loss=0.1127, over 1689055.67 frames. ], batch size: 100, lr: 2.58e-02, grad_scale: 8.0 2022-12-07 08:19:04,015 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=15551.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 08:19:56,021 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.354e+02 3.036e+02 3.993e+02 5.394e+02 1.334e+03, threshold=7.985e+02, percent-clipped=7.0 2022-12-07 08:20:07,945 INFO [train.py:873] (3/4) Epoch 3, batch 500, loss[loss=0.2395, simple_loss=0.2275, pruned_loss=0.1258, over 11972.00 frames. ], tot_loss[loss=0.2221, simple_loss=0.2201, pruned_loss=0.112, over 1774855.59 frames. ], batch size: 100, lr: 2.57e-02, grad_scale: 8.0 2022-12-07 08:20:13,779 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.02 vs. limit=2.0 2022-12-07 08:20:20,110 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7813, 2.6506, 3.9010, 2.7620, 3.6370, 3.7025, 3.4452, 3.0519], device='cuda:3'), covar=tensor([0.0093, 0.0896, 0.0064, 0.0589, 0.0221, 0.0145, 0.0654, 0.0721], device='cuda:3'), in_proj_covar=tensor([0.0206, 0.0324, 0.0153, 0.0266, 0.0215, 0.0216, 0.0232, 0.0363], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:3') 2022-12-07 08:20:48,200 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8466, 1.2809, 3.1016, 2.9366, 3.0513, 2.9030, 2.2424, 3.1776], device='cuda:3'), covar=tensor([0.0997, 0.1321, 0.0075, 0.0173, 0.0141, 0.0096, 0.0272, 0.0083], device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0135, 0.0073, 0.0102, 0.0083, 0.0089, 0.0067, 0.0071], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 08:20:49,985 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1701, 1.8516, 2.6349, 2.1141, 2.5086, 2.2047, 2.1430, 2.1568], device='cuda:3'), covar=tensor([0.0104, 0.0550, 0.0044, 0.0220, 0.0085, 0.0136, 0.0095, 0.0361], device='cuda:3'), in_proj_covar=tensor([0.0207, 0.0329, 0.0155, 0.0265, 0.0217, 0.0219, 0.0229, 0.0366], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:3') 2022-12-07 08:21:11,339 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2022-12-07 08:21:22,225 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.517e+02 2.935e+02 3.905e+02 4.916e+02 1.140e+03, threshold=7.811e+02, percent-clipped=3.0 2022-12-07 08:21:32,702 INFO [train.py:873] (3/4) Epoch 3, batch 600, loss[loss=0.2609, simple_loss=0.2374, pruned_loss=0.1422, over 8654.00 frames. ], tot_loss[loss=0.2216, simple_loss=0.2196, pruned_loss=0.1117, over 1819963.34 frames. ], batch size: 100, lr: 2.56e-02, grad_scale: 8.0 2022-12-07 08:22:18,471 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.84 vs. limit=2.0 2022-12-07 08:22:20,894 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.07 vs. limit=2.0 2022-12-07 08:22:47,078 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.484e+02 3.132e+02 3.850e+02 5.057e+02 1.369e+03, threshold=7.700e+02, percent-clipped=5.0 2022-12-07 08:22:47,343 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1989, 1.9524, 4.0640, 2.8859, 3.9206, 1.8009, 3.2923, 4.0974], device='cuda:3'), covar=tensor([0.0222, 0.4158, 0.0299, 0.6171, 0.0154, 0.3558, 0.0653, 0.0113], device='cuda:3'), in_proj_covar=tensor([0.0215, 0.0326, 0.0173, 0.0425, 0.0157, 0.0333, 0.0281, 0.0170], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0002, 0.0004, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-07 08:22:48,360 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2022-12-07 08:22:58,911 INFO [train.py:873] (3/4) Epoch 3, batch 700, loss[loss=0.1763, simple_loss=0.1932, pruned_loss=0.07968, over 13931.00 frames. ], tot_loss[loss=0.221, simple_loss=0.2192, pruned_loss=0.1114, over 1890785.79 frames. ], batch size: 20, lr: 2.56e-02, grad_scale: 8.0 2022-12-07 08:23:06,747 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6990, 1.8200, 3.7956, 2.6462, 3.5125, 1.7220, 3.0275, 3.7067], device='cuda:3'), covar=tensor([0.0243, 0.3867, 0.0250, 0.5245, 0.0170, 0.3025, 0.0632, 0.0132], device='cuda:3'), in_proj_covar=tensor([0.0216, 0.0322, 0.0174, 0.0424, 0.0156, 0.0331, 0.0280, 0.0169], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0002, 0.0004, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-07 08:23:20,521 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15851.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:23:48,049 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=15883.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:23:55,465 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9966, 3.7392, 4.2000, 3.5417, 3.8377, 4.1664, 1.4525, 3.8350], device='cuda:3'), covar=tensor([0.0196, 0.0289, 0.0310, 0.0420, 0.0295, 0.0141, 0.2911, 0.0223], device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0107, 0.0099, 0.0084, 0.0138, 0.0094, 0.0142, 0.0128], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003], device='cuda:3') 2022-12-07 08:24:01,544 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=15899.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:24:12,921 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.224e+02 3.132e+02 4.108e+02 5.472e+02 1.235e+03, threshold=8.216e+02, percent-clipped=5.0 2022-12-07 08:24:23,971 INFO [train.py:873] (3/4) Epoch 3, batch 800, loss[loss=0.1772, simple_loss=0.1931, pruned_loss=0.08067, over 14538.00 frames. ], tot_loss[loss=0.2218, simple_loss=0.2199, pruned_loss=0.1118, over 1955902.43 frames. ], batch size: 34, lr: 2.55e-02, grad_scale: 8.0 2022-12-07 08:24:40,865 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=15944.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:25:02,530 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9690, 2.5076, 2.9013, 1.7255, 2.1837, 2.2323, 1.0852, 2.2357], device='cuda:3'), covar=tensor([0.1657, 0.0823, 0.0543, 0.1888, 0.1028, 0.1003, 0.3963, 0.1253], device='cuda:3'), in_proj_covar=tensor([0.0065, 0.0068, 0.0063, 0.0072, 0.0079, 0.0062, 0.0127, 0.0073], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001], device='cuda:3') 2022-12-07 08:25:19,034 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.9124, 5.2833, 5.3141, 5.8379, 5.5684, 4.7128, 5.8306, 4.7475], device='cuda:3'), covar=tensor([0.0184, 0.0808, 0.0198, 0.0369, 0.0592, 0.0275, 0.0460, 0.0464], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0164, 0.0101, 0.0099, 0.0097, 0.0092, 0.0149, 0.0116], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 08:25:30,318 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16001.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:25:39,380 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.514e+02 2.828e+02 4.017e+02 4.886e+02 9.739e+02, threshold=8.033e+02, percent-clipped=2.0 2022-12-07 08:25:51,106 INFO [train.py:873] (3/4) Epoch 3, batch 900, loss[loss=0.1882, simple_loss=0.1692, pruned_loss=0.1036, over 1310.00 frames. ], tot_loss[loss=0.2211, simple_loss=0.2195, pruned_loss=0.1114, over 1954359.51 frames. ], batch size: 100, lr: 2.54e-02, grad_scale: 8.0 2022-12-07 08:26:22,578 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16062.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:26:42,939 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16085.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:27:06,256 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.269e+02 2.979e+02 3.878e+02 4.942e+02 1.281e+03, threshold=7.755e+02, percent-clipped=6.0 2022-12-07 08:27:16,835 INFO [train.py:873] (3/4) Epoch 3, batch 1000, loss[loss=0.2527, simple_loss=0.2331, pruned_loss=0.1361, over 10381.00 frames. ], tot_loss[loss=0.22, simple_loss=0.2186, pruned_loss=0.1106, over 1958073.86 frames. ], batch size: 100, lr: 2.54e-02, grad_scale: 8.0 2022-12-07 08:27:34,519 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7962, 1.6492, 2.1303, 1.7682, 2.1703, 1.7096, 1.8676, 1.9370], device='cuda:3'), covar=tensor([0.0076, 0.0262, 0.0026, 0.0074, 0.0043, 0.0078, 0.0048, 0.0085], device='cuda:3'), in_proj_covar=tensor([0.0201, 0.0332, 0.0155, 0.0262, 0.0210, 0.0219, 0.0227, 0.0350], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:3') 2022-12-07 08:27:35,314 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16146.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:27:37,949 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16149.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:28:29,944 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16210.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:28:31,352 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.271e+02 3.033e+02 4.056e+02 5.147e+02 1.062e+03, threshold=8.113e+02, percent-clipped=5.0 2022-12-07 08:28:42,881 INFO [train.py:873] (3/4) Epoch 3, batch 1100, loss[loss=0.2324, simple_loss=0.2318, pruned_loss=0.1165, over 14382.00 frames. ], tot_loss[loss=0.2189, simple_loss=0.2179, pruned_loss=0.11, over 1978629.54 frames. ], batch size: 73, lr: 2.53e-02, grad_scale: 8.0 2022-12-07 08:28:55,085 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16239.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:29:06,864 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.2017, 1.4691, 1.3221, 1.4338, 1.0687, 1.1368, 1.2556, 0.9691], device='cuda:3'), covar=tensor([0.2665, 0.1564, 0.0921, 0.0712, 0.1230, 0.0516, 0.1650, 0.2145], device='cuda:3'), in_proj_covar=tensor([0.0154, 0.0064, 0.0051, 0.0057, 0.0066, 0.0050, 0.0069, 0.0083], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 08:29:38,566 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0790, 1.8220, 1.9185, 1.0435, 1.6734, 1.7447, 2.0058, 1.6374], device='cuda:3'), covar=tensor([0.0681, 0.4943, 0.1077, 0.3162, 0.1242, 0.0722, 0.1184, 0.1756], device='cuda:3'), in_proj_covar=tensor([0.0071, 0.0195, 0.0095, 0.0121, 0.0079, 0.0078, 0.0069, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 08:29:55,041 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1151, 1.8730, 2.8041, 1.6155, 2.1610, 2.4356, 1.1776, 2.4875], device='cuda:3'), covar=tensor([0.0894, 0.0979, 0.0508, 0.1724, 0.0909, 0.0676, 0.3015, 0.0547], device='cuda:3'), in_proj_covar=tensor([0.0064, 0.0070, 0.0063, 0.0072, 0.0079, 0.0065, 0.0124, 0.0071], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001], device='cuda:3') 2022-12-07 08:29:58,310 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.022e+02 2.546e+02 3.661e+02 5.045e+02 9.524e+02, threshold=7.322e+02, percent-clipped=3.0 2022-12-07 08:30:09,552 INFO [train.py:873] (3/4) Epoch 3, batch 1200, loss[loss=0.218, simple_loss=0.2302, pruned_loss=0.1029, over 14351.00 frames. ], tot_loss[loss=0.2188, simple_loss=0.2182, pruned_loss=0.1097, over 1952423.69 frames. ], batch size: 28, lr: 2.52e-02, grad_scale: 8.0 2022-12-07 08:30:37,105 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16357.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:30:48,996 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16371.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:31:23,701 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.462e+02 3.060e+02 3.691e+02 4.545e+02 9.377e+02, threshold=7.381e+02, percent-clipped=3.0 2022-12-07 08:31:35,383 INFO [train.py:873] (3/4) Epoch 3, batch 1300, loss[loss=0.2863, simple_loss=0.2399, pruned_loss=0.1663, over 4993.00 frames. ], tot_loss[loss=0.2207, simple_loss=0.2194, pruned_loss=0.111, over 1951114.06 frames. ], batch size: 100, lr: 2.51e-02, grad_scale: 8.0 2022-12-07 08:31:41,405 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16432.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:31:49,055 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16441.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:31:57,562 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8566, 1.3917, 3.0618, 2.7906, 3.0746, 2.8750, 2.3853, 3.1711], device='cuda:3'), covar=tensor([0.0911, 0.1123, 0.0091, 0.0182, 0.0132, 0.0128, 0.0237, 0.0085], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0140, 0.0074, 0.0106, 0.0087, 0.0092, 0.0070, 0.0074], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 08:32:20,890 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3002, 3.0844, 2.7411, 2.8330, 3.0556, 3.1208, 3.1408, 3.2380], device='cuda:3'), covar=tensor([0.0563, 0.0565, 0.1510, 0.2062, 0.0722, 0.0499, 0.0940, 0.0715], device='cuda:3'), in_proj_covar=tensor([0.0185, 0.0167, 0.0222, 0.0274, 0.0185, 0.0197, 0.0221, 0.0180], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-07 08:32:43,950 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16505.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:32:49,716 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.482e+02 2.808e+02 3.737e+02 4.916e+02 1.118e+03, threshold=7.474e+02, percent-clipped=8.0 2022-12-07 08:32:54,399 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8515, 1.5321, 3.9675, 3.5418, 3.8602, 3.8179, 3.2948, 4.0933], device='cuda:3'), covar=tensor([0.1272, 0.1408, 0.0083, 0.0140, 0.0113, 0.0100, 0.0201, 0.0073], device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0142, 0.0075, 0.0107, 0.0088, 0.0094, 0.0071, 0.0075], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 08:33:01,054 INFO [train.py:873] (3/4) Epoch 3, batch 1400, loss[loss=0.2164, simple_loss=0.2035, pruned_loss=0.1147, over 5015.00 frames. ], tot_loss[loss=0.2219, simple_loss=0.2199, pruned_loss=0.112, over 1943530.41 frames. ], batch size: 100, lr: 2.51e-02, grad_scale: 8.0 2022-12-07 08:33:12,911 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16539.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:33:54,092 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=16587.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:34:15,419 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.141e+01 3.095e+02 4.057e+02 5.380e+02 9.685e+02, threshold=8.113e+02, percent-clipped=8.0 2022-12-07 08:34:27,198 INFO [train.py:873] (3/4) Epoch 3, batch 1500, loss[loss=0.2165, simple_loss=0.1907, pruned_loss=0.1211, over 2650.00 frames. ], tot_loss[loss=0.2205, simple_loss=0.2189, pruned_loss=0.111, over 1994578.17 frames. ], batch size: 100, lr: 2.50e-02, grad_scale: 8.0 2022-12-07 08:34:38,470 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9055, 1.5950, 2.1370, 1.8691, 2.1816, 1.7531, 1.8783, 1.9037], device='cuda:3'), covar=tensor([0.0054, 0.0204, 0.0027, 0.0055, 0.0039, 0.0066, 0.0038, 0.0089], device='cuda:3'), in_proj_covar=tensor([0.0203, 0.0334, 0.0166, 0.0267, 0.0218, 0.0221, 0.0230, 0.0359], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:3') 2022-12-07 08:34:54,234 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16657.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:35:23,461 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=9.84 vs. limit=5.0 2022-12-07 08:35:25,602 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7770, 1.6238, 4.1564, 3.7256, 3.9666, 3.9713, 3.3309, 4.1621], device='cuda:3'), covar=tensor([0.1410, 0.1422, 0.0109, 0.0186, 0.0132, 0.0130, 0.0293, 0.0120], device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0142, 0.0076, 0.0107, 0.0088, 0.0094, 0.0072, 0.0075], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 08:35:35,817 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=16705.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:35:41,765 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.257e+02 2.784e+02 3.654e+02 4.523e+02 7.286e+02, threshold=7.308e+02, percent-clipped=0.0 2022-12-07 08:35:52,815 INFO [train.py:873] (3/4) Epoch 3, batch 1600, loss[loss=0.2688, simple_loss=0.2561, pruned_loss=0.1407, over 13925.00 frames. ], tot_loss[loss=0.2194, simple_loss=0.218, pruned_loss=0.1104, over 1980846.73 frames. ], batch size: 23, lr: 2.49e-02, grad_scale: 8.0 2022-12-07 08:35:53,814 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.0919, 0.6393, 0.7108, 1.2310, 0.9089, 0.2379, 1.2050, 1.3358], device='cuda:3'), covar=tensor([0.0384, 0.0272, 0.0159, 0.0457, 0.0522, 0.0157, 0.0358, 0.0070], device='cuda:3'), in_proj_covar=tensor([0.0017, 0.0019, 0.0019, 0.0016, 0.0018, 0.0022, 0.0017, 0.0014], device='cuda:3'), out_proj_covar=tensor([4.8277e-05, 5.1143e-05, 4.8239e-05, 4.7467e-05, 4.6496e-05, 5.5532e-05, 4.8101e-05, 3.8310e-05], device='cuda:3') 2022-12-07 08:35:54,530 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16727.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:36:06,519 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16741.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:36:15,105 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.38 vs. limit=5.0 2022-12-07 08:36:48,051 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=16789.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:37:01,940 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16805.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:37:07,697 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.245e+02 3.316e+02 4.284e+02 5.393e+02 9.437e+02, threshold=8.568e+02, percent-clipped=8.0 2022-12-07 08:37:11,041 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.4571, 4.8604, 4.7770, 5.2985, 5.1734, 4.5957, 5.2658, 4.5559], device='cuda:3'), covar=tensor([0.0215, 0.0652, 0.0217, 0.0346, 0.0487, 0.0269, 0.0385, 0.0389], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0166, 0.0106, 0.0103, 0.0100, 0.0092, 0.0154, 0.0121], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 08:37:16,714 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4360, 2.0453, 2.0448, 1.7303, 1.3235, 1.6209, 1.7744, 1.1395], device='cuda:3'), covar=tensor([0.3894, 0.1670, 0.1350, 0.2043, 0.1386, 0.1195, 0.2025, 0.3492], device='cuda:3'), in_proj_covar=tensor([0.0158, 0.0065, 0.0051, 0.0059, 0.0066, 0.0053, 0.0075, 0.0086], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 08:37:19,029 INFO [train.py:873] (3/4) Epoch 3, batch 1700, loss[loss=0.1995, simple_loss=0.1746, pruned_loss=0.1122, over 2607.00 frames. ], tot_loss[loss=0.2174, simple_loss=0.2168, pruned_loss=0.109, over 1964741.72 frames. ], batch size: 100, lr: 2.49e-02, grad_scale: 8.0 2022-12-07 08:37:42,665 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=16853.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:37:55,871 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16868.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 08:38:28,173 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16906.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:38:33,243 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.516e+02 2.887e+02 3.816e+02 4.798e+02 1.128e+03, threshold=7.633e+02, percent-clipped=3.0 2022-12-07 08:38:38,655 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0596, 1.6152, 2.4561, 2.1271, 2.5617, 2.0015, 2.1189, 2.1207], device='cuda:3'), covar=tensor([0.0096, 0.0450, 0.0043, 0.0115, 0.0065, 0.0106, 0.0063, 0.0210], device='cuda:3'), in_proj_covar=tensor([0.0210, 0.0327, 0.0165, 0.0265, 0.0220, 0.0216, 0.0229, 0.0353], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:3') 2022-12-07 08:38:42,268 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0676, 2.5996, 3.4675, 2.0177, 2.5114, 2.8176, 1.0320, 2.5101], device='cuda:3'), covar=tensor([0.3491, 0.0946, 0.0748, 0.2983, 0.1188, 0.1373, 0.4860, 0.1368], device='cuda:3'), in_proj_covar=tensor([0.0064, 0.0069, 0.0063, 0.0074, 0.0078, 0.0064, 0.0125, 0.0071], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001], device='cuda:3') 2022-12-07 08:38:44,615 INFO [train.py:873] (3/4) Epoch 3, batch 1800, loss[loss=0.2225, simple_loss=0.2242, pruned_loss=0.1104, over 14242.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.2166, pruned_loss=0.1082, over 1931475.43 frames. ], batch size: 80, lr: 2.48e-02, grad_scale: 8.0 2022-12-07 08:38:48,313 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16929.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 08:39:21,135 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16967.0, num_to_drop=1, layers_to_drop={3} 2022-12-07 08:39:48,266 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.9065, 0.8318, 0.6438, 0.9387, 0.8995, 0.4543, 0.9323, 0.8570], device='cuda:3'), covar=tensor([0.0282, 0.0159, 0.0271, 0.0220, 0.0179, 0.0164, 0.0439, 0.0156], device='cuda:3'), in_proj_covar=tensor([0.0015, 0.0018, 0.0019, 0.0016, 0.0016, 0.0020, 0.0016, 0.0014], device='cuda:3'), out_proj_covar=tensor([4.5632e-05, 4.8637e-05, 4.8291e-05, 4.7488e-05, 4.3581e-05, 5.1851e-05, 4.7073e-05, 3.7430e-05], device='cuda:3') 2022-12-07 08:39:59,563 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.130e+02 3.016e+02 3.995e+02 5.163e+02 9.262e+02, threshold=7.990e+02, percent-clipped=4.0 2022-12-07 08:39:59,776 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2798, 2.7600, 3.7232, 2.5327, 2.7181, 2.2869, 1.0419, 2.4431], device='cuda:3'), covar=tensor([0.1931, 0.0893, 0.0387, 0.0947, 0.0695, 0.1943, 0.3740, 0.1768], device='cuda:3'), in_proj_covar=tensor([0.0063, 0.0071, 0.0061, 0.0071, 0.0079, 0.0065, 0.0126, 0.0071], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001], device='cuda:3') 2022-12-07 08:40:02,349 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5930, 3.2654, 4.3310, 2.7384, 2.9180, 2.8439, 1.1392, 2.8558], device='cuda:3'), covar=tensor([0.1620, 0.0702, 0.0397, 0.0969, 0.0621, 0.1387, 0.3734, 0.1245], device='cuda:3'), in_proj_covar=tensor([0.0063, 0.0071, 0.0061, 0.0071, 0.0079, 0.0065, 0.0126, 0.0071], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001], device='cuda:3') 2022-12-07 08:40:11,094 INFO [train.py:873] (3/4) Epoch 3, batch 1900, loss[loss=0.2318, simple_loss=0.2295, pruned_loss=0.1171, over 14282.00 frames. ], tot_loss[loss=0.2176, simple_loss=0.2168, pruned_loss=0.1092, over 1932665.28 frames. ], batch size: 35, lr: 2.47e-02, grad_scale: 8.0 2022-12-07 08:40:13,004 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17027.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:40:40,937 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3676, 1.8002, 2.6029, 2.1233, 2.6821, 2.1315, 2.2816, 2.0869], device='cuda:3'), covar=tensor([0.0151, 0.0907, 0.0107, 0.0490, 0.0139, 0.0273, 0.0185, 0.0806], device='cuda:3'), in_proj_covar=tensor([0.0209, 0.0327, 0.0169, 0.0270, 0.0222, 0.0218, 0.0233, 0.0359], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:3') 2022-12-07 08:40:46,411 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.69 vs. limit=2.0 2022-12-07 08:40:53,614 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=17075.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:41:05,318 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.89 vs. limit=2.0 2022-12-07 08:41:25,312 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.512e+02 2.833e+02 3.865e+02 5.052e+02 1.376e+03, threshold=7.730e+02, percent-clipped=3.0 2022-12-07 08:41:36,267 INFO [train.py:873] (3/4) Epoch 3, batch 2000, loss[loss=0.2035, simple_loss=0.1695, pruned_loss=0.1187, over 1272.00 frames. ], tot_loss[loss=0.2172, simple_loss=0.2164, pruned_loss=0.109, over 1868731.36 frames. ], batch size: 100, lr: 2.47e-02, grad_scale: 16.0 2022-12-07 08:41:38,753 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.6631, 2.4211, 1.8751, 2.1510, 1.4453, 1.7904, 2.2388, 1.2986], device='cuda:3'), covar=tensor([0.4175, 0.1327, 0.3326, 0.1649, 0.1527, 0.1313, 0.1943, 0.3095], device='cuda:3'), in_proj_covar=tensor([0.0162, 0.0069, 0.0052, 0.0062, 0.0068, 0.0054, 0.0076, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:3') 2022-12-07 08:42:30,617 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17189.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:42:49,868 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.135e+02 3.144e+02 4.016e+02 5.229e+02 1.105e+03, threshold=8.031e+02, percent-clipped=5.0 2022-12-07 08:42:55,142 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8909, 1.6120, 1.9834, 1.3137, 1.5865, 1.5558, 1.7743, 1.6628], device='cuda:3'), covar=tensor([0.0335, 0.1348, 0.0547, 0.1660, 0.0900, 0.0759, 0.0528, 0.0885], device='cuda:3'), in_proj_covar=tensor([0.0069, 0.0193, 0.0090, 0.0118, 0.0076, 0.0077, 0.0068, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 08:43:00,786 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17224.0, num_to_drop=1, layers_to_drop={3} 2022-12-07 08:43:01,531 INFO [train.py:873] (3/4) Epoch 3, batch 2100, loss[loss=0.2289, simple_loss=0.2268, pruned_loss=0.1155, over 14267.00 frames. ], tot_loss[loss=0.2177, simple_loss=0.2168, pruned_loss=0.1093, over 1917293.51 frames. ], batch size: 44, lr: 2.46e-02, grad_scale: 16.0 2022-12-07 08:43:01,942 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0 2022-12-07 08:43:04,444 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2022-12-07 08:43:22,663 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17250.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:43:25,548 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.00 vs. limit=2.0 2022-12-07 08:43:32,864 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17262.0, num_to_drop=1, layers_to_drop={3} 2022-12-07 08:44:16,384 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.689e+02 2.883e+02 3.756e+02 4.821e+02 7.319e+02, threshold=7.512e+02, percent-clipped=0.0 2022-12-07 08:44:27,587 INFO [train.py:873] (3/4) Epoch 3, batch 2200, loss[loss=0.1956, simple_loss=0.2093, pruned_loss=0.09099, over 14136.00 frames. ], tot_loss[loss=0.2178, simple_loss=0.2172, pruned_loss=0.1092, over 1927256.25 frames. ], batch size: 29, lr: 2.45e-02, grad_scale: 16.0 2022-12-07 08:44:44,637 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7050, 1.7837, 1.6347, 1.6653, 1.3746, 1.6001, 1.4760, 0.9583], device='cuda:3'), covar=tensor([0.2730, 0.1157, 0.2032, 0.0635, 0.1401, 0.0798, 0.1794, 0.3353], device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0069, 0.0054, 0.0061, 0.0069, 0.0055, 0.0079, 0.0092], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:3') 2022-12-07 08:44:59,099 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17362.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:45:16,950 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17383.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 08:45:42,085 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.435e+02 3.126e+02 3.760e+02 4.600e+02 9.821e+02, threshold=7.519e+02, percent-clipped=3.0 2022-12-07 08:45:51,704 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17423.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:45:53,186 INFO [train.py:873] (3/4) Epoch 3, batch 2300, loss[loss=0.1925, simple_loss=0.2058, pruned_loss=0.08963, over 14198.00 frames. ], tot_loss[loss=0.2164, simple_loss=0.2166, pruned_loss=0.1081, over 2002407.44 frames. ], batch size: 89, lr: 2.45e-02, grad_scale: 16.0 2022-12-07 08:45:58,976 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0 2022-12-07 08:46:09,729 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17444.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 08:46:46,282 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.3715, 0.6205, 1.0383, 0.9302, 0.9603, 0.7486, 1.3512, 1.3037], device='cuda:3'), covar=tensor([0.0518, 0.0273, 0.0304, 0.0667, 0.0519, 0.0447, 0.0530, 0.0160], device='cuda:3'), in_proj_covar=tensor([0.0015, 0.0019, 0.0020, 0.0017, 0.0017, 0.0021, 0.0016, 0.0015], device='cuda:3'), out_proj_covar=tensor([4.6413e-05, 5.1211e-05, 4.9433e-05, 5.1345e-05, 4.6064e-05, 5.4188e-05, 4.7838e-05, 3.9383e-05], device='cuda:3') 2022-12-07 08:46:46,655 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.01 vs. limit=2.0 2022-12-07 08:47:07,374 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.068e+02 2.910e+02 3.962e+02 5.251e+02 8.178e+02, threshold=7.923e+02, percent-clipped=2.0 2022-12-07 08:47:17,992 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17524.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 08:47:18,683 INFO [train.py:873] (3/4) Epoch 3, batch 2400, loss[loss=0.2282, simple_loss=0.2301, pruned_loss=0.1132, over 14287.00 frames. ], tot_loss[loss=0.2184, simple_loss=0.2182, pruned_loss=0.1093, over 2027928.53 frames. ], batch size: 39, lr: 2.44e-02, grad_scale: 16.0 2022-12-07 08:47:33,056 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6121, 3.0781, 3.4037, 3.6357, 3.4478, 3.5165, 3.5789, 3.0620], device='cuda:3'), covar=tensor([0.0826, 0.1562, 0.0737, 0.0790, 0.1036, 0.0957, 0.0991, 0.0942], device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0163, 0.0106, 0.0101, 0.0100, 0.0093, 0.0156, 0.0117], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 08:47:35,596 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17545.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:47:47,919 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 2022-12-07 08:47:50,066 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17562.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 08:47:50,089 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5820, 2.1676, 2.0271, 1.1512, 2.2369, 2.1184, 2.4475, 1.6629], device='cuda:3'), covar=tensor([0.0727, 0.5075, 0.1427, 0.3656, 0.0997, 0.0578, 0.1165, 0.2043], device='cuda:3'), in_proj_covar=tensor([0.0071, 0.0203, 0.0096, 0.0124, 0.0080, 0.0079, 0.0070, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 08:47:58,497 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=17572.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 08:48:31,674 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=17610.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:48:34,156 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.459e+02 2.796e+02 3.862e+02 5.213e+02 1.653e+03, threshold=7.725e+02, percent-clipped=7.0 2022-12-07 08:48:44,389 INFO [train.py:873] (3/4) Epoch 3, batch 2500, loss[loss=0.2618, simple_loss=0.2415, pruned_loss=0.1411, over 9480.00 frames. ], tot_loss[loss=0.2179, simple_loss=0.2175, pruned_loss=0.1091, over 1966893.71 frames. ], batch size: 100, lr: 2.43e-02, grad_scale: 8.0 2022-12-07 08:49:22,420 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.89 vs. limit=2.0 2022-12-07 08:49:25,751 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.62 vs. limit=2.0 2022-12-07 08:49:26,935 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0 2022-12-07 08:49:49,060 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17700.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:49:59,744 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.003e+02 2.968e+02 3.839e+02 5.013e+02 8.291e+02, threshold=7.678e+02, percent-clipped=4.0 2022-12-07 08:50:04,191 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17718.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:50:10,238 INFO [train.py:873] (3/4) Epoch 3, batch 2600, loss[loss=0.1832, simple_loss=0.2018, pruned_loss=0.08231, over 13959.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.217, pruned_loss=0.109, over 1966367.40 frames. ], batch size: 26, lr: 2.43e-02, grad_scale: 8.0 2022-12-07 08:50:14,753 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17730.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:50:19,316 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 2022-12-07 08:50:22,573 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17739.0, num_to_drop=1, layers_to_drop={3} 2022-12-07 08:50:41,062 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17761.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:51:06,131 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9632, 1.5557, 2.4492, 1.9132, 2.4732, 1.4902, 2.0384, 2.1058], device='cuda:3'), covar=tensor([0.0700, 0.3187, 0.0262, 0.3901, 0.0205, 0.2650, 0.0975, 0.0445], device='cuda:3'), in_proj_covar=tensor([0.0237, 0.0329, 0.0181, 0.0418, 0.0169, 0.0329, 0.0301, 0.0175], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0002, 0.0004, 0.0002, 0.0004, 0.0003, 0.0002], device='cuda:3') 2022-12-07 08:51:06,957 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17791.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:51:10,894 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2022-12-07 08:51:13,649 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.30 vs. limit=5.0 2022-12-07 08:51:25,440 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.50 vs. limit=5.0 2022-12-07 08:51:25,831 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.679e+02 2.813e+02 3.769e+02 4.969e+02 1.379e+03, threshold=7.537e+02, percent-clipped=10.0 2022-12-07 08:51:36,296 INFO [train.py:873] (3/4) Epoch 3, batch 2700, loss[loss=0.1988, simple_loss=0.1997, pruned_loss=0.0989, over 14232.00 frames. ], tot_loss[loss=0.2171, simple_loss=0.2171, pruned_loss=0.1086, over 1990080.06 frames. ], batch size: 37, lr: 2.42e-02, grad_scale: 8.0 2022-12-07 08:51:53,199 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17845.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:52:01,962 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4396, 1.4466, 1.8300, 2.0177, 1.3774, 1.4851, 1.7696, 1.7096], device='cuda:3'), covar=tensor([0.0046, 0.0061, 0.0028, 0.0019, 0.0063, 0.0098, 0.0026, 0.0032], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0154, 0.0183, 0.0163, 0.0135, 0.0199, 0.0097, 0.0188], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0001, 0.0002], device='cuda:3') 2022-12-07 08:52:34,445 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=17893.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:52:51,606 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.345e+02 3.286e+02 4.021e+02 4.924e+02 9.078e+02, threshold=8.042e+02, percent-clipped=2.0 2022-12-07 08:53:01,816 INFO [train.py:873] (3/4) Epoch 3, batch 2800, loss[loss=0.2095, simple_loss=0.2146, pruned_loss=0.1022, over 14420.00 frames. ], tot_loss[loss=0.2172, simple_loss=0.217, pruned_loss=0.1087, over 1934558.56 frames. ], batch size: 73, lr: 2.41e-02, grad_scale: 8.0 2022-12-07 08:53:31,757 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17960.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:53:42,973 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.87 vs. limit=2.0 2022-12-07 08:53:49,988 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.02 vs. limit=2.0 2022-12-07 08:54:16,684 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.275e+02 2.972e+02 3.622e+02 4.500e+02 9.762e+02, threshold=7.243e+02, percent-clipped=2.0 2022-12-07 08:54:21,522 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18018.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:54:24,236 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18021.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:54:27,361 INFO [train.py:873] (3/4) Epoch 3, batch 2900, loss[loss=0.2217, simple_loss=0.2195, pruned_loss=0.1119, over 9502.00 frames. ], tot_loss[loss=0.2164, simple_loss=0.2166, pruned_loss=0.1081, over 1941710.16 frames. ], batch size: 100, lr: 2.41e-02, grad_scale: 8.0 2022-12-07 08:54:38,868 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1549, 0.3979, 1.1104, 0.8798, 0.8139, 0.7018, 1.2796, 0.9807], device='cuda:3'), covar=tensor([0.0656, 0.0756, 0.0216, 0.0519, 0.0450, 0.0194, 0.0720, 0.0588], device='cuda:3'), in_proj_covar=tensor([0.0015, 0.0020, 0.0020, 0.0017, 0.0018, 0.0021, 0.0016, 0.0017], device='cuda:3'), out_proj_covar=tensor([4.7098e-05, 5.3700e-05, 5.2230e-05, 5.1235e-05, 4.8348e-05, 5.4010e-05, 4.8428e-05, 4.4041e-05], device='cuda:3') 2022-12-07 08:54:39,652 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18039.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 08:54:53,293 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18056.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:55:01,983 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18066.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:55:19,252 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18086.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:55:20,064 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18087.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 08:55:27,378 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2022-12-07 08:55:42,573 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.345e+02 3.066e+02 4.043e+02 5.337e+02 1.703e+03, threshold=8.087e+02, percent-clipped=7.0 2022-12-07 08:55:52,804 INFO [train.py:873] (3/4) Epoch 3, batch 3000, loss[loss=0.1649, simple_loss=0.1438, pruned_loss=0.09299, over 1281.00 frames. ], tot_loss[loss=0.2167, simple_loss=0.2165, pruned_loss=0.1085, over 1943414.30 frames. ], batch size: 100, lr: 2.40e-02, grad_scale: 8.0 2022-12-07 08:55:52,804 INFO [train.py:896] (3/4) Computing validation loss 2022-12-07 08:55:59,459 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1130, 3.9532, 3.5437, 3.7428, 3.8264, 4.1975, 4.2797, 4.1759], device='cuda:3'), covar=tensor([0.0648, 0.0526, 0.1141, 0.1827, 0.0629, 0.0355, 0.0535, 0.0674], device='cuda:3'), in_proj_covar=tensor([0.0213, 0.0176, 0.0246, 0.0309, 0.0198, 0.0220, 0.0232, 0.0201], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 08:56:00,210 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4201, 3.2419, 2.5802, 2.3113, 2.5998, 2.9009, 3.0029, 2.4661], device='cuda:3'), covar=tensor([0.0467, 0.3156, 0.1273, 0.2389, 0.1184, 0.0446, 0.0667, 0.1368], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0212, 0.0099, 0.0129, 0.0083, 0.0082, 0.0072, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:3') 2022-12-07 08:56:03,704 INFO [train.py:905] (3/4) Epoch 3, validation: loss=0.1337, simple_loss=0.176, pruned_loss=0.04573, over 857387.00 frames. 2022-12-07 08:56:03,705 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17841MB 2022-12-07 08:56:37,772 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.02 vs. limit=2.0 2022-12-07 08:56:46,099 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18174.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:56:46,940 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18175.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:57:05,869 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 2022-12-07 08:57:19,775 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.296e+02 3.113e+02 4.054e+02 5.125e+02 9.563e+02, threshold=8.108e+02, percent-clipped=4.0 2022-12-07 08:57:30,105 INFO [train.py:873] (3/4) Epoch 3, batch 3100, loss[loss=0.2139, simple_loss=0.2167, pruned_loss=0.1055, over 13997.00 frames. ], tot_loss[loss=0.215, simple_loss=0.2154, pruned_loss=0.1073, over 1919441.00 frames. ], batch size: 19, lr: 2.40e-02, grad_scale: 8.0 2022-12-07 08:57:38,707 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18235.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:57:39,529 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18236.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:58:00,415 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3834, 3.0731, 2.2219, 3.4820, 3.1822, 3.5314, 2.9777, 2.2908], device='cuda:3'), covar=tensor([0.0176, 0.0572, 0.2374, 0.0120, 0.0254, 0.0215, 0.0586, 0.2465], device='cuda:3'), in_proj_covar=tensor([0.0207, 0.0271, 0.0340, 0.0165, 0.0202, 0.0192, 0.0251, 0.0350], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0001, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:3') 2022-12-07 08:58:09,804 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.78 vs. limit=2.0 2022-12-07 08:58:39,521 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2021, 1.4766, 4.3201, 4.1797, 4.0209, 4.3341, 3.7440, 4.4023], device='cuda:3'), covar=tensor([0.1141, 0.1544, 0.0085, 0.0111, 0.0109, 0.0083, 0.0208, 0.0074], device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0142, 0.0080, 0.0110, 0.0091, 0.0098, 0.0074, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 08:58:45,517 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.240e+02 2.872e+02 3.739e+02 4.989e+02 1.887e+03, threshold=7.477e+02, percent-clipped=4.0 2022-12-07 08:58:48,231 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18316.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:58:55,687 INFO [train.py:873] (3/4) Epoch 3, batch 3200, loss[loss=0.2265, simple_loss=0.2259, pruned_loss=0.1135, over 14507.00 frames. ], tot_loss[loss=0.2142, simple_loss=0.2146, pruned_loss=0.1069, over 1905871.10 frames. ], batch size: 49, lr: 2.39e-02, grad_scale: 8.0 2022-12-07 08:59:22,765 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18356.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 08:59:48,438 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18386.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:00:03,911 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18404.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:00:11,699 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.284e+02 3.073e+02 3.846e+02 5.259e+02 1.192e+03, threshold=7.691e+02, percent-clipped=7.0 2022-12-07 09:00:13,866 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.6286, 0.5612, 0.1402, 0.7473, 0.6415, 0.0350, 0.8281, 0.7001], device='cuda:3'), covar=tensor([0.0062, 0.0053, 0.0013, 0.0074, 0.0021, 0.0018, 0.0059, 0.0035], device='cuda:3'), in_proj_covar=tensor([0.0015, 0.0021, 0.0022, 0.0017, 0.0019, 0.0023, 0.0017, 0.0017], device='cuda:3'), out_proj_covar=tensor([4.6634e-05, 5.6238e-05, 5.5761e-05, 5.3063e-05, 5.1378e-05, 5.8322e-05, 4.9101e-05, 4.5339e-05], device='cuda:3') 2022-12-07 09:00:22,199 INFO [train.py:873] (3/4) Epoch 3, batch 3300, loss[loss=0.2036, simple_loss=0.2087, pruned_loss=0.09922, over 14173.00 frames. ], tot_loss[loss=0.2152, simple_loss=0.215, pruned_loss=0.1077, over 1907023.92 frames. ], batch size: 89, lr: 2.38e-02, grad_scale: 8.0 2022-12-07 09:00:29,918 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18434.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:01:18,393 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18490.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:01:33,586 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18508.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 09:01:37,436 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.740e+02 3.014e+02 3.926e+02 4.952e+02 1.093e+03, threshold=7.852e+02, percent-clipped=5.0 2022-12-07 09:01:48,039 INFO [train.py:873] (3/4) Epoch 3, batch 3400, loss[loss=0.225, simple_loss=0.2289, pruned_loss=0.1105, over 14673.00 frames. ], tot_loss[loss=0.2135, simple_loss=0.214, pruned_loss=0.1065, over 1915671.07 frames. ], batch size: 23, lr: 2.38e-02, grad_scale: 8.0 2022-12-07 09:01:52,318 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18530.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:01:52,714 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2022-12-07 09:01:53,153 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18531.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:02:10,088 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18551.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:02:24,109 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.33 vs. limit=5.0 2022-12-07 09:02:25,463 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18569.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 09:02:33,766 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.5122, 1.3643, 1.4710, 1.2832, 1.0893, 1.4000, 1.2959, 0.6727], device='cuda:3'), covar=tensor([0.5833, 0.2531, 0.1546, 0.2018, 0.1630, 0.1142, 0.1779, 0.4192], device='cuda:3'), in_proj_covar=tensor([0.0161, 0.0068, 0.0050, 0.0059, 0.0066, 0.0057, 0.0075, 0.0092], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:3') 2022-12-07 09:02:48,482 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.5777, 5.1336, 5.1949, 5.6524, 5.4213, 4.5084, 5.7835, 5.6995], device='cuda:3'), covar=tensor([0.0636, 0.0420, 0.0606, 0.0562, 0.0466, 0.0357, 0.0481, 0.0482], device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0079, 0.0102, 0.0094, 0.0103, 0.0070, 0.0102, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2022-12-07 09:03:03,541 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.675e+02 3.062e+02 4.109e+02 5.827e+02 9.337e+02, threshold=8.217e+02, percent-clipped=6.0 2022-12-07 09:03:06,402 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18616.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:03:13,885 INFO [train.py:873] (3/4) Epoch 3, batch 3500, loss[loss=0.2074, simple_loss=0.1931, pruned_loss=0.1108, over 4979.00 frames. ], tot_loss[loss=0.2142, simple_loss=0.2144, pruned_loss=0.107, over 1913124.53 frames. ], batch size: 100, lr: 2.37e-02, grad_scale: 8.0 2022-12-07 09:03:41,197 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.72 vs. limit=2.0 2022-12-07 09:03:46,491 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18664.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:03:54,845 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18674.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:04:16,120 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2022-12-07 09:04:28,294 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.467e+02 2.925e+02 3.819e+02 4.873e+02 8.102e+02, threshold=7.639e+02, percent-clipped=0.0 2022-12-07 09:04:38,566 INFO [train.py:873] (3/4) Epoch 3, batch 3600, loss[loss=0.2038, simple_loss=0.2038, pruned_loss=0.1018, over 14029.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.2135, pruned_loss=0.1059, over 1913453.71 frames. ], batch size: 19, lr: 2.37e-02, grad_scale: 8.0 2022-12-07 09:04:47,046 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 2022-12-07 09:04:47,553 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1892, 3.7878, 4.1848, 3.9118, 4.0920, 4.3621, 1.6607, 3.8444], device='cuda:3'), covar=tensor([0.0217, 0.0358, 0.0633, 0.0370, 0.0355, 0.0167, 0.3379, 0.0307], device='cuda:3'), in_proj_covar=tensor([0.0112, 0.0115, 0.0112, 0.0094, 0.0153, 0.0101, 0.0148, 0.0142], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 09:04:47,644 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18735.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:05:01,388 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1917, 1.4524, 1.4839, 1.4908, 1.2968, 1.4893, 1.1328, 0.7890], device='cuda:3'), covar=tensor([0.2162, 0.0909, 0.0417, 0.0637, 0.0894, 0.0501, 0.1279, 0.1823], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0066, 0.0050, 0.0059, 0.0066, 0.0055, 0.0071, 0.0091], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:3') 2022-12-07 09:05:07,713 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.19 vs. limit=2.0 2022-12-07 09:05:54,916 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.731e+02 3.054e+02 3.647e+02 5.491e+02 1.580e+03, threshold=7.295e+02, percent-clipped=4.0 2022-12-07 09:06:05,677 INFO [train.py:873] (3/4) Epoch 3, batch 3700, loss[loss=0.2147, simple_loss=0.2167, pruned_loss=0.1063, over 14145.00 frames. ], tot_loss[loss=0.2141, simple_loss=0.2141, pruned_loss=0.107, over 1851858.29 frames. ], batch size: 84, lr: 2.36e-02, grad_scale: 8.0 2022-12-07 09:06:09,909 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18830.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:06:10,742 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18831.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:06:19,265 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.57 vs. limit=2.0 2022-12-07 09:06:23,471 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18846.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:06:29,937 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6497, 1.4739, 1.8583, 2.0193, 1.4568, 1.5506, 1.7990, 1.7653], device='cuda:3'), covar=tensor([0.0022, 0.0036, 0.0026, 0.0015, 0.0031, 0.0063, 0.0019, 0.0027], device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0154, 0.0186, 0.0162, 0.0139, 0.0200, 0.0103, 0.0188], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0001, 0.0002], device='cuda:3') 2022-12-07 09:06:32,709 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.5790, 5.3245, 5.2646, 5.6121, 5.4936, 4.4646, 5.7998, 5.5670], device='cuda:3'), covar=tensor([0.0585, 0.0353, 0.0559, 0.0501, 0.0467, 0.0614, 0.0569, 0.0537], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0082, 0.0103, 0.0096, 0.0106, 0.0071, 0.0104, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2022-12-07 09:06:39,780 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18864.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 09:06:51,527 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18878.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:06:52,336 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18879.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:07:02,907 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 2022-12-07 09:07:21,106 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.590e+02 2.825e+02 3.727e+02 4.870e+02 9.642e+02, threshold=7.453e+02, percent-clipped=3.0 2022-12-07 09:07:31,081 INFO [train.py:873] (3/4) Epoch 3, batch 3800, loss[loss=0.2576, simple_loss=0.2039, pruned_loss=0.1557, over 1225.00 frames. ], tot_loss[loss=0.2148, simple_loss=0.2149, pruned_loss=0.1074, over 1863083.05 frames. ], batch size: 100, lr: 2.35e-02, grad_scale: 8.0 2022-12-07 09:07:35,844 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18930.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:07:40,593 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.35 vs. limit=5.0 2022-12-07 09:07:49,877 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.3137, 1.4614, 1.5367, 1.6086, 1.3522, 1.5163, 1.1894, 0.8575], device='cuda:3'), covar=tensor([0.2803, 0.1423, 0.1129, 0.0747, 0.1356, 0.0532, 0.1518, 0.3227], device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0067, 0.0052, 0.0060, 0.0070, 0.0059, 0.0075, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:3') 2022-12-07 09:08:12,477 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.89 vs. limit=5.0 2022-12-07 09:08:27,847 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0886, 3.6395, 2.8587, 4.3075, 3.7126, 4.1595, 3.6756, 2.9962], device='cuda:3'), covar=tensor([0.0265, 0.0483, 0.2033, 0.0124, 0.0205, 0.0404, 0.0443, 0.2215], device='cuda:3'), in_proj_covar=tensor([0.0214, 0.0280, 0.0342, 0.0171, 0.0210, 0.0201, 0.0252, 0.0350], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0001, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:3') 2022-12-07 09:08:28,661 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18991.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:08:33,728 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3477, 1.7420, 3.3941, 2.4239, 3.3202, 1.8609, 2.4664, 3.0674], device='cuda:3'), covar=tensor([0.0441, 0.4335, 0.0337, 0.6034, 0.0208, 0.3196, 0.1158, 0.0258], device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0320, 0.0176, 0.0414, 0.0170, 0.0329, 0.0293, 0.0176], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0002, 0.0004, 0.0002, 0.0004, 0.0003, 0.0002], device='cuda:3') 2022-12-07 09:08:41,102 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.94 vs. limit=2.0 2022-12-07 09:08:47,291 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.617e+02 3.029e+02 3.748e+02 4.757e+02 1.336e+03, threshold=7.495e+02, percent-clipped=2.0 2022-12-07 09:08:58,126 INFO [train.py:873] (3/4) Epoch 3, batch 3900, loss[loss=0.2069, simple_loss=0.2188, pruned_loss=0.09748, over 14601.00 frames. ], tot_loss[loss=0.2129, simple_loss=0.2141, pruned_loss=0.1059, over 1928651.35 frames. ], batch size: 22, lr: 2.35e-02, grad_scale: 8.0 2022-12-07 09:09:02,362 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19030.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:10:06,044 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0795, 3.7340, 3.6851, 4.1551, 3.9183, 3.7125, 4.1558, 3.5875], device='cuda:3'), covar=tensor([0.0460, 0.0941, 0.0361, 0.0382, 0.0671, 0.0878, 0.0479, 0.0485], device='cuda:3'), in_proj_covar=tensor([0.0106, 0.0171, 0.0117, 0.0109, 0.0106, 0.0096, 0.0163, 0.0125], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 09:10:13,854 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.275e+02 2.994e+02 3.505e+02 4.615e+02 1.018e+03, threshold=7.009e+02, percent-clipped=3.0 2022-12-07 09:10:23,710 INFO [train.py:873] (3/4) Epoch 3, batch 4000, loss[loss=0.2198, simple_loss=0.1974, pruned_loss=0.1211, over 3884.00 frames. ], tot_loss[loss=0.2129, simple_loss=0.214, pruned_loss=0.1058, over 1921899.92 frames. ], batch size: 100, lr: 2.34e-02, grad_scale: 8.0 2022-12-07 09:10:27,230 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19129.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 09:10:28,326 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.59 vs. limit=2.0 2022-12-07 09:10:42,459 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19146.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:10:57,687 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19164.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 09:11:20,402 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19190.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 09:11:23,560 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19194.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:11:39,106 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19212.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 09:11:39,768 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.420e+02 2.771e+02 3.795e+02 5.006e+02 9.210e+02, threshold=7.589e+02, percent-clipped=4.0 2022-12-07 09:11:50,340 INFO [train.py:873] (3/4) Epoch 3, batch 4100, loss[loss=0.1865, simple_loss=0.1831, pruned_loss=0.09491, over 3848.00 frames. ], tot_loss[loss=0.2119, simple_loss=0.2131, pruned_loss=0.1054, over 1891831.52 frames. ], batch size: 100, lr: 2.34e-02, grad_scale: 8.0 2022-12-07 09:12:11,031 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.96 vs. limit=2.0 2022-12-07 09:12:23,724 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 2022-12-07 09:12:42,489 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19286.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:12:51,188 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.1891, 4.6014, 4.8216, 5.2060, 4.7875, 4.2000, 5.2718, 5.1339], device='cuda:3'), covar=tensor([0.0503, 0.0598, 0.0494, 0.0497, 0.0496, 0.0437, 0.0441, 0.0548], device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0082, 0.0101, 0.0097, 0.0104, 0.0070, 0.0101, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2022-12-07 09:12:53,478 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0134, 1.7686, 2.3630, 2.4732, 2.0708, 1.6575, 2.5280, 1.9235], device='cuda:3'), covar=tensor([0.0047, 0.0111, 0.0088, 0.0041, 0.0057, 0.0180, 0.0039, 0.0112], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0158, 0.0194, 0.0165, 0.0141, 0.0207, 0.0107, 0.0194], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-07 09:13:05,132 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.016e+02 2.865e+02 3.600e+02 4.882e+02 1.320e+03, threshold=7.199e+02, percent-clipped=6.0 2022-12-07 09:13:15,263 INFO [train.py:873] (3/4) Epoch 3, batch 4200, loss[loss=0.1918, simple_loss=0.2083, pruned_loss=0.08765, over 14269.00 frames. ], tot_loss[loss=0.2112, simple_loss=0.2133, pruned_loss=0.1046, over 1972556.29 frames. ], batch size: 28, lr: 2.33e-02, grad_scale: 8.0 2022-12-07 09:13:19,761 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19330.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:13:28,570 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.06 vs. limit=2.0 2022-12-07 09:14:01,572 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19378.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:14:15,210 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19394.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:14:31,171 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.339e+02 3.180e+02 3.832e+02 4.985e+02 8.220e+02, threshold=7.663e+02, percent-clipped=2.0 2022-12-07 09:14:42,548 INFO [train.py:873] (3/4) Epoch 3, batch 4300, loss[loss=0.2046, simple_loss=0.2101, pruned_loss=0.09956, over 14522.00 frames. ], tot_loss[loss=0.2125, simple_loss=0.214, pruned_loss=0.1054, over 1985330.26 frames. ], batch size: 49, lr: 2.33e-02, grad_scale: 8.0 2022-12-07 09:14:53,507 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3284, 3.8740, 4.3468, 3.8608, 4.1279, 4.3434, 1.7458, 4.1152], device='cuda:3'), covar=tensor([0.0136, 0.0294, 0.0326, 0.0338, 0.0277, 0.0147, 0.2728, 0.0220], device='cuda:3'), in_proj_covar=tensor([0.0111, 0.0117, 0.0111, 0.0093, 0.0151, 0.0103, 0.0149, 0.0141], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 09:14:54,326 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9375, 3.7804, 3.4439, 3.4819, 3.6886, 3.8564, 4.0039, 3.8857], device='cuda:3'), covar=tensor([0.0809, 0.0522, 0.1374, 0.2233, 0.0740, 0.0578, 0.0696, 0.0853], device='cuda:3'), in_proj_covar=tensor([0.0214, 0.0181, 0.0250, 0.0319, 0.0201, 0.0233, 0.0244, 0.0200], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 09:15:08,490 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19455.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:15:31,310 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.78 vs. limit=2.0 2022-12-07 09:15:33,971 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19485.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 09:15:57,829 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.656e+02 2.698e+02 3.428e+02 4.598e+02 7.579e+02, threshold=6.855e+02, percent-clipped=0.0 2022-12-07 09:16:08,044 INFO [train.py:873] (3/4) Epoch 3, batch 4400, loss[loss=0.232, simple_loss=0.2294, pruned_loss=0.1173, over 8575.00 frames. ], tot_loss[loss=0.2128, simple_loss=0.2142, pruned_loss=0.1058, over 1932481.25 frames. ], batch size: 100, lr: 2.32e-02, grad_scale: 8.0 2022-12-07 09:16:50,602 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1979, 3.4976, 2.7489, 1.8473, 3.0820, 3.0668, 3.0796, 2.2367], device='cuda:3'), covar=tensor([0.0393, 0.2243, 0.0714, 0.2190, 0.0551, 0.0303, 0.0998, 0.1111], device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0206, 0.0096, 0.0127, 0.0085, 0.0080, 0.0070, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0005, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:3') 2022-12-07 09:16:57,291 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.06 vs. limit=2.0 2022-12-07 09:17:00,308 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19586.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:17:23,968 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.141e+02 2.902e+02 3.784e+02 4.919e+02 1.720e+03, threshold=7.568e+02, percent-clipped=8.0 2022-12-07 09:17:28,292 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19618.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:17:34,597 INFO [train.py:873] (3/4) Epoch 3, batch 4500, loss[loss=0.2077, simple_loss=0.2244, pruned_loss=0.0955, over 14388.00 frames. ], tot_loss[loss=0.2119, simple_loss=0.214, pruned_loss=0.105, over 2009637.44 frames. ], batch size: 53, lr: 2.31e-02, grad_scale: 8.0 2022-12-07 09:17:42,285 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19634.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:17:58,210 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4320, 4.0476, 3.9919, 4.4341, 4.1194, 3.4831, 4.4717, 4.4143], device='cuda:3'), covar=tensor([0.0757, 0.0537, 0.0693, 0.0691, 0.0759, 0.0618, 0.0688, 0.0819], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0081, 0.0102, 0.0098, 0.0107, 0.0074, 0.0105, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2022-12-07 09:18:04,426 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.06 vs. limit=5.0 2022-12-07 09:18:18,788 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.02 vs. limit=2.0 2022-12-07 09:18:20,831 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19679.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:18:50,393 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.725e+01 3.027e+02 3.828e+02 4.881e+02 1.178e+03, threshold=7.656e+02, percent-clipped=6.0 2022-12-07 09:18:59,480 INFO [train.py:873] (3/4) Epoch 3, batch 4600, loss[loss=0.1872, simple_loss=0.1917, pruned_loss=0.09139, over 4986.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.2141, pruned_loss=0.1055, over 1953721.28 frames. ], batch size: 100, lr: 2.31e-02, grad_scale: 8.0 2022-12-07 09:19:21,732 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19750.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:19:52,088 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19785.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 09:20:16,184 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.596e+02 3.281e+02 3.991e+02 5.326e+02 1.100e+03, threshold=7.982e+02, percent-clipped=6.0 2022-12-07 09:20:26,814 INFO [train.py:873] (3/4) Epoch 3, batch 4700, loss[loss=0.1876, simple_loss=0.2018, pruned_loss=0.08673, over 14311.00 frames. ], tot_loss[loss=0.21, simple_loss=0.2125, pruned_loss=0.1038, over 1966503.29 frames. ], batch size: 31, lr: 2.30e-02, grad_scale: 8.0 2022-12-07 09:20:30,147 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8381, 4.7052, 4.4760, 5.0189, 4.6813, 3.9231, 5.1211, 4.8780], device='cuda:3'), covar=tensor([0.0755, 0.0541, 0.0592, 0.0568, 0.0593, 0.0518, 0.0567, 0.0799], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0081, 0.0100, 0.0097, 0.0107, 0.0074, 0.0104, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2022-12-07 09:20:33,478 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19833.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 09:20:55,151 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2022-12-07 09:21:19,469 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3151, 2.0513, 5.1628, 4.6268, 4.6595, 5.1694, 4.7042, 5.2561], device='cuda:3'), covar=tensor([0.0994, 0.1079, 0.0044, 0.0052, 0.0067, 0.0053, 0.0065, 0.0037], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0145, 0.0084, 0.0110, 0.0094, 0.0103, 0.0077, 0.0083], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 09:21:23,472 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.91 vs. limit=2.0 2022-12-07 09:21:43,694 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.462e+02 3.110e+02 3.871e+02 4.666e+02 7.831e+02, threshold=7.741e+02, percent-clipped=0.0 2022-12-07 09:21:47,551 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.64 vs. limit=2.0 2022-12-07 09:21:52,843 INFO [train.py:873] (3/4) Epoch 3, batch 4800, loss[loss=0.2272, simple_loss=0.1869, pruned_loss=0.1338, over 1322.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.212, pruned_loss=0.1036, over 1958070.50 frames. ], batch size: 100, lr: 2.30e-02, grad_scale: 8.0 2022-12-07 09:22:22,753 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.63 vs. limit=2.0 2022-12-07 09:22:35,306 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19974.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:22:48,576 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.2948, 3.4154, 2.8218, 4.5329, 4.0448, 4.5231, 3.4851, 3.2573], device='cuda:3'), covar=tensor([0.0273, 0.0780, 0.2719, 0.0221, 0.0265, 0.0422, 0.0758, 0.2305], device='cuda:3'), in_proj_covar=tensor([0.0208, 0.0286, 0.0336, 0.0174, 0.0224, 0.0205, 0.0258, 0.0339], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:3') 2022-12-07 09:23:13,337 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.353e+02 2.781e+02 3.610e+02 5.418e+02 1.055e+03, threshold=7.220e+02, percent-clipped=5.0 2022-12-07 09:23:18,905 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.66 vs. limit=2.0 2022-12-07 09:23:19,235 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20020.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:23:23,445 INFO [train.py:873] (3/4) Epoch 3, batch 4900, loss[loss=0.1925, simple_loss=0.205, pruned_loss=0.08997, over 14615.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.2142, pruned_loss=0.1055, over 2009118.68 frames. ], batch size: 22, lr: 2.29e-02, grad_scale: 8.0 2022-12-07 09:23:44,338 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20050.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:24:10,924 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20081.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:24:11,492 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7719, 3.4321, 3.4851, 3.7843, 3.6688, 3.5983, 3.8148, 3.2566], device='cuda:3'), covar=tensor([0.0427, 0.0795, 0.0342, 0.0410, 0.0557, 0.0553, 0.0531, 0.0572], device='cuda:3'), in_proj_covar=tensor([0.0107, 0.0174, 0.0116, 0.0109, 0.0107, 0.0099, 0.0166, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 09:24:17,514 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20089.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:24:19,088 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2111, 1.4661, 2.6645, 1.9175, 2.4890, 1.6365, 1.9516, 2.1266], device='cuda:3'), covar=tensor([0.0804, 0.4756, 0.0331, 0.4168, 0.0265, 0.3346, 0.1488, 0.0434], device='cuda:3'), in_proj_covar=tensor([0.0235, 0.0315, 0.0177, 0.0410, 0.0172, 0.0321, 0.0290, 0.0174], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0002, 0.0004, 0.0002, 0.0004, 0.0003, 0.0002], device='cuda:3') 2022-12-07 09:24:25,504 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20098.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:24:37,897 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20112.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:24:39,472 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.119e+02 3.078e+02 3.886e+02 5.087e+02 8.123e+02, threshold=7.771e+02, percent-clipped=4.0 2022-12-07 09:24:48,396 INFO [train.py:873] (3/4) Epoch 3, batch 5000, loss[loss=0.2211, simple_loss=0.2185, pruned_loss=0.1118, over 9502.00 frames. ], tot_loss[loss=0.2101, simple_loss=0.2125, pruned_loss=0.1039, over 1970915.47 frames. ], batch size: 100, lr: 2.29e-02, grad_scale: 8.0 2022-12-07 09:25:10,531 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20150.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:25:29,144 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.81 vs. limit=2.0 2022-12-07 09:25:29,554 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20173.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:25:44,627 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.83 vs. limit=2.0 2022-12-07 09:25:56,550 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20204.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:25:58,764 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=7.56 vs. limit=5.0 2022-12-07 09:26:02,421 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4603, 4.0915, 4.1978, 4.5087, 4.2768, 3.6344, 4.5064, 4.3953], device='cuda:3'), covar=tensor([0.0738, 0.0542, 0.0619, 0.0589, 0.0614, 0.0576, 0.0704, 0.0667], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0081, 0.0101, 0.0097, 0.0109, 0.0076, 0.0106, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2022-12-07 09:26:04,802 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.185e+02 2.928e+02 3.762e+02 4.834e+02 9.442e+02, threshold=7.523e+02, percent-clipped=4.0 2022-12-07 09:26:14,943 INFO [train.py:873] (3/4) Epoch 3, batch 5100, loss[loss=0.1841, simple_loss=0.2013, pruned_loss=0.08349, over 14550.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.2123, pruned_loss=0.1036, over 1990093.72 frames. ], batch size: 43, lr: 2.28e-02, grad_scale: 8.0 2022-12-07 09:26:23,968 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.08 vs. limit=5.0 2022-12-07 09:26:49,120 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20265.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:26:56,642 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20274.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:27:30,744 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.615e+02 2.933e+02 3.750e+02 4.548e+02 8.499e+02, threshold=7.500e+02, percent-clipped=1.0 2022-12-07 09:27:37,617 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20322.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:27:39,974 INFO [train.py:873] (3/4) Epoch 3, batch 5200, loss[loss=0.3084, simple_loss=0.2613, pruned_loss=0.1778, over 8597.00 frames. ], tot_loss[loss=0.211, simple_loss=0.2129, pruned_loss=0.1045, over 1997562.40 frames. ], batch size: 100, lr: 2.28e-02, grad_scale: 8.0 2022-12-07 09:28:01,308 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.82 vs. limit=2.0 2022-12-07 09:28:12,759 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.71 vs. limit=2.0 2022-12-07 09:28:23,145 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20376.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:28:34,128 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20388.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:28:56,776 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.552e+02 2.824e+02 3.945e+02 4.935e+02 1.895e+03, threshold=7.890e+02, percent-clipped=6.0 2022-12-07 09:28:57,982 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.65 vs. limit=2.0 2022-12-07 09:28:59,773 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20418.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:29:00,179 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2022-12-07 09:29:06,135 INFO [train.py:873] (3/4) Epoch 3, batch 5300, loss[loss=0.2338, simple_loss=0.2264, pruned_loss=0.1206, over 10369.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.2125, pruned_loss=0.1042, over 1940180.94 frames. ], batch size: 100, lr: 2.27e-02, grad_scale: 4.0 2022-12-07 09:29:20,888 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7893, 2.2966, 2.8252, 1.7242, 2.0630, 2.3015, 1.2568, 2.5088], device='cuda:3'), covar=tensor([0.1569, 0.0643, 0.0516, 0.1891, 0.1330, 0.0816, 0.3971, 0.0696], device='cuda:3'), in_proj_covar=tensor([0.0063, 0.0065, 0.0065, 0.0073, 0.0085, 0.0061, 0.0127, 0.0067], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 09:29:23,382 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20445.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:29:26,891 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20449.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:29:43,743 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20468.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:29:52,972 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20479.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:30:23,817 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.141e+02 2.763e+02 3.648e+02 4.928e+02 1.065e+03, threshold=7.297e+02, percent-clipped=2.0 2022-12-07 09:30:25,190 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.91 vs. limit=2.0 2022-12-07 09:30:32,155 INFO [train.py:873] (3/4) Epoch 3, batch 5400, loss[loss=0.2216, simple_loss=0.2263, pruned_loss=0.1085, over 14264.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.2113, pruned_loss=0.1019, over 1999176.30 frames. ], batch size: 57, lr: 2.27e-02, grad_scale: 4.0 2022-12-07 09:30:58,872 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5642, 4.1079, 4.0401, 4.6006, 4.3902, 4.0266, 4.5938, 3.8458], device='cuda:3'), covar=tensor([0.0381, 0.1093, 0.0396, 0.0353, 0.0613, 0.0556, 0.0568, 0.0563], device='cuda:3'), in_proj_covar=tensor([0.0109, 0.0179, 0.0124, 0.0111, 0.0113, 0.0101, 0.0174, 0.0127], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 09:31:02,909 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20560.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:31:27,483 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20588.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:31:44,522 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.81 vs. limit=5.0 2022-12-07 09:31:50,018 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.464e+02 2.948e+02 3.775e+02 4.848e+02 6.830e+02, threshold=7.550e+02, percent-clipped=0.0 2022-12-07 09:31:59,002 INFO [train.py:873] (3/4) Epoch 3, batch 5500, loss[loss=0.2396, simple_loss=0.2337, pruned_loss=0.1227, over 14407.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2125, pruned_loss=0.1034, over 1955350.82 frames. ], batch size: 41, lr: 2.26e-02, grad_scale: 4.0 2022-12-07 09:32:19,563 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20649.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:32:25,506 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=3.38 vs. limit=2.0 2022-12-07 09:32:28,584 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9484, 1.5788, 4.5591, 4.2475, 4.1401, 4.5740, 4.3293, 4.6001], device='cuda:3'), covar=tensor([0.1163, 0.1365, 0.0063, 0.0102, 0.0097, 0.0070, 0.0078, 0.0072], device='cuda:3'), in_proj_covar=tensor([0.0137, 0.0146, 0.0085, 0.0118, 0.0097, 0.0106, 0.0078, 0.0085], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-07 09:32:35,026 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9447, 4.7279, 4.4492, 4.4311, 4.3514, 4.7745, 4.9320, 4.9097], device='cuda:3'), covar=tensor([0.0630, 0.0560, 0.1300, 0.2325, 0.0614, 0.0394, 0.1000, 0.0727], device='cuda:3'), in_proj_covar=tensor([0.0217, 0.0188, 0.0266, 0.0342, 0.0214, 0.0240, 0.0255, 0.0214], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 09:32:40,295 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.33 vs. limit=5.0 2022-12-07 09:32:42,496 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20676.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:33:15,378 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.510e+02 2.742e+02 3.691e+02 4.963e+02 1.196e+03, threshold=7.382e+02, percent-clipped=6.0 2022-12-07 09:33:22,783 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20724.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:33:23,570 INFO [train.py:873] (3/4) Epoch 3, batch 5600, loss[loss=0.1826, simple_loss=0.1969, pruned_loss=0.0842, over 14365.00 frames. ], tot_loss[loss=0.2092, simple_loss=0.2119, pruned_loss=0.1032, over 1969301.91 frames. ], batch size: 55, lr: 2.26e-02, grad_scale: 8.0 2022-12-07 09:33:29,410 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.7378, 5.1220, 5.0592, 5.6466, 5.3929, 4.6417, 5.6175, 4.7294], device='cuda:3'), covar=tensor([0.0223, 0.0718, 0.0184, 0.0292, 0.0460, 0.0254, 0.0444, 0.0363], device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0174, 0.0118, 0.0107, 0.0107, 0.0096, 0.0167, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 09:33:40,098 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20744.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:33:41,300 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20745.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:33:54,917 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1678, 0.8751, 1.0013, 1.0803, 1.0802, 1.0267, 0.9938, 0.9107], device='cuda:3'), covar=tensor([0.0831, 0.0886, 0.0853, 0.0724, 0.0617, 0.0335, 0.0210, 0.1184], device='cuda:3'), in_proj_covar=tensor([0.0011, 0.0011, 0.0012, 0.0011, 0.0011, 0.0013, 0.0012, 0.0016], device='cuda:3'), out_proj_covar=tensor([3.7205e-05, 3.6060e-05, 3.9820e-05, 3.5205e-05, 3.5944e-05, 3.8355e-05, 4.5786e-05, 5.1026e-05], device='cuda:3') 2022-12-07 09:34:00,992 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20768.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:34:06,447 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20774.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:34:07,673 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.95 vs. limit=5.0 2022-12-07 09:34:24,028 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20793.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:34:43,760 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.806e+02 3.037e+02 4.375e+02 5.410e+02 1.130e+03, threshold=8.749e+02, percent-clipped=7.0 2022-12-07 09:34:44,730 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20816.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:34:49,461 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6636, 3.5153, 3.8139, 3.3066, 3.5343, 3.5852, 1.2796, 3.4204], device='cuda:3'), covar=tensor([0.0174, 0.0244, 0.0337, 0.0350, 0.0321, 0.0349, 0.3155, 0.0248], device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0113, 0.0105, 0.0091, 0.0146, 0.0099, 0.0141, 0.0136], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002, 0.0003, 0.0003], device='cuda:3') 2022-12-07 09:34:53,317 INFO [train.py:873] (3/4) Epoch 3, batch 5700, loss[loss=0.2205, simple_loss=0.1898, pruned_loss=0.1256, over 1244.00 frames. ], tot_loss[loss=0.2106, simple_loss=0.2128, pruned_loss=0.1042, over 1995631.03 frames. ], batch size: 100, lr: 2.25e-02, grad_scale: 8.0 2022-12-07 09:35:14,764 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.00 vs. limit=2.0 2022-12-07 09:35:24,798 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20860.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:35:44,564 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.90 vs. limit=5.0 2022-12-07 09:36:06,591 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20908.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:36:12,610 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.426e+01 2.770e+02 3.515e+02 4.337e+02 1.206e+03, threshold=7.030e+02, percent-clipped=1.0 2022-12-07 09:36:21,396 INFO [train.py:873] (3/4) Epoch 3, batch 5800, loss[loss=0.1914, simple_loss=0.2009, pruned_loss=0.09093, over 13987.00 frames. ], tot_loss[loss=0.2112, simple_loss=0.2129, pruned_loss=0.1048, over 1972829.32 frames. ], batch size: 19, lr: 2.25e-02, grad_scale: 8.0 2022-12-07 09:36:37,643 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20944.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:36:39,701 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.6124, 4.2441, 4.2478, 4.6767, 4.3265, 3.4150, 4.7074, 4.6230], device='cuda:3'), covar=tensor([0.0704, 0.0613, 0.0669, 0.0555, 0.0849, 0.0742, 0.0797, 0.0715], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0083, 0.0100, 0.0099, 0.0109, 0.0074, 0.0105, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2022-12-07 09:37:39,555 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.364e+02 2.882e+02 3.467e+02 4.356e+02 1.149e+03, threshold=6.934e+02, percent-clipped=1.0 2022-12-07 09:37:47,926 INFO [train.py:873] (3/4) Epoch 3, batch 5900, loss[loss=0.194, simple_loss=0.2013, pruned_loss=0.09331, over 14260.00 frames. ], tot_loss[loss=0.209, simple_loss=0.2112, pruned_loss=0.1034, over 1979280.30 frames. ], batch size: 37, lr: 2.24e-02, grad_scale: 8.0 2022-12-07 09:37:48,434 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2022-12-07 09:37:50,148 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.9726, 0.7146, 1.0512, 0.9854, 0.9211, 0.6140, 1.3136, 1.1645], device='cuda:3'), covar=tensor([0.0337, 0.0358, 0.0145, 0.0772, 0.0433, 0.0158, 0.0154, 0.0155], device='cuda:3'), in_proj_covar=tensor([0.0016, 0.0021, 0.0020, 0.0016, 0.0018, 0.0023, 0.0018, 0.0017], device='cuda:3'), out_proj_covar=tensor([5.4052e-05, 6.1737e-05, 5.3935e-05, 5.3077e-05, 5.3867e-05, 6.3567e-05, 5.6732e-05, 5.1537e-05], device='cuda:3') 2022-12-07 09:38:04,809 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21044.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:38:11,876 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21052.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:38:29,593 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4889, 1.5456, 1.9114, 1.0189, 1.3287, 1.6040, 1.1793, 1.6109], device='cuda:3'), covar=tensor([0.0893, 0.1665, 0.0620, 0.2163, 0.2075, 0.0906, 0.3935, 0.0784], device='cuda:3'), in_proj_covar=tensor([0.0064, 0.0068, 0.0069, 0.0076, 0.0091, 0.0064, 0.0137, 0.0071], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 09:38:32,106 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21074.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:38:47,677 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21092.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:39:07,996 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21113.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:39:09,614 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.016e+02 3.189e+02 4.370e+02 6.428e+02 1.112e+03, threshold=8.740e+02, percent-clipped=21.0 2022-12-07 09:39:12,456 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21118.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:39:16,019 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21122.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:39:18,491 INFO [train.py:873] (3/4) Epoch 3, batch 6000, loss[loss=0.1754, simple_loss=0.1948, pruned_loss=0.07803, over 14253.00 frames. ], tot_loss[loss=0.2089, simple_loss=0.2111, pruned_loss=0.1033, over 1940857.61 frames. ], batch size: 60, lr: 2.24e-02, grad_scale: 8.0 2022-12-07 09:39:18,491 INFO [train.py:896] (3/4) Computing validation loss 2022-12-07 09:39:27,745 INFO [train.py:905] (3/4) Epoch 3, validation: loss=0.1295, simple_loss=0.172, pruned_loss=0.04354, over 857387.00 frames. 2022-12-07 09:39:27,746 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17841MB 2022-12-07 09:39:37,710 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.58 vs. limit=2.0 2022-12-07 09:39:54,533 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8629, 1.3829, 1.7982, 1.1531, 1.6039, 1.6658, 1.8587, 1.6436], device='cuda:3'), covar=tensor([0.0436, 0.2364, 0.1002, 0.2532, 0.0882, 0.0933, 0.0436, 0.1335], device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0225, 0.0104, 0.0134, 0.0091, 0.0089, 0.0076, 0.0108], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:3') 2022-12-07 09:40:02,102 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0021, 1.9275, 1.9337, 2.0976, 1.9532, 1.8075, 1.1393, 1.7849], device='cuda:3'), covar=tensor([0.0252, 0.0365, 0.0524, 0.0171, 0.0271, 0.0571, 0.1717, 0.0330], device='cuda:3'), in_proj_covar=tensor([0.0108, 0.0117, 0.0107, 0.0092, 0.0147, 0.0100, 0.0145, 0.0138], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002, 0.0003, 0.0003], device='cuda:3') 2022-12-07 09:40:16,050 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21179.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:40:47,318 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.278e+02 2.723e+02 3.536e+02 4.487e+02 9.124e+02, threshold=7.073e+02, percent-clipped=3.0 2022-12-07 09:40:56,102 INFO [train.py:873] (3/4) Epoch 3, batch 6100, loss[loss=0.2297, simple_loss=0.2217, pruned_loss=0.1188, over 13530.00 frames. ], tot_loss[loss=0.2083, simple_loss=0.2111, pruned_loss=0.1028, over 1925538.81 frames. ], batch size: 100, lr: 2.23e-02, grad_scale: 8.0 2022-12-07 09:41:04,978 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 2022-12-07 09:41:12,172 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21244.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:41:27,524 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9841, 3.7567, 3.4477, 3.6049, 3.7770, 3.7493, 3.9672, 3.8916], device='cuda:3'), covar=tensor([0.0652, 0.0574, 0.1398, 0.1992, 0.0642, 0.0664, 0.0781, 0.0731], device='cuda:3'), in_proj_covar=tensor([0.0223, 0.0191, 0.0269, 0.0349, 0.0220, 0.0253, 0.0265, 0.0208], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 09:41:46,136 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.07 vs. limit=2.0 2022-12-07 09:41:50,747 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1026, 3.7665, 3.6683, 4.0978, 4.0580, 3.6918, 4.1382, 3.5315], device='cuda:3'), covar=tensor([0.0436, 0.0987, 0.0347, 0.0432, 0.0594, 0.0973, 0.0483, 0.0497], device='cuda:3'), in_proj_covar=tensor([0.0115, 0.0189, 0.0124, 0.0115, 0.0117, 0.0103, 0.0174, 0.0130], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 09:41:54,619 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21292.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:41:57,662 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=8.78 vs. limit=5.0 2022-12-07 09:42:08,431 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5330, 1.6920, 3.5491, 2.5546, 3.4603, 1.7167, 2.8808, 3.3425], device='cuda:3'), covar=tensor([0.0389, 0.4667, 0.0337, 0.6353, 0.0208, 0.3748, 0.0934, 0.0234], device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0299, 0.0171, 0.0402, 0.0174, 0.0313, 0.0282, 0.0175], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0002, 0.0004, 0.0002, 0.0004, 0.0003, 0.0002], device='cuda:3') 2022-12-07 09:42:11,704 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21312.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:42:14,077 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.862e+02 3.459e+02 4.303e+02 5.209e+02 1.109e+03, threshold=8.606e+02, percent-clipped=4.0 2022-12-07 09:42:22,733 INFO [train.py:873] (3/4) Epoch 3, batch 6200, loss[loss=0.1963, simple_loss=0.1913, pruned_loss=0.1006, over 4968.00 frames. ], tot_loss[loss=0.2092, simple_loss=0.2116, pruned_loss=0.1034, over 1979946.56 frames. ], batch size: 100, lr: 2.23e-02, grad_scale: 8.0 2022-12-07 09:42:25,351 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21328.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:43:04,507 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21373.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:43:18,529 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21389.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:43:34,681 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21408.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:43:41,035 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.343e+02 2.777e+02 3.357e+02 4.403e+02 6.655e+02, threshold=6.715e+02, percent-clipped=0.0 2022-12-07 09:43:49,957 INFO [train.py:873] (3/4) Epoch 3, batch 6300, loss[loss=0.2348, simple_loss=0.2158, pruned_loss=0.1269, over 4929.00 frames. ], tot_loss[loss=0.2099, simple_loss=0.2122, pruned_loss=0.1038, over 1941110.06 frames. ], batch size: 100, lr: 2.22e-02, grad_scale: 8.0 2022-12-07 09:43:57,718 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.9980, 0.6900, 0.9232, 1.0578, 0.8590, 0.5191, 0.8964, 1.0810], device='cuda:3'), covar=tensor([0.0299, 0.0312, 0.0182, 0.0336, 0.0190, 0.0226, 0.0155, 0.0240], device='cuda:3'), in_proj_covar=tensor([0.0015, 0.0020, 0.0020, 0.0017, 0.0018, 0.0024, 0.0019, 0.0017], device='cuda:3'), out_proj_covar=tensor([5.3177e-05, 5.9728e-05, 5.4763e-05, 5.5813e-05, 5.3360e-05, 6.5483e-05, 6.0283e-05, 5.2476e-05], device='cuda:3') 2022-12-07 09:44:32,742 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21474.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:44:35,274 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7749, 2.6118, 2.0578, 2.3348, 1.7447, 2.4151, 2.5647, 0.9557], device='cuda:3'), covar=tensor([0.3846, 0.1226, 0.1874, 0.2040, 0.1181, 0.0515, 0.1272, 0.3784], device='cuda:3'), in_proj_covar=tensor([0.0159, 0.0063, 0.0050, 0.0057, 0.0068, 0.0057, 0.0076, 0.0093], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2022-12-07 09:45:07,700 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.209e+02 2.996e+02 3.929e+02 5.404e+02 9.451e+02, threshold=7.857e+02, percent-clipped=12.0 2022-12-07 09:45:16,310 INFO [train.py:873] (3/4) Epoch 3, batch 6400, loss[loss=0.2077, simple_loss=0.2139, pruned_loss=0.1007, over 14178.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.2111, pruned_loss=0.1027, over 1947283.07 frames. ], batch size: 84, lr: 2.22e-02, grad_scale: 8.0 2022-12-07 09:45:45,085 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0938, 3.1269, 2.5548, 2.6238, 1.8857, 2.7697, 2.9843, 1.1280], device='cuda:3'), covar=tensor([0.2891, 0.0661, 0.0996, 0.1496, 0.1114, 0.0574, 0.1357, 0.3315], device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0060, 0.0049, 0.0056, 0.0067, 0.0057, 0.0076, 0.0092], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2022-12-07 09:46:16,519 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21595.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:46:31,499 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8436, 2.7003, 2.8604, 2.7986, 2.7492, 2.6119, 1.3477, 2.5605], device='cuda:3'), covar=tensor([0.0178, 0.0319, 0.0348, 0.0255, 0.0263, 0.0487, 0.2037, 0.0273], device='cuda:3'), in_proj_covar=tensor([0.0110, 0.0118, 0.0113, 0.0094, 0.0144, 0.0103, 0.0144, 0.0138], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 09:46:32,445 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21613.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:46:34,034 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.437e+02 2.688e+02 3.501e+02 4.661e+02 7.945e+02, threshold=7.001e+02, percent-clipped=1.0 2022-12-07 09:46:43,050 INFO [train.py:873] (3/4) Epoch 3, batch 6500, loss[loss=0.1832, simple_loss=0.1641, pruned_loss=0.1012, over 2670.00 frames. ], tot_loss[loss=0.2089, simple_loss=0.2112, pruned_loss=0.1033, over 1916387.86 frames. ], batch size: 100, lr: 2.21e-02, grad_scale: 8.0 2022-12-07 09:46:56,747 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1518, 2.1580, 2.0699, 2.1096, 1.4943, 1.9254, 1.9094, 0.9696], device='cuda:3'), covar=tensor([0.2648, 0.1028, 0.1012, 0.0514, 0.1008, 0.0491, 0.1691, 0.2847], device='cuda:3'), in_proj_covar=tensor([0.0158, 0.0062, 0.0050, 0.0056, 0.0068, 0.0057, 0.0078, 0.0091], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2022-12-07 09:47:03,413 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4770, 1.6346, 1.9883, 1.1460, 1.3334, 1.7539, 1.2022, 1.7158], device='cuda:3'), covar=tensor([0.0640, 0.1709, 0.0324, 0.1708, 0.1372, 0.0542, 0.2536, 0.0674], device='cuda:3'), in_proj_covar=tensor([0.0059, 0.0068, 0.0067, 0.0074, 0.0086, 0.0060, 0.0131, 0.0071], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 09:47:09,433 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21656.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:47:19,586 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21668.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:47:24,839 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21674.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:47:33,133 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21684.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:47:54,491 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21708.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:48:00,153 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.571e+02 3.136e+02 4.193e+02 5.448e+02 1.293e+03, threshold=8.387e+02, percent-clipped=8.0 2022-12-07 09:48:08,228 INFO [train.py:873] (3/4) Epoch 3, batch 6600, loss[loss=0.2108, simple_loss=0.2123, pruned_loss=0.1047, over 13542.00 frames. ], tot_loss[loss=0.2083, simple_loss=0.2107, pruned_loss=0.103, over 1921515.26 frames. ], batch size: 100, lr: 2.21e-02, grad_scale: 8.0 2022-12-07 09:48:34,766 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21756.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:48:38,076 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.2029, 4.0368, 4.4101, 3.6856, 4.1864, 4.4720, 1.6403, 4.1109], device='cuda:3'), covar=tensor([0.0145, 0.0210, 0.0345, 0.0377, 0.0216, 0.0149, 0.2651, 0.0174], device='cuda:3'), in_proj_covar=tensor([0.0110, 0.0118, 0.0112, 0.0092, 0.0146, 0.0100, 0.0144, 0.0137], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002, 0.0003, 0.0003], device='cuda:3') 2022-12-07 09:48:40,825 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21763.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:48:49,859 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5853, 1.8792, 3.7295, 2.5610, 3.7511, 2.0051, 2.6090, 3.5332], device='cuda:3'), covar=tensor([0.0313, 0.4572, 0.0400, 0.7648, 0.0211, 0.3756, 0.1172, 0.0185], device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0288, 0.0173, 0.0388, 0.0171, 0.0306, 0.0266, 0.0169], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0002, 0.0004, 0.0002, 0.0004, 0.0003, 0.0002], device='cuda:3') 2022-12-07 09:48:51,195 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21774.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:49:15,184 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9767, 2.1690, 2.8054, 1.5808, 1.9568, 2.4457, 1.2303, 2.2035], device='cuda:3'), covar=tensor([0.0912, 0.0739, 0.0451, 0.1753, 0.1084, 0.0576, 0.3090, 0.0654], device='cuda:3'), in_proj_covar=tensor([0.0060, 0.0066, 0.0067, 0.0076, 0.0085, 0.0059, 0.0132, 0.0069], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 09:49:27,684 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.343e+02 2.849e+02 3.705e+02 5.166e+02 8.548e+02, threshold=7.409e+02, percent-clipped=1.0 2022-12-07 09:49:34,266 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21822.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:49:36,025 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21824.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:49:36,766 INFO [train.py:873] (3/4) Epoch 3, batch 6700, loss[loss=0.1654, simple_loss=0.1533, pruned_loss=0.08873, over 1249.00 frames. ], tot_loss[loss=0.2077, simple_loss=0.2108, pruned_loss=0.1023, over 1967205.94 frames. ], batch size: 100, lr: 2.20e-02, grad_scale: 8.0 2022-12-07 09:49:57,946 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21850.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:50:42,868 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8679, 2.7954, 1.8586, 3.0346, 2.7180, 2.9295, 2.5286, 2.1828], device='cuda:3'), covar=tensor([0.0283, 0.0737, 0.2437, 0.0152, 0.0338, 0.0342, 0.0930, 0.2512], device='cuda:3'), in_proj_covar=tensor([0.0214, 0.0285, 0.0331, 0.0178, 0.0228, 0.0217, 0.0265, 0.0327], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:3') 2022-12-07 09:50:45,918 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=3.38 vs. limit=2.0 2022-12-07 09:50:51,641 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21911.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:50:54,926 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.098e+02 2.679e+02 3.700e+02 5.034e+02 8.925e+02, threshold=7.401e+02, percent-clipped=4.0 2022-12-07 09:51:03,450 INFO [train.py:873] (3/4) Epoch 3, batch 6800, loss[loss=0.2348, simple_loss=0.2274, pruned_loss=0.1212, over 14593.00 frames. ], tot_loss[loss=0.2076, simple_loss=0.2105, pruned_loss=0.1024, over 1969852.35 frames. ], batch size: 23, lr: 2.20e-02, grad_scale: 8.0 2022-12-07 09:51:05,301 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8541, 1.9974, 4.4105, 2.3719, 4.3974, 4.5600, 4.2626, 5.0224], device='cuda:3'), covar=tensor([0.0100, 0.2045, 0.0199, 0.1491, 0.0173, 0.0156, 0.0197, 0.0081], device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0149, 0.0111, 0.0159, 0.0130, 0.0120, 0.0101, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-07 09:51:26,000 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21951.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:51:29,809 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2022-12-07 09:51:40,233 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21968.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:51:41,062 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21969.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:51:42,284 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21970.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:51:45,774 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.84 vs. limit=2.0 2022-12-07 09:51:54,846 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21984.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:52:07,443 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2022-12-07 09:52:21,335 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.545e+02 2.962e+02 3.808e+02 5.028e+02 7.344e+02, threshold=7.616e+02, percent-clipped=0.0 2022-12-07 09:52:22,245 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22016.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:52:29,727 INFO [train.py:873] (3/4) Epoch 3, batch 6900, loss[loss=0.2102, simple_loss=0.213, pruned_loss=0.1038, over 14403.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2104, pruned_loss=0.1028, over 1952794.87 frames. ], batch size: 53, lr: 2.19e-02, grad_scale: 8.0 2022-12-07 09:52:35,025 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22031.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:52:35,754 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22032.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:53:24,702 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1785, 1.4447, 3.7908, 1.4750, 3.8214, 4.0908, 3.1730, 4.4870], device='cuda:3'), covar=tensor([0.0145, 0.2775, 0.0313, 0.2412, 0.0297, 0.0257, 0.0569, 0.0097], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0152, 0.0113, 0.0164, 0.0134, 0.0122, 0.0105, 0.0104], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 09:53:46,970 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.783e+02 2.990e+02 4.011e+02 5.296e+02 1.371e+03, threshold=8.022e+02, percent-clipped=6.0 2022-12-07 09:53:50,231 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22119.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:53:55,141 INFO [train.py:873] (3/4) Epoch 3, batch 7000, loss[loss=0.2366, simple_loss=0.2231, pruned_loss=0.1251, over 8569.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2105, pruned_loss=0.1027, over 1925314.64 frames. ], batch size: 100, lr: 2.19e-02, grad_scale: 8.0 2022-12-07 09:53:59,830 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22130.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:54:13,864 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8130, 1.5244, 2.0932, 1.7014, 2.1705, 1.8549, 1.7953, 1.9062], device='cuda:3'), covar=tensor([0.0094, 0.0349, 0.0049, 0.0093, 0.0055, 0.0085, 0.0058, 0.0154], device='cuda:3'), in_proj_covar=tensor([0.0224, 0.0336, 0.0230, 0.0294, 0.0259, 0.0237, 0.0262, 0.0353], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-07 09:54:39,474 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22176.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:54:52,514 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22191.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:55:05,092 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22206.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:55:06,091 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4495, 2.8662, 4.1763, 2.8250, 4.0996, 4.0950, 3.5229, 3.2800], device='cuda:3'), covar=tensor([0.0098, 0.1511, 0.0189, 0.0932, 0.0282, 0.0251, 0.1150, 0.1465], device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0349, 0.0241, 0.0302, 0.0268, 0.0246, 0.0270, 0.0370], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-07 09:55:11,357 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.2906, 4.8579, 4.7965, 5.2942, 5.0833, 4.5236, 5.3550, 4.4212], device='cuda:3'), covar=tensor([0.0259, 0.0839, 0.0236, 0.0381, 0.0530, 0.0307, 0.0350, 0.0432], device='cuda:3'), in_proj_covar=tensor([0.0115, 0.0187, 0.0123, 0.0114, 0.0121, 0.0101, 0.0175, 0.0128], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 09:55:12,926 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.440e+02 2.681e+02 3.418e+02 4.673e+02 9.486e+02, threshold=6.836e+02, percent-clipped=4.0 2022-12-07 09:55:13,439 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22215.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:55:21,728 INFO [train.py:873] (3/4) Epoch 3, batch 7100, loss[loss=0.172, simple_loss=0.1512, pruned_loss=0.09635, over 1271.00 frames. ], tot_loss[loss=0.2076, simple_loss=0.2104, pruned_loss=0.1024, over 1938677.28 frames. ], batch size: 100, lr: 2.18e-02, grad_scale: 8.0 2022-12-07 09:55:29,838 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9446, 2.3812, 4.9090, 3.2300, 4.8288, 2.4313, 3.2741, 4.6702], device='cuda:3'), covar=tensor([0.0306, 0.4852, 0.0362, 0.9324, 0.0114, 0.3817, 0.1154, 0.0131], device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0294, 0.0180, 0.0400, 0.0177, 0.0320, 0.0279, 0.0170], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0002, 0.0005, 0.0002, 0.0004, 0.0003, 0.0002], device='cuda:3') 2022-12-07 09:55:32,448 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22237.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:55:32,469 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22237.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:55:44,104 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22251.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:56:00,020 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22269.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:56:06,012 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22276.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:56:24,833 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22298.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:56:25,593 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22299.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:56:38,993 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.756e+01 2.795e+02 3.480e+02 4.240e+02 8.610e+02, threshold=6.960e+02, percent-clipped=4.0 2022-12-07 09:56:40,761 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22317.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:56:47,594 INFO [train.py:873] (3/4) Epoch 3, batch 7200, loss[loss=0.2181, simple_loss=0.2153, pruned_loss=0.1104, over 14163.00 frames. ], tot_loss[loss=0.2054, simple_loss=0.209, pruned_loss=0.101, over 1967094.31 frames. ], batch size: 99, lr: 2.18e-02, grad_scale: 8.0 2022-12-07 09:56:48,445 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22326.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:56:56,148 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=7.02 vs. limit=5.0 2022-12-07 09:56:57,120 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.6227, 0.6915, 0.2177, 0.7838, 0.7489, 0.0942, 0.6171, 0.6871], device='cuda:3'), covar=tensor([0.0080, 0.0027, 0.0015, 0.0045, 0.0026, 0.0020, 0.0087, 0.0027], device='cuda:3'), in_proj_covar=tensor([0.0016, 0.0018, 0.0019, 0.0015, 0.0016, 0.0022, 0.0018, 0.0016], device='cuda:3'), out_proj_covar=tensor([5.5010e-05, 5.5969e-05, 5.3968e-05, 5.1798e-05, 5.0537e-05, 6.1622e-05, 5.8768e-05, 5.0590e-05], device='cuda:3') 2022-12-07 09:57:10,949 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.5911, 5.1419, 5.1893, 5.8116, 5.3189, 4.4934, 5.8001, 5.7399], device='cuda:3'), covar=tensor([0.0709, 0.0537, 0.0531, 0.0562, 0.0580, 0.0390, 0.0547, 0.0565], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0079, 0.0097, 0.0098, 0.0108, 0.0074, 0.0104, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2022-12-07 09:57:22,340 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.81 vs. limit=5.0 2022-12-07 09:57:38,929 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.63 vs. limit=5.0 2022-12-07 09:57:39,313 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22385.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:57:43,136 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2022-12-07 09:58:05,272 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.487e+02 2.951e+02 3.875e+02 4.878e+02 1.117e+03, threshold=7.750e+02, percent-clipped=7.0 2022-12-07 09:58:09,162 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22419.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:58:13,857 INFO [train.py:873] (3/4) Epoch 3, batch 7300, loss[loss=0.2043, simple_loss=0.1808, pruned_loss=0.1139, over 2658.00 frames. ], tot_loss[loss=0.2045, simple_loss=0.2085, pruned_loss=0.1003, over 1988524.33 frames. ], batch size: 100, lr: 2.17e-02, grad_scale: 16.0 2022-12-07 09:58:18,546 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=3.34 vs. limit=2.0 2022-12-07 09:58:31,545 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22446.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:58:49,760 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22467.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:59:06,176 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22486.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:59:06,508 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.50 vs. limit=5.0 2022-12-07 09:59:20,461 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6217, 1.4968, 1.8878, 2.0605, 1.5036, 1.5759, 1.8825, 1.8325], device='cuda:3'), covar=tensor([0.0030, 0.0056, 0.0024, 0.0013, 0.0037, 0.0077, 0.0021, 0.0024], device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0158, 0.0206, 0.0173, 0.0145, 0.0201, 0.0111, 0.0193], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0001, 0.0002], device='cuda:3') 2022-12-07 09:59:23,424 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22506.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:59:31,053 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.364e+02 2.878e+02 3.644e+02 4.636e+02 1.010e+03, threshold=7.288e+02, percent-clipped=2.0 2022-12-07 09:59:39,445 INFO [train.py:873] (3/4) Epoch 3, batch 7400, loss[loss=0.2587, simple_loss=0.239, pruned_loss=0.1392, over 9485.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.2079, pruned_loss=0.09959, over 1972339.32 frames. ], batch size: 100, lr: 2.17e-02, grad_scale: 16.0 2022-12-07 09:59:45,736 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22532.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 09:59:53,895 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8427, 0.6333, 0.8569, 0.8532, 1.1560, 0.4384, 1.0224, 1.0809], device='cuda:3'), covar=tensor([0.0364, 0.0588, 0.0164, 0.0904, 0.0369, 0.0221, 0.0336, 0.0458], device='cuda:3'), in_proj_covar=tensor([0.0015, 0.0018, 0.0019, 0.0016, 0.0016, 0.0022, 0.0017, 0.0017], device='cuda:3'), out_proj_covar=tensor([5.3920e-05, 5.5950e-05, 5.4280e-05, 5.4459e-05, 4.9741e-05, 6.3233e-05, 5.7416e-05, 5.2540e-05], device='cuda:3') 2022-12-07 09:59:55,686 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.7214, 1.8312, 4.4783, 1.6481, 4.2662, 4.3683, 4.1698, 4.8877], device='cuda:3'), covar=tensor([0.0119, 0.2316, 0.0190, 0.2125, 0.0191, 0.0226, 0.0210, 0.0105], device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0145, 0.0111, 0.0157, 0.0129, 0.0120, 0.0101, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-07 10:00:05,237 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22554.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:00:10,193 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4403, 2.1089, 4.9677, 4.4188, 4.4336, 4.9100, 4.7694, 5.0015], device='cuda:3'), covar=tensor([0.1059, 0.1155, 0.0056, 0.0107, 0.0108, 0.0081, 0.0053, 0.0066], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0145, 0.0084, 0.0117, 0.0098, 0.0101, 0.0077, 0.0083], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 10:00:19,734 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22571.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:00:38,841 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22593.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:00:58,627 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.482e+02 2.799e+02 3.655e+02 4.825e+02 8.691e+02, threshold=7.309e+02, percent-clipped=2.0 2022-12-07 10:01:06,423 INFO [train.py:873] (3/4) Epoch 3, batch 7500, loss[loss=0.2125, simple_loss=0.2125, pruned_loss=0.1063, over 14330.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.2078, pruned_loss=0.09927, over 1953709.70 frames. ], batch size: 44, lr: 2.16e-02, grad_scale: 8.0 2022-12-07 10:01:07,392 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22626.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:01:17,949 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22639.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:01:21,350 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.6659, 1.9997, 2.7118, 2.3621, 2.8366, 2.6213, 2.5187, 2.3119], device='cuda:3'), covar=tensor([0.0141, 0.1249, 0.0198, 0.0545, 0.0207, 0.0291, 0.0283, 0.0888], device='cuda:3'), in_proj_covar=tensor([0.0234, 0.0350, 0.0249, 0.0302, 0.0271, 0.0248, 0.0271, 0.0366], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-07 10:01:44,222 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22674.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:02:33,693 INFO [train.py:873] (3/4) Epoch 4, batch 0, loss[loss=0.2805, simple_loss=0.2576, pruned_loss=0.1517, over 9511.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.2576, pruned_loss=0.1517, over 9511.00 frames. ], batch size: 100, lr: 2.02e-02, grad_scale: 8.0 2022-12-07 10:02:33,694 INFO [train.py:896] (3/4) Computing validation loss 2022-12-07 10:02:40,698 INFO [train.py:905] (3/4) Epoch 4, validation: loss=0.1426, simple_loss=0.185, pruned_loss=0.0501, over 857387.00 frames. 2022-12-07 10:02:40,699 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17841MB 2022-12-07 10:02:52,907 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22700.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 10:03:06,590 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 5.563e+01 2.095e+02 3.755e+02 5.061e+02 1.365e+03, threshold=7.510e+02, percent-clipped=11.0 2022-12-07 10:03:20,096 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2840, 3.1190, 2.9364, 3.3940, 2.8656, 2.6522, 3.3145, 3.3317], device='cuda:3'), covar=tensor([0.0696, 0.0644, 0.0754, 0.0620, 0.0852, 0.0736, 0.0648, 0.0708], device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0081, 0.0098, 0.0098, 0.0107, 0.0074, 0.0103, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2022-12-07 10:03:28,849 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22741.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:03:52,740 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5587, 3.1338, 2.2893, 3.4673, 3.3316, 3.4697, 3.1356, 2.2730], device='cuda:3'), covar=tensor([0.0364, 0.0951, 0.3292, 0.0232, 0.0365, 0.0533, 0.0826, 0.3397], device='cuda:3'), in_proj_covar=tensor([0.0224, 0.0290, 0.0335, 0.0185, 0.0230, 0.0224, 0.0259, 0.0331], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:3') 2022-12-07 10:04:08,189 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22786.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:04:08,914 INFO [train.py:873] (3/4) Epoch 4, batch 100, loss[loss=0.1597, simple_loss=0.1788, pruned_loss=0.0703, over 13963.00 frames. ], tot_loss[loss=0.205, simple_loss=0.2093, pruned_loss=0.1004, over 832986.50 frames. ], batch size: 19, lr: 2.02e-02, grad_scale: 8.0 2022-12-07 10:04:15,019 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22794.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:04:33,903 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.273e+02 2.903e+02 3.464e+02 4.472e+02 7.664e+02, threshold=6.927e+02, percent-clipped=2.0 2022-12-07 10:04:41,877 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22825.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:04:48,111 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22832.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:04:49,753 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22834.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:05:08,076 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22855.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:05:22,202 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22871.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:05:28,032 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.3533, 4.7118, 4.9968, 5.3219, 4.8775, 4.4019, 5.2845, 4.5519], device='cuda:3'), covar=tensor([0.0279, 0.0913, 0.0209, 0.0369, 0.0708, 0.0380, 0.0421, 0.0451], device='cuda:3'), in_proj_covar=tensor([0.0113, 0.0185, 0.0121, 0.0115, 0.0119, 0.0100, 0.0174, 0.0125], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 10:05:29,640 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22880.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:05:34,953 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22886.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:05:35,728 INFO [train.py:873] (3/4) Epoch 4, batch 200, loss[loss=0.1896, simple_loss=0.1856, pruned_loss=0.09683, over 3862.00 frames. ], tot_loss[loss=0.2042, simple_loss=0.2084, pruned_loss=0.09999, over 1282699.87 frames. ], batch size: 100, lr: 2.01e-02, grad_scale: 8.0 2022-12-07 10:05:40,927 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22893.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:06:00,925 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.239e+02 2.534e+02 3.313e+02 4.252e+02 7.669e+02, threshold=6.626e+02, percent-clipped=3.0 2022-12-07 10:06:03,423 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22919.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:06:15,182 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0848, 1.6385, 2.4294, 2.0087, 2.3321, 1.7567, 1.8376, 2.1404], device='cuda:3'), covar=tensor([0.1133, 0.1943, 0.0219, 0.2319, 0.0160, 0.1718, 0.1459, 0.0336], device='cuda:3'), in_proj_covar=tensor([0.0243, 0.0298, 0.0176, 0.0406, 0.0175, 0.0322, 0.0293, 0.0182], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0002, 0.0005, 0.0002, 0.0004, 0.0004, 0.0002], device='cuda:3') 2022-12-07 10:06:22,000 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22941.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:07:01,608 INFO [train.py:873] (3/4) Epoch 4, batch 300, loss[loss=0.2237, simple_loss=0.2282, pruned_loss=0.1096, over 14230.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.2069, pruned_loss=0.0988, over 1511755.61 frames. ], batch size: 76, lr: 2.01e-02, grad_scale: 8.0 2022-12-07 10:07:07,397 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.62 vs. limit=2.0 2022-12-07 10:07:08,443 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22995.0, num_to_drop=1, layers_to_drop={3} 2022-12-07 10:07:12,896 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0657, 2.8950, 2.8008, 3.1619, 2.7611, 2.5811, 3.0921, 3.0722], device='cuda:3'), covar=tensor([0.0646, 0.0693, 0.0689, 0.0566, 0.0837, 0.0698, 0.0681, 0.0716], device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0084, 0.0098, 0.0100, 0.0107, 0.0075, 0.0105, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2022-12-07 10:07:26,196 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.872e+01 2.724e+02 3.752e+02 4.773e+02 9.337e+02, threshold=7.504e+02, percent-clipped=9.0 2022-12-07 10:07:46,316 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2022-12-07 10:07:47,359 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2176, 2.1562, 2.0407, 2.2890, 1.9213, 1.9751, 2.2239, 2.2677], device='cuda:3'), covar=tensor([0.0729, 0.0778, 0.0804, 0.0600, 0.0956, 0.0684, 0.0837, 0.0703], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0084, 0.0099, 0.0101, 0.0108, 0.0074, 0.0106, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2022-12-07 10:07:48,189 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23041.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:08:17,892 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.98 vs. limit=2.0 2022-12-07 10:08:27,272 INFO [train.py:873] (3/4) Epoch 4, batch 400, loss[loss=0.1937, simple_loss=0.1879, pruned_loss=0.09975, over 3876.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2069, pruned_loss=0.09816, over 1736498.90 frames. ], batch size: 100, lr: 2.00e-02, grad_scale: 8.0 2022-12-07 10:08:29,409 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23089.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:08:46,286 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23108.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:08:52,656 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.082e+02 2.809e+02 3.789e+02 4.581e+02 1.009e+03, threshold=7.577e+02, percent-clipped=4.0 2022-12-07 10:09:22,122 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23150.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:09:27,955 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3249, 1.3959, 2.5398, 1.3446, 2.4149, 2.4390, 1.6360, 2.5523], device='cuda:3'), covar=tensor([0.0157, 0.1503, 0.0167, 0.1376, 0.0238, 0.0303, 0.0899, 0.0151], device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0150, 0.0117, 0.0161, 0.0132, 0.0125, 0.0109, 0.0105], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 10:09:32,718 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2022-12-07 10:09:33,963 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1201, 1.8804, 2.0289, 2.1147, 2.0656, 2.0296, 2.1892, 1.8339], device='cuda:3'), covar=tensor([0.0675, 0.1314, 0.0575, 0.0636, 0.0829, 0.0710, 0.0879, 0.0773], device='cuda:3'), in_proj_covar=tensor([0.0117, 0.0190, 0.0126, 0.0118, 0.0123, 0.0102, 0.0183, 0.0129], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 10:09:38,616 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23169.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:09:48,740 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23181.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:09:54,464 INFO [train.py:873] (3/4) Epoch 4, batch 500, loss[loss=0.2488, simple_loss=0.2436, pruned_loss=0.127, over 14088.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2066, pruned_loss=0.09842, over 1828395.24 frames. ], batch size: 29, lr: 2.00e-02, grad_scale: 8.0 2022-12-07 10:10:19,852 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.033e+01 2.776e+02 3.736e+02 4.463e+02 9.726e+02, threshold=7.473e+02, percent-clipped=4.0 2022-12-07 10:10:20,802 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23217.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:10:31,921 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23229.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:11:05,713 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9138, 1.9521, 1.8069, 2.0672, 1.6312, 1.8131, 1.9636, 2.0306], device='cuda:3'), covar=tensor([0.1096, 0.0786, 0.0966, 0.0876, 0.1233, 0.0863, 0.1192, 0.0925], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0082, 0.0101, 0.0098, 0.0107, 0.0075, 0.0106, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2022-12-07 10:11:13,518 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23278.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:11:20,903 INFO [train.py:873] (3/4) Epoch 4, batch 600, loss[loss=0.2193, simple_loss=0.184, pruned_loss=0.1273, over 2653.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.207, pruned_loss=0.09895, over 1843265.09 frames. ], batch size: 100, lr: 2.00e-02, grad_scale: 8.0 2022-12-07 10:11:23,692 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23290.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:11:27,759 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23295.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:11:28,513 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5796, 3.3905, 3.2927, 3.7120, 3.2108, 2.9195, 3.6465, 3.5946], device='cuda:3'), covar=tensor([0.0720, 0.0557, 0.0665, 0.0591, 0.0842, 0.0673, 0.0642, 0.0698], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0082, 0.0102, 0.0099, 0.0109, 0.0075, 0.0108, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2022-12-07 10:11:45,413 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.072e+02 2.720e+02 3.786e+02 4.640e+02 1.060e+03, threshold=7.571e+02, percent-clipped=5.0 2022-12-07 10:12:08,388 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23343.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:12:46,509 INFO [train.py:873] (3/4) Epoch 4, batch 700, loss[loss=0.1978, simple_loss=0.2064, pruned_loss=0.09463, over 10376.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.206, pruned_loss=0.09813, over 1891935.98 frames. ], batch size: 100, lr: 1.99e-02, grad_scale: 8.0 2022-12-07 10:13:08,891 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3510, 1.8608, 2.5780, 2.1765, 2.6250, 2.3341, 2.1532, 2.1419], device='cuda:3'), covar=tensor([0.0127, 0.1048, 0.0241, 0.0535, 0.0177, 0.0209, 0.0241, 0.0618], device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0354, 0.0260, 0.0310, 0.0288, 0.0250, 0.0283, 0.0375], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-07 10:13:11,266 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.426e+02 2.855e+02 3.544e+02 4.274e+02 7.481e+02, threshold=7.089e+02, percent-clipped=0.0 2022-12-07 10:13:32,232 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.7796, 4.5542, 4.2334, 4.2811, 4.3570, 4.5751, 4.7591, 4.7247], device='cuda:3'), covar=tensor([0.0675, 0.0514, 0.1593, 0.2234, 0.0655, 0.0536, 0.0779, 0.0778], device='cuda:3'), in_proj_covar=tensor([0.0239, 0.0204, 0.0297, 0.0381, 0.0224, 0.0268, 0.0273, 0.0220], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2022-12-07 10:13:40,549 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23450.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:13:49,978 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 2022-12-07 10:13:52,733 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23464.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:14:07,373 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23481.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:14:12,390 INFO [train.py:873] (3/4) Epoch 4, batch 800, loss[loss=0.1855, simple_loss=0.2094, pruned_loss=0.08076, over 14098.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2059, pruned_loss=0.09804, over 1928812.56 frames. ], batch size: 29, lr: 1.99e-02, grad_scale: 8.0 2022-12-07 10:14:13,325 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23488.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:14:22,032 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23498.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:14:37,484 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.307e+02 3.074e+02 3.598e+02 4.736e+02 1.088e+03, threshold=7.196e+02, percent-clipped=3.0 2022-12-07 10:14:48,557 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23529.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:14:49,032 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=12.06 vs. limit=5.0 2022-12-07 10:14:58,830 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.6943, 2.7432, 3.4108, 2.3689, 2.3352, 2.6814, 1.1119, 2.7673], device='cuda:3'), covar=tensor([0.1128, 0.0779, 0.0574, 0.1766, 0.1497, 0.1321, 0.5840, 0.0848], device='cuda:3'), in_proj_covar=tensor([0.0068, 0.0074, 0.0070, 0.0083, 0.0093, 0.0069, 0.0146, 0.0077], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 10:15:02,823 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.91 vs. limit=2.0 2022-12-07 10:15:05,953 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23549.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 10:15:12,850 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8560, 4.4201, 4.3162, 4.8535, 4.6443, 4.2497, 4.8322, 4.0302], device='cuda:3'), covar=tensor([0.0280, 0.0814, 0.0298, 0.0323, 0.0604, 0.0376, 0.0405, 0.0438], device='cuda:3'), in_proj_covar=tensor([0.0115, 0.0184, 0.0121, 0.0114, 0.0122, 0.0098, 0.0177, 0.0125], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 10:15:17,846 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.2246, 1.2785, 1.4828, 0.9224, 1.0113, 1.3938, 1.0200, 1.3338], device='cuda:3'), covar=tensor([0.0886, 0.2121, 0.0380, 0.1714, 0.2122, 0.0531, 0.1892, 0.0648], device='cuda:3'), in_proj_covar=tensor([0.0066, 0.0072, 0.0068, 0.0081, 0.0090, 0.0066, 0.0142, 0.0075], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 10:15:26,545 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23573.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:15:36,759 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23585.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:15:37,683 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1733, 1.0496, 0.9539, 0.8889, 0.6506, 0.9773, 1.0058, 0.5618], device='cuda:3'), covar=tensor([0.0475, 0.0309, 0.0527, 0.0173, 0.0653, 0.0303, 0.0303, 0.1003], device='cuda:3'), in_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0010, 0.0011, 0.0013, 0.0013, 0.0016], device='cuda:3'), out_proj_covar=tensor([3.9704e-05, 3.6809e-05, 4.4969e-05, 3.7260e-05, 3.9416e-05, 4.4668e-05, 5.4325e-05, 5.7003e-05], device='cuda:3') 2022-12-07 10:15:38,303 INFO [train.py:873] (3/4) Epoch 4, batch 900, loss[loss=0.1885, simple_loss=0.2005, pruned_loss=0.08821, over 6949.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.2059, pruned_loss=0.09923, over 1791832.86 frames. ], batch size: 100, lr: 1.98e-02, grad_scale: 8.0 2022-12-07 10:15:49,319 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.08 vs. limit=2.0 2022-12-07 10:15:58,155 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.47 vs. limit=5.0 2022-12-07 10:16:03,091 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.303e+02 2.960e+02 3.909e+02 5.159e+02 1.247e+03, threshold=7.817e+02, percent-clipped=4.0 2022-12-07 10:16:31,828 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.6483, 4.3778, 4.9295, 3.9577, 4.6164, 4.9215, 1.6155, 4.4299], device='cuda:3'), covar=tensor([0.0120, 0.0196, 0.0222, 0.0349, 0.0211, 0.0089, 0.2816, 0.0182], device='cuda:3'), in_proj_covar=tensor([0.0113, 0.0120, 0.0114, 0.0094, 0.0152, 0.0106, 0.0144, 0.0143], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 10:17:03,224 INFO [train.py:873] (3/4) Epoch 4, batch 1000, loss[loss=0.1973, simple_loss=0.2068, pruned_loss=0.09387, over 14257.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2061, pruned_loss=0.09868, over 1843580.60 frames. ], batch size: 76, lr: 1.98e-02, grad_scale: 8.0 2022-12-07 10:17:15,877 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.35 vs. limit=5.0 2022-12-07 10:17:28,303 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.149e+02 2.746e+02 3.575e+02 4.376e+02 8.279e+02, threshold=7.149e+02, percent-clipped=2.0 2022-12-07 10:17:41,297 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5649, 2.4713, 3.3138, 2.4392, 3.3194, 3.4164, 3.1424, 2.8142], device='cuda:3'), covar=tensor([0.0146, 0.1385, 0.0256, 0.1067, 0.0333, 0.0247, 0.0553, 0.1244], device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0346, 0.0260, 0.0304, 0.0286, 0.0243, 0.0276, 0.0367], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-07 10:18:09,515 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23764.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:18:11,840 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8242, 4.6184, 4.4765, 4.8984, 4.4744, 4.1527, 5.0131, 4.7240], device='cuda:3'), covar=tensor([0.0597, 0.0491, 0.0532, 0.0567, 0.0584, 0.0408, 0.0432, 0.0689], device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0084, 0.0106, 0.0103, 0.0112, 0.0077, 0.0111, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2022-12-07 10:18:28,812 INFO [train.py:873] (3/4) Epoch 4, batch 1100, loss[loss=0.1997, simple_loss=0.1748, pruned_loss=0.1123, over 1184.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2056, pruned_loss=0.09733, over 1919259.12 frames. ], batch size: 100, lr: 1.98e-02, grad_scale: 8.0 2022-12-07 10:18:47,663 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2022-12-07 10:18:50,542 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23812.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:18:53,787 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.712e+02 2.927e+02 4.016e+02 5.307e+02 1.051e+03, threshold=8.031e+02, percent-clipped=8.0 2022-12-07 10:19:15,061 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8713, 1.9171, 2.5040, 1.6098, 1.7070, 2.2480, 1.2249, 1.9672], device='cuda:3'), covar=tensor([0.2068, 0.1443, 0.0669, 0.2680, 0.2778, 0.1386, 0.5672, 0.1591], device='cuda:3'), in_proj_covar=tensor([0.0065, 0.0072, 0.0065, 0.0080, 0.0093, 0.0066, 0.0139, 0.0076], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 10:19:18,365 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23844.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 10:19:25,313 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2022-12-07 10:19:43,526 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23873.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:19:53,189 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0674, 1.5314, 3.9964, 3.9623, 3.8841, 4.0854, 3.6511, 4.1979], device='cuda:3'), covar=tensor([0.1012, 0.1257, 0.0080, 0.0089, 0.0114, 0.0073, 0.0112, 0.0069], device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0146, 0.0088, 0.0120, 0.0102, 0.0106, 0.0076, 0.0086], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-07 10:19:53,967 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23885.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:19:55,486 INFO [train.py:873] (3/4) Epoch 4, batch 1200, loss[loss=0.2076, simple_loss=0.2075, pruned_loss=0.1039, over 14497.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.2044, pruned_loss=0.09558, over 1965277.05 frames. ], batch size: 49, lr: 1.97e-02, grad_scale: 8.0 2022-12-07 10:20:21,446 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.594e+02 2.900e+02 3.609e+02 4.734e+02 8.045e+02, threshold=7.219e+02, percent-clipped=1.0 2022-12-07 10:20:24,887 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23921.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:20:34,961 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23933.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:20:59,306 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7813, 2.1234, 2.8105, 2.3459, 2.7794, 2.6383, 2.6405, 2.3708], device='cuda:3'), covar=tensor([0.0116, 0.1005, 0.0226, 0.0579, 0.0199, 0.0227, 0.0298, 0.0830], device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0341, 0.0255, 0.0296, 0.0281, 0.0240, 0.0275, 0.0362], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-07 10:21:08,908 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.25 vs. limit=5.0 2022-12-07 10:21:20,674 INFO [train.py:873] (3/4) Epoch 4, batch 1300, loss[loss=0.2424, simple_loss=0.2153, pruned_loss=0.1348, over 4993.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2046, pruned_loss=0.09681, over 1940656.17 frames. ], batch size: 100, lr: 1.97e-02, grad_scale: 4.0 2022-12-07 10:21:28,208 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7936, 1.8641, 2.8298, 2.7631, 2.7799, 2.0017, 2.8670, 2.2868], device='cuda:3'), covar=tensor([0.0055, 0.0169, 0.0124, 0.0074, 0.0051, 0.0264, 0.0039, 0.0183], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0169, 0.0226, 0.0188, 0.0153, 0.0211, 0.0123, 0.0205], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:3') 2022-12-07 10:21:46,583 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.354e+02 2.737e+02 3.421e+02 4.208e+02 7.387e+02, threshold=6.842e+02, percent-clipped=1.0 2022-12-07 10:22:04,344 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24037.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:22:22,800 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2959, 3.3171, 3.4478, 3.0455, 3.2795, 2.9243, 1.3422, 3.2581], device='cuda:3'), covar=tensor([0.0185, 0.0228, 0.0323, 0.0314, 0.0251, 0.0529, 0.2482, 0.0207], device='cuda:3'), in_proj_covar=tensor([0.0115, 0.0123, 0.0115, 0.0096, 0.0153, 0.0108, 0.0143, 0.0142], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 10:22:46,182 INFO [train.py:873] (3/4) Epoch 4, batch 1400, loss[loss=0.2279, simple_loss=0.22, pruned_loss=0.1179, over 9522.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2061, pruned_loss=0.09786, over 1986343.43 frames. ], batch size: 100, lr: 1.96e-02, grad_scale: 4.0 2022-12-07 10:22:55,445 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24098.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:23:12,193 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.457e+02 2.769e+02 3.319e+02 4.343e+02 7.993e+02, threshold=6.638e+02, percent-clipped=1.0 2022-12-07 10:23:18,785 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.15 vs. limit=2.0 2022-12-07 10:23:35,092 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=24144.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 10:23:53,457 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8513, 0.7756, 0.6045, 1.0133, 0.7591, 0.4090, 0.6678, 0.8537], device='cuda:3'), covar=tensor([0.0128, 0.0153, 0.0040, 0.0133, 0.0072, 0.0066, 0.0469, 0.0176], device='cuda:3'), in_proj_covar=tensor([0.0014, 0.0018, 0.0019, 0.0015, 0.0016, 0.0021, 0.0019, 0.0016], device='cuda:3'), out_proj_covar=tensor([5.3644e-05, 5.9654e-05, 5.6227e-05, 5.5631e-05, 5.3889e-05, 6.5957e-05, 6.4192e-05, 5.3096e-05], device='cuda:3') 2022-12-07 10:23:56,050 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2684, 3.2138, 3.9749, 2.8273, 2.6253, 3.2067, 1.6288, 3.0244], device='cuda:3'), covar=tensor([0.0979, 0.0584, 0.0602, 0.1526, 0.1421, 0.1074, 0.4252, 0.1144], device='cuda:3'), in_proj_covar=tensor([0.0064, 0.0071, 0.0065, 0.0077, 0.0090, 0.0063, 0.0135, 0.0073], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 10:24:08,775 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.5574, 5.6070, 5.2930, 5.7381, 5.4912, 4.8601, 5.6929, 5.6438], device='cuda:3'), covar=tensor([0.0711, 0.0296, 0.0418, 0.0441, 0.0396, 0.0287, 0.0536, 0.0517], device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0083, 0.0103, 0.0100, 0.0109, 0.0075, 0.0109, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2022-12-07 10:24:11,899 INFO [train.py:873] (3/4) Epoch 4, batch 1500, loss[loss=0.1609, simple_loss=0.14, pruned_loss=0.09085, over 1342.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2054, pruned_loss=0.09699, over 1975317.54 frames. ], batch size: 100, lr: 1.96e-02, grad_scale: 4.0 2022-12-07 10:24:16,765 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=24192.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:24:38,553 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.592e+02 2.634e+02 3.340e+02 4.470e+02 8.988e+02, threshold=6.680e+02, percent-clipped=4.0 2022-12-07 10:24:44,783 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7364, 2.6229, 2.2487, 2.3631, 2.6264, 2.6034, 2.6922, 2.6650], device='cuda:3'), covar=tensor([0.0856, 0.0747, 0.1555, 0.2583, 0.0838, 0.0858, 0.1044, 0.0976], device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0209, 0.0303, 0.0387, 0.0237, 0.0276, 0.0280, 0.0229], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2022-12-07 10:24:52,890 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.7432, 4.5246, 4.4084, 4.9003, 4.3684, 4.0418, 4.8385, 4.7974], device='cuda:3'), covar=tensor([0.0681, 0.0501, 0.0560, 0.0486, 0.0622, 0.0387, 0.0557, 0.0604], device='cuda:3'), in_proj_covar=tensor([0.0102, 0.0083, 0.0103, 0.0099, 0.0109, 0.0075, 0.0108, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2022-12-07 10:25:10,179 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2022-12-07 10:25:39,350 INFO [train.py:873] (3/4) Epoch 4, batch 1600, loss[loss=0.2238, simple_loss=0.2298, pruned_loss=0.1089, over 14312.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2051, pruned_loss=0.0969, over 1945282.85 frames. ], batch size: 46, lr: 1.96e-02, grad_scale: 8.0 2022-12-07 10:25:50,212 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.86 vs. limit=2.0 2022-12-07 10:26:05,382 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.425e+02 2.687e+02 3.404e+02 4.412e+02 1.604e+03, threshold=6.807e+02, percent-clipped=2.0 2022-12-07 10:26:24,291 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7866, 1.7344, 2.9036, 2.0304, 2.7796, 1.8009, 2.2101, 2.6074], device='cuda:3'), covar=tensor([0.0532, 0.4912, 0.0387, 0.7343, 0.0202, 0.3781, 0.1557, 0.0536], device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0296, 0.0175, 0.0394, 0.0172, 0.0306, 0.0276, 0.0178], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0002, 0.0005, 0.0002, 0.0004, 0.0003, 0.0002], device='cuda:3') 2022-12-07 10:27:05,337 INFO [train.py:873] (3/4) Epoch 4, batch 1700, loss[loss=0.1603, simple_loss=0.1761, pruned_loss=0.07223, over 13664.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2051, pruned_loss=0.0969, over 1893387.87 frames. ], batch size: 17, lr: 1.95e-02, grad_scale: 8.0 2022-12-07 10:27:10,495 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24393.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:27:31,011 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.561e+02 2.793e+02 3.554e+02 4.451e+02 1.467e+03, threshold=7.108e+02, percent-clipped=5.0 2022-12-07 10:28:31,822 INFO [train.py:873] (3/4) Epoch 4, batch 1800, loss[loss=0.2635, simple_loss=0.2443, pruned_loss=0.1413, over 10366.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2045, pruned_loss=0.09613, over 1898672.44 frames. ], batch size: 100, lr: 1.95e-02, grad_scale: 8.0 2022-12-07 10:28:40,477 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0420, 3.7948, 3.7189, 4.0930, 4.0504, 3.6391, 4.1576, 3.5075], device='cuda:3'), covar=tensor([0.0415, 0.0906, 0.0338, 0.0393, 0.0633, 0.0973, 0.0475, 0.0516], device='cuda:3'), in_proj_covar=tensor([0.0119, 0.0189, 0.0125, 0.0117, 0.0126, 0.0103, 0.0183, 0.0131], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 10:28:44,408 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2022-12-07 10:28:57,764 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.360e+02 2.645e+02 3.516e+02 4.741e+02 1.240e+03, threshold=7.031e+02, percent-clipped=7.0 2022-12-07 10:29:12,343 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.7101, 4.4218, 4.3143, 4.7099, 4.3476, 4.0977, 4.7771, 4.7233], device='cuda:3'), covar=tensor([0.0591, 0.0420, 0.0519, 0.0518, 0.0564, 0.0375, 0.0475, 0.0551], device='cuda:3'), in_proj_covar=tensor([0.0106, 0.0084, 0.0106, 0.0103, 0.0112, 0.0078, 0.0112, 0.0103], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2022-12-07 10:29:23,728 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24547.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:29:43,010 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24569.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:29:58,459 INFO [train.py:873] (3/4) Epoch 4, batch 1900, loss[loss=0.1818, simple_loss=0.1923, pruned_loss=0.08566, over 14257.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.2042, pruned_loss=0.09576, over 1970389.03 frames. ], batch size: 57, lr: 1.94e-02, grad_scale: 8.0 2022-12-07 10:30:17,385 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24608.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:30:23,219 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8253, 3.5068, 2.8911, 2.3089, 2.7944, 2.9977, 3.3432, 2.5731], device='cuda:3'), covar=tensor([0.0265, 0.2392, 0.0980, 0.2313, 0.0938, 0.0372, 0.1038, 0.1587], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0229, 0.0109, 0.0135, 0.0101, 0.0095, 0.0082, 0.0118], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0006, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-07 10:30:24,686 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.791e+02 2.734e+02 3.498e+02 4.101e+02 8.612e+02, threshold=6.995e+02, percent-clipped=1.0 2022-12-07 10:30:36,051 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24630.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:31:24,624 INFO [train.py:873] (3/4) Epoch 4, batch 2000, loss[loss=0.1723, simple_loss=0.178, pruned_loss=0.08331, over 5991.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2051, pruned_loss=0.09647, over 1945824.97 frames. ], batch size: 100, lr: 1.94e-02, grad_scale: 8.0 2022-12-07 10:31:29,956 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=24693.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:31:50,652 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.066e+02 2.930e+02 3.788e+02 4.800e+02 1.678e+03, threshold=7.576e+02, percent-clipped=5.0 2022-12-07 10:32:11,146 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=24741.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:32:14,386 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=20.95 vs. limit=5.0 2022-12-07 10:32:32,519 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24765.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:32:49,629 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8620, 2.8859, 1.9851, 3.0333, 2.8027, 2.9843, 2.4894, 2.2358], device='cuda:3'), covar=tensor([0.0263, 0.0850, 0.3116, 0.0210, 0.0422, 0.0406, 0.0949, 0.3296], device='cuda:3'), in_proj_covar=tensor([0.0216, 0.0291, 0.0330, 0.0185, 0.0231, 0.0225, 0.0262, 0.0325], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2022-12-07 10:32:51,149 INFO [train.py:873] (3/4) Epoch 4, batch 2100, loss[loss=0.2073, simple_loss=0.2132, pruned_loss=0.1007, over 14313.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.2045, pruned_loss=0.09569, over 1953943.76 frames. ], batch size: 66, lr: 1.94e-02, grad_scale: 8.0 2022-12-07 10:33:17,441 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.596e+02 2.608e+02 3.300e+02 4.069e+02 1.028e+03, threshold=6.599e+02, percent-clipped=1.0 2022-12-07 10:33:25,512 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24826.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:33:44,873 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24848.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:34:18,570 INFO [train.py:873] (3/4) Epoch 4, batch 2200, loss[loss=0.2093, simple_loss=0.2152, pruned_loss=0.1018, over 14218.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.2049, pruned_loss=0.0959, over 2033862.33 frames. ], batch size: 94, lr: 1.93e-02, grad_scale: 8.0 2022-12-07 10:34:31,894 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24903.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:34:37,043 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24909.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:34:43,926 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.370e+02 2.748e+02 3.566e+02 4.781e+02 9.343e+02, threshold=7.131e+02, percent-clipped=9.0 2022-12-07 10:34:50,879 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24925.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:35:01,356 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3670, 1.7957, 2.3700, 2.1673, 2.5648, 2.2634, 2.1946, 2.1754], device='cuda:3'), covar=tensor([0.0146, 0.0911, 0.0186, 0.0370, 0.0163, 0.0238, 0.0210, 0.0523], device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0359, 0.0266, 0.0308, 0.0290, 0.0254, 0.0287, 0.0367], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-07 10:35:28,769 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2022-12-07 10:35:44,685 INFO [train.py:873] (3/4) Epoch 4, batch 2300, loss[loss=0.2148, simple_loss=0.2181, pruned_loss=0.1057, over 12788.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.204, pruned_loss=0.09553, over 1946927.63 frames. ], batch size: 100, lr: 1.93e-02, grad_scale: 8.0 2022-12-07 10:35:51,907 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24995.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:36:03,833 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.19 vs. limit=5.0 2022-12-07 10:36:14,179 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.386e+02 2.734e+02 3.348e+02 4.422e+02 1.416e+03, threshold=6.696e+02, percent-clipped=4.0 2022-12-07 10:36:18,192 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 2022-12-07 10:36:48,424 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25056.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 10:37:08,164 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2322, 3.1088, 3.3951, 3.1476, 3.2692, 2.8020, 1.3007, 3.1011], device='cuda:3'), covar=tensor([0.0221, 0.0316, 0.0301, 0.0319, 0.0299, 0.0704, 0.2699, 0.0282], device='cuda:3'), in_proj_covar=tensor([0.0118, 0.0125, 0.0113, 0.0098, 0.0158, 0.0110, 0.0147, 0.0148], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 10:37:15,453 INFO [train.py:873] (3/4) Epoch 4, batch 2400, loss[loss=0.1954, simple_loss=0.2087, pruned_loss=0.09108, over 14287.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.2038, pruned_loss=0.0964, over 1862686.12 frames. ], batch size: 39, lr: 1.93e-02, grad_scale: 8.0 2022-12-07 10:37:22,739 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.06 vs. limit=2.0 2022-12-07 10:37:23,228 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8342, 1.3660, 2.0993, 1.6794, 2.0436, 1.3974, 1.7469, 1.8415], device='cuda:3'), covar=tensor([0.0655, 0.1342, 0.0194, 0.1288, 0.0267, 0.1172, 0.0741, 0.0374], device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0286, 0.0171, 0.0386, 0.0175, 0.0299, 0.0272, 0.0174], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0002, 0.0005, 0.0003, 0.0004, 0.0003, 0.0002], device='cuda:3') 2022-12-07 10:37:30,864 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.71 vs. limit=2.0 2022-12-07 10:37:41,017 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.389e+02 2.617e+02 3.623e+02 4.580e+02 1.018e+03, threshold=7.246e+02, percent-clipped=1.0 2022-12-07 10:37:44,904 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25121.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:38:37,212 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1984, 0.9372, 1.1556, 1.0454, 0.8784, 0.5990, 1.0356, 1.3580], device='cuda:3'), covar=tensor([0.0469, 0.0365, 0.0239, 0.0908, 0.0478, 0.0278, 0.0713, 0.0283], device='cuda:3'), in_proj_covar=tensor([0.0015, 0.0019, 0.0020, 0.0016, 0.0016, 0.0023, 0.0019, 0.0017], device='cuda:3'), out_proj_covar=tensor([5.7326e-05, 6.1615e-05, 6.0095e-05, 5.7833e-05, 5.5074e-05, 7.0861e-05, 6.6356e-05, 5.7593e-05], device='cuda:3') 2022-12-07 10:38:41,251 INFO [train.py:873] (3/4) Epoch 4, batch 2500, loss[loss=0.1964, simple_loss=0.1731, pruned_loss=0.1098, over 1304.00 frames. ], tot_loss[loss=0.1971, simple_loss=0.2036, pruned_loss=0.09528, over 1954409.34 frames. ], batch size: 100, lr: 1.92e-02, grad_scale: 8.0 2022-12-07 10:38:42,367 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7787, 1.9464, 2.9883, 2.9724, 2.6126, 1.8989, 2.9060, 2.1125], device='cuda:3'), covar=tensor([0.0095, 0.0263, 0.0197, 0.0111, 0.0116, 0.0458, 0.0055, 0.0362], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0170, 0.0229, 0.0189, 0.0157, 0.0217, 0.0128, 0.0208], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:3') 2022-12-07 10:38:55,810 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25203.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:38:56,517 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25204.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:38:58,474 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.05 vs. limit=5.0 2022-12-07 10:39:07,387 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.756e+02 2.934e+02 3.822e+02 4.724e+02 8.736e+02, threshold=7.644e+02, percent-clipped=3.0 2022-12-07 10:39:14,144 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25225.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:39:37,058 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25251.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:39:50,435 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.0953, 4.8953, 4.5556, 4.5832, 4.5099, 4.8642, 5.0583, 5.0631], device='cuda:3'), covar=tensor([0.0479, 0.0343, 0.1256, 0.2037, 0.0598, 0.0416, 0.0472, 0.0511], device='cuda:3'), in_proj_covar=tensor([0.0247, 0.0217, 0.0309, 0.0397, 0.0243, 0.0278, 0.0284, 0.0232], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0005, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2022-12-07 10:39:56,070 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25273.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:40:08,587 INFO [train.py:873] (3/4) Epoch 4, batch 2600, loss[loss=0.1761, simple_loss=0.1589, pruned_loss=0.09668, over 2629.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2042, pruned_loss=0.0959, over 1965286.85 frames. ], batch size: 100, lr: 1.92e-02, grad_scale: 8.0 2022-12-07 10:40:18,018 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2935, 2.1367, 3.1746, 2.1496, 2.2545, 2.4261, 1.2415, 2.6672], device='cuda:3'), covar=tensor([0.2325, 0.1807, 0.1311, 0.2844, 0.2421, 0.2137, 0.7738, 0.1569], device='cuda:3'), in_proj_covar=tensor([0.0068, 0.0072, 0.0065, 0.0079, 0.0088, 0.0065, 0.0138, 0.0072], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 10:40:21,600 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.0804, 1.0455, 1.0506, 1.0769, 0.8926, 0.7557, 0.8428, 0.6096], device='cuda:3'), covar=tensor([0.0400, 0.0486, 0.0385, 0.0256, 0.0499, 0.0419, 0.0255, 0.0886], device='cuda:3'), in_proj_covar=tensor([0.0012, 0.0011, 0.0012, 0.0010, 0.0012, 0.0015, 0.0013, 0.0018], device='cuda:3'), out_proj_covar=tensor([4.4787e-05, 4.2964e-05, 4.8607e-05, 3.9695e-05, 4.5624e-05, 5.3403e-05, 5.7706e-05, 6.6126e-05], device='cuda:3') 2022-12-07 10:40:34,421 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.970e+01 2.618e+02 3.461e+02 4.480e+02 1.117e+03, threshold=6.922e+02, percent-clipped=3.0 2022-12-07 10:40:51,827 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=11.76 vs. limit=5.0 2022-12-07 10:41:02,995 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25351.0, num_to_drop=1, layers_to_drop={3} 2022-12-07 10:41:24,849 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2837, 2.1702, 2.9534, 2.5031, 3.0929, 3.0524, 2.8776, 2.4313], device='cuda:3'), covar=tensor([0.0146, 0.1466, 0.0234, 0.0857, 0.0242, 0.0275, 0.0519, 0.1252], device='cuda:3'), in_proj_covar=tensor([0.0243, 0.0360, 0.0280, 0.0316, 0.0300, 0.0257, 0.0305, 0.0374], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-07 10:41:33,623 INFO [train.py:873] (3/4) Epoch 4, batch 2700, loss[loss=0.2053, simple_loss=0.213, pruned_loss=0.09873, over 14212.00 frames. ], tot_loss[loss=0.1976, simple_loss=0.2036, pruned_loss=0.0958, over 1911635.37 frames. ], batch size: 46, lr: 1.92e-02, grad_scale: 8.0 2022-12-07 10:41:59,633 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.191e+02 2.867e+02 3.536e+02 4.473e+02 9.787e+02, threshold=7.071e+02, percent-clipped=5.0 2022-12-07 10:42:03,224 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25421.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:42:09,219 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25428.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 10:42:17,153 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.6547, 2.6030, 2.7160, 2.5728, 2.6385, 2.3993, 1.3458, 2.4298], device='cuda:3'), covar=tensor([0.0264, 0.0318, 0.0401, 0.0318, 0.0338, 0.0748, 0.2302, 0.0339], device='cuda:3'), in_proj_covar=tensor([0.0118, 0.0127, 0.0115, 0.0099, 0.0158, 0.0112, 0.0147, 0.0148], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 10:42:44,239 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25469.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:42:44,438 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3978, 2.7871, 4.0369, 2.9738, 4.1094, 4.0822, 3.8734, 3.4357], device='cuda:3'), covar=tensor([0.0144, 0.1514, 0.0321, 0.1030, 0.0275, 0.0244, 0.0798, 0.1242], device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0357, 0.0278, 0.0315, 0.0298, 0.0255, 0.0300, 0.0372], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-07 10:43:00,507 INFO [train.py:873] (3/4) Epoch 4, batch 2800, loss[loss=0.1891, simple_loss=0.1993, pruned_loss=0.08947, over 5979.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2034, pruned_loss=0.09492, over 1941137.54 frames. ], batch size: 100, lr: 1.91e-02, grad_scale: 8.0 2022-12-07 10:43:02,451 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25489.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 10:43:04,189 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4082, 2.2245, 4.9436, 4.6032, 4.5239, 4.9244, 4.7931, 5.0572], device='cuda:3'), covar=tensor([0.1049, 0.1115, 0.0055, 0.0087, 0.0088, 0.0081, 0.0051, 0.0068], device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0157, 0.0091, 0.0126, 0.0107, 0.0113, 0.0083, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-07 10:43:06,721 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0079, 2.0433, 1.8833, 2.1467, 1.7089, 1.8686, 2.0215, 2.0607], device='cuda:3'), covar=tensor([0.0988, 0.0851, 0.0897, 0.0782, 0.1135, 0.0705, 0.0935, 0.0919], device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0085, 0.0103, 0.0100, 0.0112, 0.0078, 0.0112, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2022-12-07 10:43:14,985 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25504.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:43:17,422 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0240, 2.0863, 1.9589, 1.9591, 1.5412, 1.9591, 1.8383, 1.0205], device='cuda:3'), covar=tensor([0.2663, 0.0784, 0.1222, 0.0796, 0.1283, 0.0590, 0.1521, 0.3509], device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0059, 0.0051, 0.0053, 0.0068, 0.0054, 0.0078, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004], device='cuda:3') 2022-12-07 10:43:25,891 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.068e+02 2.496e+02 3.123e+02 4.474e+02 9.387e+02, threshold=6.246e+02, percent-clipped=3.0 2022-12-07 10:43:55,788 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25552.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:44:17,950 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7493, 1.5019, 1.7792, 2.1036, 1.4024, 1.5701, 1.6821, 1.8860], device='cuda:3'), covar=tensor([0.0033, 0.0062, 0.0027, 0.0017, 0.0055, 0.0071, 0.0033, 0.0025], device='cuda:3'), in_proj_covar=tensor([0.0154, 0.0169, 0.0228, 0.0190, 0.0158, 0.0215, 0.0133, 0.0208], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:3') 2022-12-07 10:44:26,292 INFO [train.py:873] (3/4) Epoch 4, batch 2900, loss[loss=0.1794, simple_loss=0.1825, pruned_loss=0.08819, over 5987.00 frames. ], tot_loss[loss=0.197, simple_loss=0.2037, pruned_loss=0.09513, over 1987775.34 frames. ], batch size: 100, lr: 1.91e-02, grad_scale: 8.0 2022-12-07 10:44:52,742 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.030e+02 2.734e+02 3.612e+02 4.564e+02 7.787e+02, threshold=7.225e+02, percent-clipped=1.0 2022-12-07 10:45:09,120 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.83 vs. limit=2.0 2022-12-07 10:45:22,065 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25651.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 10:45:53,507 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.89 vs. limit=2.0 2022-12-07 10:45:53,685 INFO [train.py:873] (3/4) Epoch 4, batch 3000, loss[loss=0.186, simple_loss=0.1924, pruned_loss=0.08982, over 14406.00 frames. ], tot_loss[loss=0.1971, simple_loss=0.2035, pruned_loss=0.09532, over 1940144.50 frames. ], batch size: 53, lr: 1.90e-02, grad_scale: 8.0 2022-12-07 10:45:53,686 INFO [train.py:896] (3/4) Computing validation loss 2022-12-07 10:46:03,274 INFO [train.py:905] (3/4) Epoch 4, validation: loss=0.1268, simple_loss=0.1698, pruned_loss=0.0419, over 857387.00 frames. 2022-12-07 10:46:03,275 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17841MB 2022-12-07 10:46:14,147 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25699.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:46:17,486 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.22 vs. limit=2.0 2022-12-07 10:46:29,652 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.391e+02 2.570e+02 3.689e+02 4.619e+02 9.776e+02, threshold=7.378e+02, percent-clipped=3.0 2022-12-07 10:46:50,373 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.90 vs. limit=2.0 2022-12-07 10:47:03,900 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.7872, 4.2937, 4.4504, 4.6858, 4.5342, 4.1319, 4.7281, 3.9395], device='cuda:3'), covar=tensor([0.0302, 0.0811, 0.0234, 0.0360, 0.0559, 0.0576, 0.0488, 0.0531], device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0200, 0.0128, 0.0126, 0.0130, 0.0109, 0.0192, 0.0131], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 10:47:14,742 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5143, 4.0810, 4.1747, 4.4729, 4.4069, 3.9606, 4.5410, 3.7119], device='cuda:3'), covar=tensor([0.0461, 0.0961, 0.0287, 0.0411, 0.0597, 0.0756, 0.0521, 0.0572], device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0202, 0.0129, 0.0126, 0.0130, 0.0110, 0.0193, 0.0131], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 10:47:20,732 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.23 vs. limit=2.0 2022-12-07 10:47:28,455 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25784.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 10:47:30,952 INFO [train.py:873] (3/4) Epoch 4, batch 3100, loss[loss=0.1722, simple_loss=0.1916, pruned_loss=0.07639, over 14016.00 frames. ], tot_loss[loss=0.1961, simple_loss=0.2027, pruned_loss=0.09481, over 1948270.21 frames. ], batch size: 29, lr: 1.90e-02, grad_scale: 8.0 2022-12-07 10:47:57,004 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.131e+02 2.578e+02 3.545e+02 4.492e+02 7.913e+02, threshold=7.089e+02, percent-clipped=1.0 2022-12-07 10:48:27,094 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0013, 1.8368, 2.1981, 1.5684, 1.5827, 2.0911, 1.2486, 1.9373], device='cuda:3'), covar=tensor([0.1036, 0.2255, 0.0716, 0.1755, 0.2489, 0.0846, 0.4985, 0.1032], device='cuda:3'), in_proj_covar=tensor([0.0067, 0.0073, 0.0067, 0.0078, 0.0090, 0.0061, 0.0136, 0.0073], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 10:48:33,003 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=7.43 vs. limit=5.0 2022-12-07 10:48:33,406 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25858.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 10:48:54,294 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.94 vs. limit=2.0 2022-12-07 10:48:57,813 INFO [train.py:873] (3/4) Epoch 4, batch 3200, loss[loss=0.2173, simple_loss=0.2218, pruned_loss=0.1064, over 14507.00 frames. ], tot_loss[loss=0.1968, simple_loss=0.2034, pruned_loss=0.09516, over 1928518.35 frames. ], batch size: 49, lr: 1.90e-02, grad_scale: 8.0 2022-12-07 10:49:24,095 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.295e+02 2.801e+02 3.510e+02 4.555e+02 1.491e+03, threshold=7.021e+02, percent-clipped=7.0 2022-12-07 10:49:26,030 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25919.0, num_to_drop=1, layers_to_drop={3} 2022-12-07 10:49:34,041 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2022-12-07 10:49:45,128 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.48 vs. limit=5.0 2022-12-07 10:50:25,176 INFO [train.py:873] (3/4) Epoch 4, batch 3300, loss[loss=0.1805, simple_loss=0.1953, pruned_loss=0.08289, over 14173.00 frames. ], tot_loss[loss=0.1947, simple_loss=0.2019, pruned_loss=0.09369, over 1980748.59 frames. ], batch size: 89, lr: 1.89e-02, grad_scale: 8.0 2022-12-07 10:50:51,957 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.542e+02 2.499e+02 3.235e+02 4.436e+02 1.031e+03, threshold=6.470e+02, percent-clipped=3.0 2022-12-07 10:51:37,966 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3112, 1.5206, 2.6990, 2.0950, 2.3987, 1.7202, 1.9286, 2.2708], device='cuda:3'), covar=tensor([0.0791, 0.3683, 0.0276, 0.5499, 0.0260, 0.3472, 0.1711, 0.0251], device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0286, 0.0178, 0.0392, 0.0179, 0.0298, 0.0273, 0.0176], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 10:51:48,817 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26084.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 10:51:51,018 INFO [train.py:873] (3/4) Epoch 4, batch 3400, loss[loss=0.1916, simple_loss=0.2021, pruned_loss=0.09054, over 11214.00 frames. ], tot_loss[loss=0.1942, simple_loss=0.2015, pruned_loss=0.0934, over 1963928.14 frames. ], batch size: 100, lr: 1.89e-02, grad_scale: 8.0 2022-12-07 10:52:04,862 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26102.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:52:07,622 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.7957, 4.3433, 3.1043, 4.9325, 4.2849, 4.5965, 3.9963, 3.7758], device='cuda:3'), covar=tensor([0.0196, 0.0634, 0.3607, 0.0178, 0.0550, 0.0735, 0.0820, 0.2460], device='cuda:3'), in_proj_covar=tensor([0.0220, 0.0303, 0.0322, 0.0193, 0.0238, 0.0231, 0.0264, 0.0316], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:3') 2022-12-07 10:52:18,305 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.312e+02 2.979e+02 3.585e+02 4.923e+02 1.213e+03, threshold=7.170e+02, percent-clipped=7.0 2022-12-07 10:52:30,057 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26132.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 10:52:54,739 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.2983, 1.1194, 1.4715, 1.1838, 1.0378, 1.4033, 0.9276, 1.3069], device='cuda:3'), covar=tensor([0.1244, 0.2724, 0.0489, 0.1382, 0.2163, 0.0579, 0.1865, 0.0775], device='cuda:3'), in_proj_covar=tensor([0.0070, 0.0075, 0.0070, 0.0082, 0.0095, 0.0063, 0.0143, 0.0073], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 10:52:57,468 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26163.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:53:01,645 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26168.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:53:02,089 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.85 vs. limit=2.0 2022-12-07 10:53:05,820 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8646, 0.8023, 0.7530, 0.9456, 0.8426, 0.4822, 1.0276, 1.0203], device='cuda:3'), covar=tensor([0.0363, 0.0606, 0.0234, 0.0289, 0.0489, 0.0214, 0.0368, 0.0288], device='cuda:3'), in_proj_covar=tensor([0.0016, 0.0019, 0.0021, 0.0017, 0.0019, 0.0023, 0.0019, 0.0018], device='cuda:3'), out_proj_covar=tensor([6.0264e-05, 6.4714e-05, 6.4669e-05, 6.0981e-05, 6.3529e-05, 7.6410e-05, 6.8617e-05, 6.1205e-05], device='cuda:3') 2022-12-07 10:53:18,123 INFO [train.py:873] (3/4) Epoch 4, batch 3500, loss[loss=0.203, simple_loss=0.1888, pruned_loss=0.1086, over 4979.00 frames. ], tot_loss[loss=0.1942, simple_loss=0.2016, pruned_loss=0.09338, over 1936859.62 frames. ], batch size: 100, lr: 1.89e-02, grad_scale: 8.0 2022-12-07 10:53:32,049 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5091, 3.0110, 4.3729, 3.1182, 4.2251, 4.1874, 3.9926, 3.4757], device='cuda:3'), covar=tensor([0.0169, 0.1799, 0.0250, 0.0993, 0.0362, 0.0275, 0.0894, 0.1645], device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0349, 0.0282, 0.0307, 0.0295, 0.0254, 0.0294, 0.0357], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-07 10:53:41,589 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26214.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 10:53:44,755 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.551e+02 2.647e+02 3.463e+02 4.115e+02 9.483e+02, threshold=6.926e+02, percent-clipped=4.0 2022-12-07 10:53:53,234 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.00 vs. limit=5.0 2022-12-07 10:53:54,565 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26229.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:54:12,016 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26250.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:54:12,489 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2022-12-07 10:54:44,200 INFO [train.py:873] (3/4) Epoch 4, batch 3600, loss[loss=0.1782, simple_loss=0.1871, pruned_loss=0.08472, over 10311.00 frames. ], tot_loss[loss=0.1931, simple_loss=0.2012, pruned_loss=0.09246, over 2016579.66 frames. ], batch size: 100, lr: 1.88e-02, grad_scale: 8.0 2022-12-07 10:55:03,081 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.01 vs. limit=2.0 2022-12-07 10:55:05,369 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26311.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:55:10,969 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.347e+02 2.617e+02 3.267e+02 4.043e+02 9.670e+02, threshold=6.534e+02, percent-clipped=3.0 2022-12-07 10:55:28,031 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.67 vs. limit=5.0 2022-12-07 10:55:36,350 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.41 vs. limit=5.0 2022-12-07 10:56:06,821 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.05 vs. limit=2.0 2022-12-07 10:56:10,706 INFO [train.py:873] (3/4) Epoch 4, batch 3700, loss[loss=0.2124, simple_loss=0.1998, pruned_loss=0.1125, over 3903.00 frames. ], tot_loss[loss=0.1947, simple_loss=0.2021, pruned_loss=0.09361, over 1995806.41 frames. ], batch size: 100, lr: 1.88e-02, grad_scale: 8.0 2022-12-07 10:56:34,085 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26414.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:56:37,278 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.464e+02 2.639e+02 3.495e+02 4.493e+02 9.568e+02, threshold=6.990e+02, percent-clipped=8.0 2022-12-07 10:56:49,763 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5571, 4.1332, 4.1835, 4.5721, 4.3522, 4.0208, 4.6077, 3.8718], device='cuda:3'), covar=tensor([0.0398, 0.0969, 0.0287, 0.0417, 0.0727, 0.0746, 0.0459, 0.0469], device='cuda:3'), in_proj_covar=tensor([0.0119, 0.0199, 0.0128, 0.0123, 0.0131, 0.0109, 0.0190, 0.0129], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 10:56:59,196 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8142, 2.4522, 2.2874, 1.3505, 2.3762, 2.4783, 2.9216, 2.0881], device='cuda:3'), covar=tensor([0.0543, 0.2243, 0.1268, 0.3557, 0.0800, 0.0415, 0.0477, 0.1929], device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0223, 0.0109, 0.0130, 0.0099, 0.0094, 0.0079, 0.0121], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0006, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-07 10:57:01,614 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5450, 2.2201, 3.6055, 3.7252, 3.6479, 2.2290, 3.4907, 2.8797], device='cuda:3'), covar=tensor([0.0081, 0.0259, 0.0203, 0.0105, 0.0072, 0.0393, 0.0059, 0.0277], device='cuda:3'), in_proj_covar=tensor([0.0160, 0.0174, 0.0240, 0.0201, 0.0160, 0.0222, 0.0136, 0.0215], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:3') 2022-12-07 10:57:11,875 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26458.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:57:26,716 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26475.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:57:36,083 INFO [train.py:873] (3/4) Epoch 4, batch 3800, loss[loss=0.1991, simple_loss=0.1779, pruned_loss=0.1102, over 3883.00 frames. ], tot_loss[loss=0.1949, simple_loss=0.2023, pruned_loss=0.09379, over 1980455.56 frames. ], batch size: 100, lr: 1.88e-02, grad_scale: 8.0 2022-12-07 10:58:00,272 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26514.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 10:58:03,348 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.297e+02 2.752e+02 3.266e+02 4.284e+02 8.876e+02, threshold=6.533e+02, percent-clipped=2.0 2022-12-07 10:58:08,566 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26524.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:58:41,459 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26562.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 10:58:52,846 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6849, 2.1998, 3.7054, 3.7727, 3.9278, 2.2956, 3.8889, 2.8419], device='cuda:3'), covar=tensor([0.0082, 0.0270, 0.0267, 0.0145, 0.0062, 0.0431, 0.0054, 0.0345], device='cuda:3'), in_proj_covar=tensor([0.0161, 0.0174, 0.0239, 0.0201, 0.0160, 0.0223, 0.0135, 0.0216], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:3') 2022-12-07 10:59:03,303 INFO [train.py:873] (3/4) Epoch 4, batch 3900, loss[loss=0.177, simple_loss=0.1977, pruned_loss=0.07812, over 14254.00 frames. ], tot_loss[loss=0.1931, simple_loss=0.2008, pruned_loss=0.09272, over 1927583.83 frames. ], batch size: 57, lr: 1.87e-02, grad_scale: 8.0 2022-12-07 10:59:19,329 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26606.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 10:59:29,338 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.374e+02 2.794e+02 3.682e+02 4.681e+02 9.431e+02, threshold=7.364e+02, percent-clipped=3.0 2022-12-07 10:59:40,042 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.03 vs. limit=2.0 2022-12-07 11:00:28,849 INFO [train.py:873] (3/4) Epoch 4, batch 4000, loss[loss=0.1752, simple_loss=0.1539, pruned_loss=0.09827, over 1193.00 frames. ], tot_loss[loss=0.1936, simple_loss=0.2012, pruned_loss=0.093, over 1933004.47 frames. ], batch size: 100, lr: 1.87e-02, grad_scale: 8.0 2022-12-07 11:00:50,790 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8015, 4.4976, 5.0114, 3.6390, 4.6074, 5.1266, 1.9437, 4.3762], device='cuda:3'), covar=tensor([0.0146, 0.0271, 0.0321, 0.0498, 0.0280, 0.0097, 0.2833, 0.0281], device='cuda:3'), in_proj_covar=tensor([0.0117, 0.0127, 0.0119, 0.0100, 0.0157, 0.0112, 0.0149, 0.0151], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 11:00:55,535 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.732e+02 2.728e+02 3.637e+02 4.905e+02 8.607e+02, threshold=7.274e+02, percent-clipped=2.0 2022-12-07 11:01:25,482 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2174, 2.1799, 2.1585, 1.9950, 1.6191, 2.1858, 2.1064, 1.0988], device='cuda:3'), covar=tensor([0.4016, 0.1026, 0.1306, 0.1151, 0.1279, 0.0618, 0.1612, 0.4036], device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0060, 0.0053, 0.0054, 0.0069, 0.0055, 0.0079, 0.0103], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004], device='cuda:3') 2022-12-07 11:01:29,177 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26758.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:01:39,470 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26770.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:01:53,836 INFO [train.py:873] (3/4) Epoch 4, batch 4100, loss[loss=0.2131, simple_loss=0.218, pruned_loss=0.1041, over 12737.00 frames. ], tot_loss[loss=0.1944, simple_loss=0.2014, pruned_loss=0.09375, over 1864409.47 frames. ], batch size: 100, lr: 1.87e-02, grad_scale: 8.0 2022-12-07 11:02:10,439 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26806.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:02:20,840 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.226e+02 2.498e+02 3.224e+02 4.280e+02 1.572e+03, threshold=6.447e+02, percent-clipped=4.0 2022-12-07 11:02:26,166 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26824.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:02:52,169 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4156, 1.2968, 3.5264, 1.3288, 3.3560, 3.4570, 2.4666, 3.7111], device='cuda:3'), covar=tensor([0.0170, 0.2755, 0.0250, 0.2326, 0.0594, 0.0258, 0.0666, 0.0126], device='cuda:3'), in_proj_covar=tensor([0.0137, 0.0156, 0.0125, 0.0165, 0.0139, 0.0132, 0.0114, 0.0111], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 11:03:07,339 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26872.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:03:19,883 INFO [train.py:873] (3/4) Epoch 4, batch 4200, loss[loss=0.1657, simple_loss=0.1453, pruned_loss=0.09308, over 1248.00 frames. ], tot_loss[loss=0.194, simple_loss=0.2014, pruned_loss=0.09331, over 1919879.08 frames. ], batch size: 100, lr: 1.86e-02, grad_scale: 8.0 2022-12-07 11:03:33,874 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.7672, 3.0323, 4.4733, 3.3016, 4.3056, 4.4508, 3.9044, 3.7681], device='cuda:3'), covar=tensor([0.0200, 0.1555, 0.0284, 0.0905, 0.0380, 0.0247, 0.1338, 0.0974], device='cuda:3'), in_proj_covar=tensor([0.0249, 0.0346, 0.0286, 0.0309, 0.0305, 0.0256, 0.0301, 0.0364], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-07 11:03:34,643 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26904.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 11:03:36,778 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26906.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:03:46,350 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.525e+02 2.657e+02 3.822e+02 4.895e+02 8.860e+02, threshold=7.643e+02, percent-clipped=9.0 2022-12-07 11:04:01,731 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9881, 1.5232, 2.4289, 1.9684, 2.2686, 1.5738, 1.9299, 2.0110], device='cuda:3'), covar=tensor([0.0876, 0.3067, 0.0207, 0.2520, 0.0268, 0.2167, 0.0905, 0.0316], device='cuda:3'), in_proj_covar=tensor([0.0226, 0.0282, 0.0177, 0.0380, 0.0177, 0.0293, 0.0261, 0.0176], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0005, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2022-12-07 11:04:13,545 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26950.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:04:16,891 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26954.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:04:26,341 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26965.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 11:04:45,369 INFO [train.py:873] (3/4) Epoch 4, batch 4300, loss[loss=0.1852, simple_loss=0.2069, pruned_loss=0.08176, over 14002.00 frames. ], tot_loss[loss=0.1948, simple_loss=0.2021, pruned_loss=0.09374, over 1968127.77 frames. ], batch size: 26, lr: 1.86e-02, grad_scale: 8.0 2022-12-07 11:05:05,722 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27011.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:05:11,530 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.439e+02 2.870e+02 3.578e+02 4.388e+02 9.558e+02, threshold=7.157e+02, percent-clipped=3.0 2022-12-07 11:05:28,110 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0804, 1.8469, 2.0273, 2.0756, 2.0011, 1.9784, 2.1607, 1.8326], device='cuda:3'), covar=tensor([0.0579, 0.1253, 0.0489, 0.0599, 0.0932, 0.0565, 0.0577, 0.0608], device='cuda:3'), in_proj_covar=tensor([0.0120, 0.0200, 0.0129, 0.0124, 0.0133, 0.0108, 0.0193, 0.0132], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 11:05:57,165 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27070.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:06:11,314 INFO [train.py:873] (3/4) Epoch 4, batch 4400, loss[loss=0.1656, simple_loss=0.1863, pruned_loss=0.07251, over 14338.00 frames. ], tot_loss[loss=0.1937, simple_loss=0.2019, pruned_loss=0.0928, over 2035443.84 frames. ], batch size: 39, lr: 1.86e-02, grad_scale: 8.0 2022-12-07 11:06:22,618 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9953, 4.8197, 4.6349, 5.1109, 4.6818, 4.0450, 5.2282, 5.0838], device='cuda:3'), covar=tensor([0.0718, 0.0503, 0.0650, 0.0636, 0.0612, 0.0530, 0.0526, 0.0655], device='cuda:3'), in_proj_covar=tensor([0.0106, 0.0085, 0.0103, 0.0103, 0.0111, 0.0081, 0.0112, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2022-12-07 11:06:37,228 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.88 vs. limit=2.0 2022-12-07 11:06:38,400 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.237e+02 2.559e+02 3.461e+02 4.606e+02 7.213e+02, threshold=6.922e+02, percent-clipped=1.0 2022-12-07 11:06:38,482 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=27118.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:06:52,826 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.7790, 0.6136, 0.7403, 0.5601, 0.6215, 0.3884, 0.5767, 0.5423], device='cuda:3'), covar=tensor([0.0120, 0.0049, 0.0100, 0.0118, 0.0337, 0.0242, 0.0086, 0.0247], device='cuda:3'), in_proj_covar=tensor([0.0011, 0.0011, 0.0010, 0.0010, 0.0011, 0.0015, 0.0011, 0.0017], device='cuda:3'), out_proj_covar=tensor([4.4557e-05, 4.5048e-05, 4.4904e-05, 4.2342e-05, 4.7016e-05, 5.6676e-05, 5.3007e-05, 6.5058e-05], device='cuda:3') 2022-12-07 11:07:04,488 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27148.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:07:37,305 INFO [train.py:873] (3/4) Epoch 4, batch 4500, loss[loss=0.1857, simple_loss=0.2, pruned_loss=0.08571, over 14269.00 frames. ], tot_loss[loss=0.1942, simple_loss=0.2022, pruned_loss=0.09305, over 2005420.78 frames. ], batch size: 46, lr: 1.85e-02, grad_scale: 8.0 2022-12-07 11:07:39,174 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.32 vs. limit=5.0 2022-12-07 11:07:57,172 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27209.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:08:04,791 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.722e+01 2.647e+02 3.341e+02 4.493e+02 1.287e+03, threshold=6.683e+02, percent-clipped=3.0 2022-12-07 11:08:37,444 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8976, 3.5269, 3.5877, 3.8896, 3.7238, 3.4753, 3.9281, 3.4273], device='cuda:3'), covar=tensor([0.0375, 0.0970, 0.0346, 0.0393, 0.0676, 0.1052, 0.0517, 0.0445], device='cuda:3'), in_proj_covar=tensor([0.0119, 0.0198, 0.0130, 0.0124, 0.0131, 0.0108, 0.0192, 0.0132], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 11:08:40,900 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=27260.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 11:09:04,081 INFO [train.py:873] (3/4) Epoch 4, batch 4600, loss[loss=0.1881, simple_loss=0.1672, pruned_loss=0.1045, over 1301.00 frames. ], tot_loss[loss=0.1936, simple_loss=0.2019, pruned_loss=0.0926, over 2007763.56 frames. ], batch size: 100, lr: 1.85e-02, grad_scale: 8.0 2022-12-07 11:09:20,190 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=27306.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:09:30,612 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.463e+02 3.048e+02 4.012e+02 4.961e+02 8.817e+02, threshold=8.024e+02, percent-clipped=5.0 2022-12-07 11:09:54,156 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2022-12-07 11:09:55,428 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0292, 1.6957, 2.2423, 2.3765, 1.8646, 1.7139, 2.2291, 2.0164], device='cuda:3'), covar=tensor([0.0046, 0.0111, 0.0051, 0.0035, 0.0056, 0.0140, 0.0050, 0.0065], device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0171, 0.0240, 0.0199, 0.0156, 0.0220, 0.0132, 0.0213], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:3') 2022-12-07 11:10:29,693 INFO [train.py:873] (3/4) Epoch 4, batch 4700, loss[loss=0.1965, simple_loss=0.1694, pruned_loss=0.1118, over 1249.00 frames. ], tot_loss[loss=0.1946, simple_loss=0.2024, pruned_loss=0.09339, over 1993786.21 frames. ], batch size: 100, lr: 1.85e-02, grad_scale: 8.0 2022-12-07 11:10:42,338 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7718, 1.2587, 3.3794, 3.2604, 3.3603, 3.4671, 2.9796, 3.4971], device='cuda:3'), covar=tensor([0.1176, 0.1461, 0.0128, 0.0168, 0.0162, 0.0107, 0.0181, 0.0110], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0150, 0.0093, 0.0127, 0.0108, 0.0113, 0.0081, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-07 11:10:56,345 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.523e+02 2.544e+02 3.410e+02 4.417e+02 8.558e+02, threshold=6.819e+02, percent-clipped=1.0 2022-12-07 11:11:22,447 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.0061, 0.6853, 0.6348, 0.9965, 1.2119, 0.6931, 1.0821, 0.9929], device='cuda:3'), covar=tensor([0.0777, 0.0611, 0.0321, 0.0594, 0.0343, 0.0317, 0.0757, 0.0509], device='cuda:3'), in_proj_covar=tensor([0.0016, 0.0017, 0.0019, 0.0015, 0.0017, 0.0022, 0.0017, 0.0017], device='cuda:3'), out_proj_covar=tensor([5.8867e-05, 5.8346e-05, 6.0581e-05, 5.6928e-05, 5.6250e-05, 7.3308e-05, 6.3654e-05, 5.8287e-05], device='cuda:3') 2022-12-07 11:11:22,995 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.91 vs. limit=2.0 2022-12-07 11:11:27,939 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.67 vs. limit=2.0 2022-12-07 11:11:56,835 INFO [train.py:873] (3/4) Epoch 4, batch 4800, loss[loss=0.2166, simple_loss=0.2084, pruned_loss=0.1124, over 5998.00 frames. ], tot_loss[loss=0.1927, simple_loss=0.2014, pruned_loss=0.09201, over 2026523.37 frames. ], batch size: 100, lr: 1.84e-02, grad_scale: 8.0 2022-12-07 11:12:11,709 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=27504.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:12:24,599 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.288e+02 3.057e+02 3.716e+02 4.446e+02 1.010e+03, threshold=7.432e+02, percent-clipped=2.0 2022-12-07 11:12:28,275 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.2170, 0.7462, 0.7060, 1.1739, 1.2588, 0.7055, 1.2070, 0.9168], device='cuda:3'), covar=tensor([0.0518, 0.0454, 0.0246, 0.0306, 0.0288, 0.0308, 0.0206, 0.0701], device='cuda:3'), in_proj_covar=tensor([0.0016, 0.0017, 0.0019, 0.0016, 0.0017, 0.0023, 0.0017, 0.0018], device='cuda:3'), out_proj_covar=tensor([6.0327e-05, 5.9941e-05, 6.1741e-05, 5.8924e-05, 5.6906e-05, 7.5762e-05, 6.4614e-05, 6.1001e-05], device='cuda:3') 2022-12-07 11:13:00,418 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27560.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 11:13:07,910 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0673, 1.8744, 1.9822, 2.0769, 2.0443, 1.9534, 2.1798, 1.7707], device='cuda:3'), covar=tensor([0.0593, 0.1239, 0.0574, 0.0556, 0.0761, 0.0553, 0.0681, 0.0731], device='cuda:3'), in_proj_covar=tensor([0.0120, 0.0197, 0.0133, 0.0126, 0.0130, 0.0107, 0.0194, 0.0133], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 11:13:23,629 INFO [train.py:873] (3/4) Epoch 4, batch 4900, loss[loss=0.173, simple_loss=0.192, pruned_loss=0.077, over 14355.00 frames. ], tot_loss[loss=0.1935, simple_loss=0.2013, pruned_loss=0.09279, over 1956127.55 frames. ], batch size: 28, lr: 1.84e-02, grad_scale: 8.0 2022-12-07 11:13:40,775 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27606.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:13:42,438 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=27608.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 11:13:51,358 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.582e+02 2.749e+02 3.689e+02 4.632e+02 8.661e+02, threshold=7.378e+02, percent-clipped=1.0 2022-12-07 11:14:02,340 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4467, 0.7839, 0.8507, 1.1321, 1.0724, 0.9179, 1.3020, 1.2760], device='cuda:3'), covar=tensor([0.0524, 0.0445, 0.0234, 0.0577, 0.0780, 0.0331, 0.0555, 0.0463], device='cuda:3'), in_proj_covar=tensor([0.0016, 0.0017, 0.0019, 0.0015, 0.0017, 0.0022, 0.0017, 0.0017], device='cuda:3'), out_proj_covar=tensor([5.7896e-05, 5.8373e-05, 6.0646e-05, 5.6571e-05, 5.6295e-05, 7.5003e-05, 6.3381e-05, 5.8541e-05], device='cuda:3') 2022-12-07 11:14:05,910 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7880, 3.3857, 3.4939, 3.9220, 3.5091, 2.9923, 3.8691, 3.7960], device='cuda:3'), covar=tensor([0.0811, 0.0789, 0.0690, 0.0554, 0.0815, 0.0710, 0.0698, 0.0819], device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0089, 0.0105, 0.0103, 0.0113, 0.0081, 0.0114, 0.0106], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2022-12-07 11:14:21,723 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=27654.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:14:50,478 INFO [train.py:873] (3/4) Epoch 4, batch 5000, loss[loss=0.1979, simple_loss=0.208, pruned_loss=0.0939, over 14256.00 frames. ], tot_loss[loss=0.1943, simple_loss=0.2017, pruned_loss=0.09342, over 1918468.94 frames. ], batch size: 80, lr: 1.84e-02, grad_scale: 8.0 2022-12-07 11:15:01,419 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.57 vs. limit=2.0 2022-12-07 11:15:14,333 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27714.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 11:15:18,161 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.210e+02 2.479e+02 3.081e+02 3.772e+02 7.331e+02, threshold=6.161e+02, percent-clipped=0.0 2022-12-07 11:16:06,288 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27775.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 11:16:16,333 INFO [train.py:873] (3/4) Epoch 4, batch 5100, loss[loss=0.153, simple_loss=0.1815, pruned_loss=0.06224, over 14371.00 frames. ], tot_loss[loss=0.1936, simple_loss=0.2009, pruned_loss=0.09311, over 1956756.76 frames. ], batch size: 18, lr: 1.83e-02, grad_scale: 8.0 2022-12-07 11:16:31,665 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27804.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:16:37,339 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8096, 1.2954, 1.8047, 1.0348, 1.3982, 1.6657, 1.6641, 1.5205], device='cuda:3'), covar=tensor([0.0552, 0.0987, 0.0624, 0.0999, 0.0957, 0.0702, 0.0493, 0.1291], device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0225, 0.0109, 0.0129, 0.0100, 0.0097, 0.0080, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0006, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-07 11:16:43,937 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.421e+02 2.535e+02 3.343e+02 4.225e+02 1.167e+03, threshold=6.685e+02, percent-clipped=6.0 2022-12-07 11:17:12,654 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=27852.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:17:43,050 INFO [train.py:873] (3/4) Epoch 4, batch 5200, loss[loss=0.1737, simple_loss=0.1988, pruned_loss=0.07425, over 14301.00 frames. ], tot_loss[loss=0.1933, simple_loss=0.2011, pruned_loss=0.09272, over 1969712.31 frames. ], batch size: 60, lr: 1.83e-02, grad_scale: 8.0 2022-12-07 11:18:04,427 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3328, 2.1850, 2.3097, 1.9143, 1.4743, 2.1808, 2.1171, 0.9412], device='cuda:3'), covar=tensor([0.3422, 0.1057, 0.1678, 0.1274, 0.1358, 0.0617, 0.1560, 0.4351], device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0058, 0.0051, 0.0051, 0.0070, 0.0056, 0.0081, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004], device='cuda:3') 2022-12-07 11:18:10,409 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.340e+02 2.745e+02 3.569e+02 4.514e+02 6.730e+02, threshold=7.138e+02, percent-clipped=2.0 2022-12-07 11:18:21,354 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0875, 3.0225, 3.1508, 2.9371, 3.0040, 2.9949, 1.2087, 2.7655], device='cuda:3'), covar=tensor([0.0182, 0.0253, 0.0310, 0.0275, 0.0272, 0.0431, 0.2652, 0.0261], device='cuda:3'), in_proj_covar=tensor([0.0115, 0.0126, 0.0116, 0.0101, 0.0157, 0.0109, 0.0144, 0.0148], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 11:19:08,658 INFO [train.py:873] (3/4) Epoch 4, batch 5300, loss[loss=0.1854, simple_loss=0.1933, pruned_loss=0.08876, over 14226.00 frames. ], tot_loss[loss=0.1931, simple_loss=0.2011, pruned_loss=0.09253, over 1964217.77 frames. ], batch size: 89, lr: 1.83e-02, grad_scale: 8.0 2022-12-07 11:19:29,643 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5850, 2.4174, 2.1871, 2.2855, 2.4251, 2.4252, 2.5262, 2.5017], device='cuda:3'), covar=tensor([0.0769, 0.0864, 0.1766, 0.2031, 0.0863, 0.0840, 0.1047, 0.0800], device='cuda:3'), in_proj_covar=tensor([0.0255, 0.0216, 0.0324, 0.0405, 0.0243, 0.0287, 0.0302, 0.0240], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0005, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2022-12-07 11:19:36,271 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.465e+02 2.605e+02 3.441e+02 4.511e+02 8.751e+02, threshold=6.883e+02, percent-clipped=1.0 2022-12-07 11:20:20,866 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28070.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 11:20:35,860 INFO [train.py:873] (3/4) Epoch 4, batch 5400, loss[loss=0.226, simple_loss=0.1931, pruned_loss=0.1295, over 1172.00 frames. ], tot_loss[loss=0.1925, simple_loss=0.2006, pruned_loss=0.09222, over 1966466.90 frames. ], batch size: 100, lr: 1.82e-02, grad_scale: 8.0 2022-12-07 11:21:03,785 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.374e+02 2.371e+02 3.121e+02 4.059e+02 8.001e+02, threshold=6.242e+02, percent-clipped=5.0 2022-12-07 11:21:07,630 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3992, 4.3238, 4.2967, 4.0192, 4.3010, 4.7190, 1.4535, 4.1581], device='cuda:3'), covar=tensor([0.0243, 0.0357, 0.0773, 0.0579, 0.0409, 0.0179, 0.4131, 0.0349], device='cuda:3'), in_proj_covar=tensor([0.0119, 0.0131, 0.0121, 0.0104, 0.0162, 0.0114, 0.0151, 0.0156], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 11:21:26,492 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7016, 2.8059, 3.4967, 2.5411, 2.5747, 2.5048, 1.2904, 2.5430], device='cuda:3'), covar=tensor([0.2074, 0.1213, 0.0633, 0.1707, 0.1704, 0.1621, 0.6179, 0.1421], device='cuda:3'), in_proj_covar=tensor([0.0071, 0.0076, 0.0070, 0.0083, 0.0098, 0.0065, 0.0142, 0.0072], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 11:21:42,259 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1088, 1.5470, 2.4011, 1.9093, 2.1971, 1.5875, 1.8237, 1.9576], device='cuda:3'), covar=tensor([0.0920, 0.2759, 0.0290, 0.3254, 0.0267, 0.2297, 0.1335, 0.0716], device='cuda:3'), in_proj_covar=tensor([0.0230, 0.0276, 0.0175, 0.0371, 0.0171, 0.0291, 0.0264, 0.0179], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 11:22:02,364 INFO [train.py:873] (3/4) Epoch 4, batch 5500, loss[loss=0.1942, simple_loss=0.1663, pruned_loss=0.1111, over 1242.00 frames. ], tot_loss[loss=0.1908, simple_loss=0.1996, pruned_loss=0.09094, over 1977522.96 frames. ], batch size: 100, lr: 1.82e-02, grad_scale: 8.0 2022-12-07 11:22:06,522 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.36 vs. limit=2.0 2022-12-07 11:22:24,568 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1571, 1.5805, 2.5204, 1.8812, 2.1748, 1.5856, 1.9395, 2.0034], device='cuda:3'), covar=tensor([0.0828, 0.3205, 0.0274, 0.4295, 0.0253, 0.2983, 0.1308, 0.0560], device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0279, 0.0176, 0.0372, 0.0171, 0.0292, 0.0266, 0.0178], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 11:22:30,324 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.463e+02 2.632e+02 3.347e+02 4.419e+02 1.188e+03, threshold=6.694e+02, percent-clipped=11.0 2022-12-07 11:22:39,479 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28229.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:23:01,978 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28255.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 11:23:10,294 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4104, 0.9423, 0.8511, 0.9887, 1.3641, 1.0802, 1.1644, 1.3861], device='cuda:3'), covar=tensor([0.0289, 0.0448, 0.0243, 0.0331, 0.0409, 0.0291, 0.0312, 0.0247], device='cuda:3'), in_proj_covar=tensor([0.0015, 0.0018, 0.0020, 0.0017, 0.0017, 0.0023, 0.0016, 0.0017], device='cuda:3'), out_proj_covar=tensor([5.7489e-05, 6.2892e-05, 6.3638e-05, 6.1393e-05, 5.9296e-05, 7.8500e-05, 6.3388e-05, 6.0964e-05], device='cuda:3') 2022-12-07 11:23:29,472 INFO [train.py:873] (3/4) Epoch 4, batch 5600, loss[loss=0.1815, simple_loss=0.1872, pruned_loss=0.08786, over 4936.00 frames. ], tot_loss[loss=0.1906, simple_loss=0.1994, pruned_loss=0.09096, over 1947046.25 frames. ], batch size: 100, lr: 1.82e-02, grad_scale: 8.0 2022-12-07 11:23:29,670 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28287.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:23:32,025 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28290.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:23:54,694 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28316.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 11:23:57,191 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.230e+02 2.669e+02 3.555e+02 4.575e+02 8.799e+02, threshold=7.110e+02, percent-clipped=5.0 2022-12-07 11:24:21,862 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28348.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:24:22,195 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.02 vs. limit=5.0 2022-12-07 11:24:37,868 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8962, 2.7385, 2.0574, 2.9638, 2.6007, 2.8322, 2.3996, 2.2563], device='cuda:3'), covar=tensor([0.0320, 0.0824, 0.2659, 0.0223, 0.0567, 0.0353, 0.0956, 0.2440], device='cuda:3'), in_proj_covar=tensor([0.0222, 0.0302, 0.0315, 0.0190, 0.0238, 0.0239, 0.0260, 0.0311], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 11:24:41,074 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28370.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 11:24:50,577 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.7874, 1.1381, 1.2988, 1.3036, 1.1492, 1.2509, 1.0966, 0.8543], device='cuda:3'), covar=tensor([0.2606, 0.0642, 0.0417, 0.0487, 0.1160, 0.0308, 0.1845, 0.1442], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0054, 0.0051, 0.0048, 0.0070, 0.0052, 0.0079, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004], device='cuda:3') 2022-12-07 11:24:55,322 INFO [train.py:873] (3/4) Epoch 4, batch 5700, loss[loss=0.1969, simple_loss=0.2061, pruned_loss=0.09388, over 11977.00 frames. ], tot_loss[loss=0.1918, simple_loss=0.2002, pruned_loss=0.09168, over 2008620.46 frames. ], batch size: 100, lr: 1.81e-02, grad_scale: 8.0 2022-12-07 11:25:04,267 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.20 vs. limit=5.0 2022-12-07 11:25:14,569 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4568, 1.6853, 4.1687, 1.6927, 4.0779, 4.2151, 3.7780, 4.7051], device='cuda:3'), covar=tensor([0.0143, 0.2471, 0.0224, 0.2113, 0.0282, 0.0284, 0.0303, 0.0108], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0155, 0.0127, 0.0164, 0.0143, 0.0135, 0.0114, 0.0114], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 11:25:22,820 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28418.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 11:25:23,566 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.172e+02 2.835e+02 3.398e+02 4.455e+02 8.619e+02, threshold=6.796e+02, percent-clipped=2.0 2022-12-07 11:25:34,792 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 2022-12-07 11:25:55,398 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.36 vs. limit=2.0 2022-12-07 11:26:22,883 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28486.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:26:22,934 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.6450, 1.9130, 2.5191, 2.1853, 2.6400, 2.4578, 2.4884, 2.2132], device='cuda:3'), covar=tensor([0.0289, 0.1825, 0.0526, 0.1181, 0.0448, 0.0469, 0.0596, 0.1170], device='cuda:3'), in_proj_covar=tensor([0.0257, 0.0346, 0.0314, 0.0313, 0.0317, 0.0267, 0.0307, 0.0361], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-07 11:26:23,534 INFO [train.py:873] (3/4) Epoch 4, batch 5800, loss[loss=0.2, simple_loss=0.2018, pruned_loss=0.09913, over 6906.00 frames. ], tot_loss[loss=0.1898, simple_loss=0.1993, pruned_loss=0.09012, over 1984148.21 frames. ], batch size: 100, lr: 1.81e-02, grad_scale: 8.0 2022-12-07 11:26:27,938 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28492.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:26:52,116 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.461e+02 2.705e+02 3.370e+02 4.199e+02 7.683e+02, threshold=6.740e+02, percent-clipped=3.0 2022-12-07 11:27:16,541 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28547.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:27:21,797 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28553.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:27:49,622 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28585.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:27:51,187 INFO [train.py:873] (3/4) Epoch 4, batch 5900, loss[loss=0.2064, simple_loss=0.1727, pruned_loss=0.12, over 1314.00 frames. ], tot_loss[loss=0.1883, simple_loss=0.1981, pruned_loss=0.08928, over 1973257.02 frames. ], batch size: 100, lr: 1.81e-02, grad_scale: 8.0 2022-12-07 11:28:02,200 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.3685, 1.1386, 0.8727, 0.9593, 1.3655, 0.9335, 1.0919, 1.3005], device='cuda:3'), covar=tensor([0.0257, 0.0246, 0.0199, 0.0277, 0.0220, 0.0182, 0.0256, 0.0299], device='cuda:3'), in_proj_covar=tensor([0.0014, 0.0017, 0.0018, 0.0015, 0.0016, 0.0021, 0.0016, 0.0016], device='cuda:3'), out_proj_covar=tensor([5.4380e-05, 5.9333e-05, 5.8745e-05, 5.7470e-05, 5.5432e-05, 7.3302e-05, 6.2281e-05, 5.7679e-05], device='cuda:3') 2022-12-07 11:28:13,036 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28611.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 11:28:19,509 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.061e+02 2.929e+02 3.624e+02 4.382e+02 1.301e+03, threshold=7.248e+02, percent-clipped=4.0 2022-12-07 11:28:40,776 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28643.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:28:56,091 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.85 vs. limit=5.0 2022-12-07 11:29:09,042 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9695, 3.6457, 4.0877, 3.5065, 3.9031, 4.0721, 1.3716, 3.6860], device='cuda:3'), covar=tensor([0.0189, 0.0303, 0.0330, 0.0529, 0.0251, 0.0252, 0.2982, 0.0251], device='cuda:3'), in_proj_covar=tensor([0.0118, 0.0128, 0.0117, 0.0101, 0.0157, 0.0112, 0.0146, 0.0152], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 11:29:19,381 INFO [train.py:873] (3/4) Epoch 4, batch 6000, loss[loss=0.1856, simple_loss=0.1713, pruned_loss=0.09994, over 3868.00 frames. ], tot_loss[loss=0.1882, simple_loss=0.198, pruned_loss=0.08921, over 1988556.24 frames. ], batch size: 100, lr: 1.81e-02, grad_scale: 8.0 2022-12-07 11:29:19,381 INFO [train.py:896] (3/4) Computing validation loss 2022-12-07 11:29:30,470 INFO [train.py:905] (3/4) Epoch 4, validation: loss=0.1258, simple_loss=0.1688, pruned_loss=0.04138, over 857387.00 frames. 2022-12-07 11:29:30,471 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17841MB 2022-12-07 11:29:58,583 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.235e+02 2.688e+02 3.368e+02 4.685e+02 1.249e+03, threshold=6.736e+02, percent-clipped=4.0 2022-12-07 11:30:58,676 INFO [train.py:873] (3/4) Epoch 4, batch 6100, loss[loss=0.2035, simple_loss=0.2066, pruned_loss=0.1002, over 13530.00 frames. ], tot_loss[loss=0.1889, simple_loss=0.1983, pruned_loss=0.08977, over 1969372.22 frames. ], batch size: 100, lr: 1.80e-02, grad_scale: 8.0 2022-12-07 11:31:27,398 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.104e+02 2.717e+02 3.239e+02 4.102e+02 7.823e+02, threshold=6.479e+02, percent-clipped=2.0 2022-12-07 11:31:47,361 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28842.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:31:52,799 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28848.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:32:04,640 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28861.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:32:11,849 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4670, 3.2588, 2.7973, 1.9122, 2.6427, 2.9810, 3.2892, 2.5204], device='cuda:3'), covar=tensor([0.0511, 0.2602, 0.1307, 0.3003, 0.0859, 0.0523, 0.0742, 0.1597], device='cuda:3'), in_proj_covar=tensor([0.0095, 0.0223, 0.0112, 0.0130, 0.0098, 0.0099, 0.0083, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0006, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-07 11:32:25,334 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28885.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:32:27,172 INFO [train.py:873] (3/4) Epoch 4, batch 6200, loss[loss=0.1974, simple_loss=0.2007, pruned_loss=0.09712, over 14268.00 frames. ], tot_loss[loss=0.1878, simple_loss=0.1978, pruned_loss=0.08895, over 1997849.69 frames. ], batch size: 89, lr: 1.80e-02, grad_scale: 8.0 2022-12-07 11:32:45,061 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9201, 2.7336, 4.7363, 3.2440, 4.7472, 2.1021, 3.7596, 4.3976], device='cuda:3'), covar=tensor([0.0236, 0.3955, 0.0298, 0.8800, 0.0154, 0.3655, 0.0993, 0.0187], device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0275, 0.0169, 0.0370, 0.0174, 0.0282, 0.0261, 0.0171], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 11:32:48,587 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28911.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 11:32:55,225 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.253e+02 2.759e+02 3.767e+02 5.172e+02 1.797e+03, threshold=7.535e+02, percent-clipped=14.0 2022-12-07 11:32:57,818 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28922.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:33:07,816 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28933.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:33:16,552 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28943.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:33:30,099 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28959.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 11:33:55,197 INFO [train.py:873] (3/4) Epoch 4, batch 6300, loss[loss=0.2189, simple_loss=0.2036, pruned_loss=0.117, over 4929.00 frames. ], tot_loss[loss=0.1885, simple_loss=0.198, pruned_loss=0.08953, over 1946522.72 frames. ], batch size: 100, lr: 1.80e-02, grad_scale: 8.0 2022-12-07 11:33:58,984 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28991.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:34:08,944 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29002.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:34:23,887 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.333e+02 2.322e+02 3.000e+02 3.617e+02 7.085e+02, threshold=6.000e+02, percent-clipped=0.0 2022-12-07 11:34:33,369 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1965, 1.0223, 0.7594, 1.0539, 0.9999, 0.8246, 1.0651, 1.2254], device='cuda:3'), covar=tensor([0.0593, 0.0673, 0.0349, 0.0484, 0.0512, 0.0264, 0.0470, 0.0251], device='cuda:3'), in_proj_covar=tensor([0.0015, 0.0018, 0.0018, 0.0016, 0.0016, 0.0022, 0.0017, 0.0016], device='cuda:3'), out_proj_covar=tensor([5.8185e-05, 6.4385e-05, 5.9615e-05, 6.0320e-05, 5.7132e-05, 7.7122e-05, 6.3996e-05, 5.7949e-05], device='cuda:3') 2022-12-07 11:34:46,521 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.02 vs. limit=2.0 2022-12-07 11:35:03,257 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29063.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:35:19,362 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.56 vs. limit=5.0 2022-12-07 11:35:24,088 INFO [train.py:873] (3/4) Epoch 4, batch 6400, loss[loss=0.1818, simple_loss=0.1817, pruned_loss=0.09093, over 4936.00 frames. ], tot_loss[loss=0.1878, simple_loss=0.1978, pruned_loss=0.08894, over 1981434.58 frames. ], batch size: 100, lr: 1.79e-02, grad_scale: 8.0 2022-12-07 11:35:46,744 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8172, 1.5146, 2.0156, 1.7359, 1.9946, 1.4186, 1.6776, 1.7840], device='cuda:3'), covar=tensor([0.0697, 0.1685, 0.0157, 0.1005, 0.0202, 0.0802, 0.0759, 0.0251], device='cuda:3'), in_proj_covar=tensor([0.0222, 0.0272, 0.0169, 0.0372, 0.0175, 0.0286, 0.0258, 0.0173], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 11:35:52,467 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.074e+02 2.620e+02 3.469e+02 4.373e+02 1.002e+03, threshold=6.937e+02, percent-clipped=9.0 2022-12-07 11:36:12,703 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29142.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:36:16,998 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5134, 1.4560, 2.6633, 1.3859, 2.7430, 2.6540, 1.9733, 2.7786], device='cuda:3'), covar=tensor([0.0194, 0.1970, 0.0232, 0.1523, 0.0256, 0.0295, 0.0607, 0.0167], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0156, 0.0127, 0.0164, 0.0144, 0.0137, 0.0114, 0.0114], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 11:36:17,942 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29148.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:36:39,260 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4801, 1.3996, 2.6623, 1.2739, 2.7265, 2.6558, 1.8360, 2.7904], device='cuda:3'), covar=tensor([0.0224, 0.2036, 0.0223, 0.1702, 0.0262, 0.0299, 0.0698, 0.0161], device='cuda:3'), in_proj_covar=tensor([0.0140, 0.0156, 0.0126, 0.0163, 0.0143, 0.0136, 0.0114, 0.0114], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 11:36:52,305 INFO [train.py:873] (3/4) Epoch 4, batch 6500, loss[loss=0.176, simple_loss=0.1587, pruned_loss=0.09664, over 2670.00 frames. ], tot_loss[loss=0.189, simple_loss=0.1985, pruned_loss=0.08973, over 1949720.54 frames. ], batch size: 100, lr: 1.79e-02, grad_scale: 8.0 2022-12-07 11:36:54,886 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29190.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:37:00,040 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29196.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:37:18,817 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29217.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:37:20,462 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.481e+02 2.712e+02 3.572e+02 4.670e+02 1.139e+03, threshold=7.145e+02, percent-clipped=7.0 2022-12-07 11:37:35,590 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29236.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:37:52,369 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.2745, 1.5416, 4.1325, 1.7046, 3.9622, 4.0315, 3.3209, 4.5556], device='cuda:3'), covar=tensor([0.0146, 0.2652, 0.0243, 0.2114, 0.0258, 0.0281, 0.0412, 0.0112], device='cuda:3'), in_proj_covar=tensor([0.0140, 0.0157, 0.0127, 0.0165, 0.0143, 0.0136, 0.0115, 0.0115], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 11:37:54,098 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4041, 1.1280, 1.4128, 0.7398, 1.1897, 1.1892, 1.2430, 1.2308], device='cuda:3'), covar=tensor([0.0218, 0.0562, 0.0348, 0.0276, 0.0360, 0.0434, 0.0156, 0.0544], device='cuda:3'), in_proj_covar=tensor([0.0095, 0.0222, 0.0110, 0.0128, 0.0095, 0.0096, 0.0083, 0.0120], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0006, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-07 11:37:58,248 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.97 vs. limit=5.0 2022-12-07 11:38:19,446 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5668, 2.5928, 3.3835, 2.1447, 2.2260, 2.4476, 1.5231, 2.4630], device='cuda:3'), covar=tensor([0.1157, 0.0820, 0.0400, 0.1957, 0.1982, 0.1038, 0.4910, 0.1614], device='cuda:3'), in_proj_covar=tensor([0.0070, 0.0074, 0.0071, 0.0084, 0.0101, 0.0063, 0.0139, 0.0072], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 11:38:20,130 INFO [train.py:873] (3/4) Epoch 4, batch 6600, loss[loss=0.1924, simple_loss=0.208, pruned_loss=0.08844, over 14375.00 frames. ], tot_loss[loss=0.1897, simple_loss=0.199, pruned_loss=0.09023, over 1944562.66 frames. ], batch size: 73, lr: 1.79e-02, grad_scale: 8.0 2022-12-07 11:38:29,475 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29297.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:38:48,061 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.105e+02 2.617e+02 3.366e+02 4.262e+02 9.069e+02, threshold=6.731e+02, percent-clipped=3.0 2022-12-07 11:39:22,480 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29358.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:39:32,649 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.2596, 2.4331, 4.2680, 2.9815, 4.1246, 1.9284, 3.0360, 3.9616], device='cuda:3'), covar=tensor([0.0249, 0.4528, 0.0293, 0.8399, 0.0266, 0.4262, 0.1418, 0.0193], device='cuda:3'), in_proj_covar=tensor([0.0226, 0.0277, 0.0170, 0.0371, 0.0179, 0.0282, 0.0262, 0.0177], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 11:39:47,260 INFO [train.py:873] (3/4) Epoch 4, batch 6700, loss[loss=0.168, simple_loss=0.1838, pruned_loss=0.07615, over 13881.00 frames. ], tot_loss[loss=0.1885, simple_loss=0.1985, pruned_loss=0.08927, over 2019577.54 frames. ], batch size: 23, lr: 1.78e-02, grad_scale: 8.0 2022-12-07 11:39:48,613 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 2022-12-07 11:40:14,795 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.599e+02 2.696e+02 3.507e+02 4.703e+02 1.037e+03, threshold=7.015e+02, percent-clipped=5.0 2022-12-07 11:40:16,647 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.8016, 5.2426, 5.2142, 5.7188, 5.5027, 4.7598, 5.7809, 4.9480], device='cuda:3'), covar=tensor([0.0267, 0.0759, 0.0232, 0.0368, 0.0562, 0.0242, 0.0319, 0.0322], device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0206, 0.0138, 0.0130, 0.0139, 0.0109, 0.0198, 0.0135], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 11:40:22,018 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2253, 2.9016, 2.9176, 3.1769, 3.1387, 3.1991, 3.2467, 2.7448], device='cuda:3'), covar=tensor([0.0387, 0.1057, 0.0471, 0.0481, 0.0686, 0.0337, 0.0604, 0.0553], device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0207, 0.0138, 0.0130, 0.0140, 0.0109, 0.0199, 0.0135], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 11:40:45,084 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2931, 3.0789, 3.0241, 3.4028, 2.9459, 2.7860, 3.3738, 3.3541], device='cuda:3'), covar=tensor([0.0717, 0.0682, 0.0761, 0.0619, 0.0866, 0.0720, 0.0634, 0.0690], device='cuda:3'), in_proj_covar=tensor([0.0109, 0.0088, 0.0108, 0.0107, 0.0115, 0.0084, 0.0118, 0.0106], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 11:40:50,996 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=16.85 vs. limit=5.0 2022-12-07 11:41:13,502 INFO [train.py:873] (3/4) Epoch 4, batch 6800, loss[loss=0.2089, simple_loss=0.2003, pruned_loss=0.1087, over 4942.00 frames. ], tot_loss[loss=0.1901, simple_loss=0.1993, pruned_loss=0.09039, over 1999880.91 frames. ], batch size: 100, lr: 1.78e-02, grad_scale: 8.0 2022-12-07 11:41:27,870 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7802, 1.8683, 3.6962, 2.4002, 3.5833, 1.6873, 2.5086, 3.3616], device='cuda:3'), covar=tensor([0.0556, 0.6777, 0.0575, 1.1758, 0.0500, 0.6022, 0.2147, 0.0391], device='cuda:3'), in_proj_covar=tensor([0.0226, 0.0282, 0.0170, 0.0376, 0.0181, 0.0285, 0.0264, 0.0181], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 11:41:39,792 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29517.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:41:42,141 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.212e+02 2.836e+02 3.618e+02 4.474e+02 6.792e+02, threshold=7.235e+02, percent-clipped=0.0 2022-12-07 11:42:04,808 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.99 vs. limit=2.0 2022-12-07 11:42:05,224 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7778, 2.6414, 2.3761, 2.4821, 2.6979, 2.6176, 2.7587, 2.7142], device='cuda:3'), covar=tensor([0.0724, 0.0745, 0.1833, 0.2421, 0.0795, 0.0848, 0.1145, 0.0821], device='cuda:3'), in_proj_covar=tensor([0.0249, 0.0215, 0.0320, 0.0409, 0.0242, 0.0285, 0.0299, 0.0238], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0005, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2022-12-07 11:42:09,645 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29551.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:42:21,757 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29565.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:42:41,253 INFO [train.py:873] (3/4) Epoch 4, batch 6900, loss[loss=0.1959, simple_loss=0.2033, pruned_loss=0.09425, over 14403.00 frames. ], tot_loss[loss=0.1893, simple_loss=0.1985, pruned_loss=0.09007, over 1959457.71 frames. ], batch size: 53, lr: 1.78e-02, grad_scale: 8.0 2022-12-07 11:42:45,777 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29592.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:43:03,011 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29612.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:43:09,418 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.448e+02 2.737e+02 3.624e+02 4.962e+02 1.279e+03, threshold=7.248e+02, percent-clipped=6.0 2022-12-07 11:43:24,276 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2022-12-07 11:43:40,060 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1547, 1.6157, 1.6899, 1.5617, 1.4581, 1.6416, 1.3381, 0.9162], device='cuda:3'), covar=tensor([0.2352, 0.0702, 0.0312, 0.0364, 0.0813, 0.0265, 0.1104, 0.1942], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0058, 0.0050, 0.0050, 0.0067, 0.0052, 0.0079, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0004], device='cuda:3') 2022-12-07 11:43:42,690 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29658.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:43:45,781 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29662.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:44:04,639 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9725, 1.8831, 2.0019, 2.0386, 2.0180, 1.7553, 1.0518, 1.7631], device='cuda:3'), covar=tensor([0.0324, 0.0327, 0.0442, 0.0241, 0.0314, 0.0812, 0.1919, 0.0341], device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0134, 0.0122, 0.0105, 0.0163, 0.0119, 0.0153, 0.0157], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 11:44:07,055 INFO [train.py:873] (3/4) Epoch 4, batch 7000, loss[loss=0.2047, simple_loss=0.1763, pruned_loss=0.1166, over 1302.00 frames. ], tot_loss[loss=0.1907, simple_loss=0.1992, pruned_loss=0.0911, over 1944219.96 frames. ], batch size: 100, lr: 1.78e-02, grad_scale: 8.0 2022-12-07 11:44:23,787 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29706.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:44:35,728 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.520e+02 2.834e+02 3.387e+02 4.469e+02 7.718e+02, threshold=6.775e+02, percent-clipped=1.0 2022-12-07 11:44:38,292 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29723.0, num_to_drop=1, layers_to_drop={3} 2022-12-07 11:44:38,531 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.00 vs. limit=2.0 2022-12-07 11:44:43,677 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.84 vs. limit=5.0 2022-12-07 11:44:55,525 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5315, 2.5019, 2.0708, 1.9852, 2.5776, 2.5528, 2.5736, 2.5209], device='cuda:3'), covar=tensor([0.1398, 0.1297, 0.2967, 0.4776, 0.1097, 0.1189, 0.1765, 0.1383], device='cuda:3'), in_proj_covar=tensor([0.0256, 0.0215, 0.0320, 0.0414, 0.0247, 0.0287, 0.0300, 0.0245], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0005, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2022-12-07 11:45:33,333 INFO [train.py:873] (3/4) Epoch 4, batch 7100, loss[loss=0.1834, simple_loss=0.2088, pruned_loss=0.07901, over 14487.00 frames. ], tot_loss[loss=0.1876, simple_loss=0.1973, pruned_loss=0.08893, over 1918975.81 frames. ], batch size: 49, lr: 1.77e-02, grad_scale: 8.0 2022-12-07 11:45:46,790 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29802.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:46:02,683 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.326e+02 2.583e+02 3.436e+02 4.712e+02 1.166e+03, threshold=6.872e+02, percent-clipped=4.0 2022-12-07 11:46:04,826 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.32 vs. limit=5.0 2022-12-07 11:46:40,720 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29863.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:47:01,596 INFO [train.py:873] (3/4) Epoch 4, batch 7200, loss[loss=0.2202, simple_loss=0.2182, pruned_loss=0.1111, over 12745.00 frames. ], tot_loss[loss=0.1883, simple_loss=0.1978, pruned_loss=0.08942, over 1954411.62 frames. ], batch size: 100, lr: 1.77e-02, grad_scale: 8.0 2022-12-07 11:47:06,080 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29892.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:47:08,824 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([6.0463, 5.5000, 5.4632, 5.9305, 5.4806, 5.0367, 6.0655, 6.0075], device='cuda:3'), covar=tensor([0.0426, 0.0442, 0.0429, 0.0383, 0.0453, 0.0255, 0.0416, 0.0369], device='cuda:3'), in_proj_covar=tensor([0.0108, 0.0089, 0.0106, 0.0107, 0.0114, 0.0086, 0.0115, 0.0104], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 11:47:09,798 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4620, 4.3242, 4.7370, 4.0730, 4.4413, 4.8600, 1.7371, 4.2286], device='cuda:3'), covar=tensor([0.0168, 0.0237, 0.0316, 0.0472, 0.0302, 0.0132, 0.2869, 0.0225], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0130, 0.0120, 0.0104, 0.0160, 0.0112, 0.0148, 0.0153], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 11:47:14,386 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.6161, 2.6307, 4.5713, 2.9800, 4.3683, 2.1409, 3.2099, 4.0756], device='cuda:3'), covar=tensor([0.0244, 0.4737, 0.0400, 1.0079, 0.0213, 0.4168, 0.1483, 0.0286], device='cuda:3'), in_proj_covar=tensor([0.0230, 0.0282, 0.0169, 0.0372, 0.0178, 0.0286, 0.0261, 0.0174], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 11:47:19,546 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29907.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:47:23,648 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4408, 1.6668, 1.6295, 1.7010, 1.5642, 1.7671, 1.4495, 0.9616], device='cuda:3'), covar=tensor([0.2330, 0.0941, 0.0952, 0.0504, 0.1051, 0.0383, 0.1707, 0.3156], device='cuda:3'), in_proj_covar=tensor([0.0154, 0.0060, 0.0053, 0.0051, 0.0070, 0.0055, 0.0081, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0004], device='cuda:3') 2022-12-07 11:47:24,009 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.97 vs. limit=2.0 2022-12-07 11:47:30,663 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.357e+02 2.626e+02 3.344e+02 4.272e+02 1.259e+03, threshold=6.687e+02, percent-clipped=4.0 2022-12-07 11:47:47,977 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29940.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:48:21,756 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 2022-12-07 11:48:28,906 INFO [train.py:873] (3/4) Epoch 4, batch 7300, loss[loss=0.1876, simple_loss=0.1916, pruned_loss=0.09176, over 6908.00 frames. ], tot_loss[loss=0.1868, simple_loss=0.1971, pruned_loss=0.08822, over 1982988.25 frames. ], batch size: 100, lr: 1.77e-02, grad_scale: 8.0 2022-12-07 11:48:59,755 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30018.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 11:49:01,143 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.428e+02 2.508e+02 3.092e+02 3.972e+02 1.080e+03, threshold=6.184e+02, percent-clipped=3.0 2022-12-07 11:49:18,755 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8523, 2.5648, 2.5955, 1.5539, 2.4190, 2.4084, 2.8249, 2.3309], device='cuda:3'), covar=tensor([0.0684, 0.1944, 0.1123, 0.2730, 0.0850, 0.0478, 0.0715, 0.1746], device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0215, 0.0112, 0.0125, 0.0095, 0.0096, 0.0083, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0006, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-07 11:49:32,631 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=12.34 vs. limit=5.0 2022-12-07 11:49:59,921 INFO [train.py:873] (3/4) Epoch 4, batch 7400, loss[loss=0.1701, simple_loss=0.198, pruned_loss=0.07108, over 14439.00 frames. ], tot_loss[loss=0.1869, simple_loss=0.1975, pruned_loss=0.08818, over 1997223.73 frames. ], batch size: 53, lr: 1.76e-02, grad_scale: 8.0 2022-12-07 11:50:29,607 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.686e+02 2.725e+02 3.785e+02 4.899e+02 8.566e+02, threshold=7.571e+02, percent-clipped=11.0 2022-12-07 11:50:33,926 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30125.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:51:02,892 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30158.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:51:27,349 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30186.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:51:28,006 INFO [train.py:873] (3/4) Epoch 4, batch 7500, loss[loss=0.1626, simple_loss=0.1876, pruned_loss=0.06877, over 14248.00 frames. ], tot_loss[loss=0.1864, simple_loss=0.1971, pruned_loss=0.08791, over 2016148.82 frames. ], batch size: 57, lr: 1.76e-02, grad_scale: 8.0 2022-12-07 11:51:46,028 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30207.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:51:47,613 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30209.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:51:56,539 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.420e+02 2.609e+02 3.309e+02 4.020e+02 7.663e+02, threshold=6.618e+02, percent-clipped=1.0 2022-12-07 11:52:55,166 INFO [train.py:873] (3/4) Epoch 5, batch 0, loss[loss=0.254, simple_loss=0.2522, pruned_loss=0.1279, over 13973.00 frames. ], tot_loss[loss=0.254, simple_loss=0.2522, pruned_loss=0.1279, over 13973.00 frames. ], batch size: 19, lr: 1.64e-02, grad_scale: 8.0 2022-12-07 11:52:55,167 INFO [train.py:896] (3/4) Computing validation loss 2022-12-07 11:53:02,316 INFO [train.py:905] (3/4) Epoch 5, validation: loss=0.1366, simple_loss=0.1807, pruned_loss=0.04626, over 857387.00 frames. 2022-12-07 11:53:02,317 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17841MB 2022-12-07 11:53:07,523 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30255.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:53:21,389 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30270.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:54:03,618 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30318.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 11:54:05,438 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 5.465e+01 2.449e+02 3.602e+02 5.226e+02 1.817e+03, threshold=7.204e+02, percent-clipped=11.0 2022-12-07 11:54:25,548 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2022-12-07 11:54:31,492 INFO [train.py:873] (3/4) Epoch 5, batch 100, loss[loss=0.1622, simple_loss=0.1887, pruned_loss=0.06786, over 14612.00 frames. ], tot_loss[loss=0.1887, simple_loss=0.1999, pruned_loss=0.08876, over 968661.24 frames. ], batch size: 22, lr: 1.64e-02, grad_scale: 8.0 2022-12-07 11:54:40,829 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30360.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:54:45,804 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30366.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:55:32,796 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.580e+02 2.778e+02 3.417e+02 4.208e+02 9.792e+02, threshold=6.833e+02, percent-clipped=3.0 2022-12-07 11:55:33,885 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30421.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:55:43,665 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4947, 2.8086, 4.2363, 2.8777, 4.0653, 4.0962, 3.7630, 3.4694], device='cuda:3'), covar=tensor([0.0322, 0.3033, 0.0729, 0.2020, 0.0727, 0.0482, 0.2436, 0.2094], device='cuda:3'), in_proj_covar=tensor([0.0274, 0.0352, 0.0337, 0.0318, 0.0336, 0.0271, 0.0326, 0.0368], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-07 11:55:58,009 INFO [train.py:873] (3/4) Epoch 5, batch 200, loss[loss=0.2177, simple_loss=0.189, pruned_loss=0.1232, over 2599.00 frames. ], tot_loss[loss=0.1875, simple_loss=0.1973, pruned_loss=0.08882, over 1302501.46 frames. ], batch size: 100, lr: 1.63e-02, grad_scale: 8.0 2022-12-07 11:56:06,130 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30458.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:56:12,132 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.49 vs. limit=5.0 2022-12-07 11:56:26,420 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30481.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:56:32,995 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2022-12-07 11:56:48,238 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30506.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:57:00,757 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.239e+02 2.564e+02 3.377e+02 4.486e+02 9.729e+02, threshold=6.754e+02, percent-clipped=3.0 2022-12-07 11:57:25,912 INFO [train.py:873] (3/4) Epoch 5, batch 300, loss[loss=0.1909, simple_loss=0.1967, pruned_loss=0.0925, over 14244.00 frames. ], tot_loss[loss=0.1869, simple_loss=0.1968, pruned_loss=0.08847, over 1550237.44 frames. ], batch size: 80, lr: 1.63e-02, grad_scale: 8.0 2022-12-07 11:57:39,910 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30565.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:58:27,948 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.393e+02 2.626e+02 3.332e+02 4.486e+02 9.530e+02, threshold=6.664e+02, percent-clipped=3.0 2022-12-07 11:58:53,707 INFO [train.py:873] (3/4) Epoch 5, batch 400, loss[loss=0.2151, simple_loss=0.1828, pruned_loss=0.1237, over 1159.00 frames. ], tot_loss[loss=0.185, simple_loss=0.1958, pruned_loss=0.08709, over 1661372.97 frames. ], batch size: 100, lr: 1.63e-02, grad_scale: 8.0 2022-12-07 11:58:57,554 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=10.52 vs. limit=5.0 2022-12-07 11:58:58,950 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30655.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:59:52,632 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30716.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 11:59:52,747 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30716.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 11:59:56,041 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.257e+02 2.557e+02 3.010e+02 3.908e+02 7.694e+02, threshold=6.020e+02, percent-clipped=1.0 2022-12-07 12:00:17,566 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5346, 4.2206, 4.1888, 4.5719, 4.4535, 4.0494, 4.6304, 3.7915], device='cuda:3'), covar=tensor([0.0359, 0.0928, 0.0331, 0.0381, 0.0600, 0.0583, 0.0429, 0.0540], device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0202, 0.0139, 0.0130, 0.0138, 0.0107, 0.0198, 0.0134], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 12:00:21,155 INFO [train.py:873] (3/4) Epoch 5, batch 500, loss[loss=0.1761, simple_loss=0.1533, pruned_loss=0.09941, over 1199.00 frames. ], tot_loss[loss=0.184, simple_loss=0.1954, pruned_loss=0.08634, over 1756976.39 frames. ], batch size: 100, lr: 1.63e-02, grad_scale: 8.0 2022-12-07 12:00:24,973 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2022-12-07 12:00:49,080 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30781.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:01:23,929 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.147e+02 2.567e+02 3.338e+02 3.998e+02 8.568e+02, threshold=6.676e+02, percent-clipped=3.0 2022-12-07 12:01:31,371 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30829.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:01:48,857 INFO [train.py:873] (3/4) Epoch 5, batch 600, loss[loss=0.181, simple_loss=0.1982, pruned_loss=0.08187, over 14471.00 frames. ], tot_loss[loss=0.1843, simple_loss=0.1955, pruned_loss=0.08657, over 1864157.43 frames. ], batch size: 24, lr: 1.62e-02, grad_scale: 4.0 2022-12-07 12:02:02,894 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30865.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:02:36,294 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30903.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:02:45,025 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30913.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:02:51,899 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.252e+02 2.564e+02 3.109e+02 3.998e+02 8.746e+02, threshold=6.217e+02, percent-clipped=4.0 2022-12-07 12:02:55,483 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30925.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:03:03,318 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30934.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:03:16,252 INFO [train.py:873] (3/4) Epoch 5, batch 700, loss[loss=0.1857, simple_loss=0.1913, pruned_loss=0.09004, over 14194.00 frames. ], tot_loss[loss=0.1834, simple_loss=0.1942, pruned_loss=0.08628, over 1853724.64 frames. ], batch size: 99, lr: 1.62e-02, grad_scale: 4.0 2022-12-07 12:03:26,725 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.01 vs. limit=2.0 2022-12-07 12:03:29,739 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30964.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:03:49,048 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30986.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:03:56,483 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30995.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:04:10,464 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31011.0, num_to_drop=1, layers_to_drop={3} 2022-12-07 12:04:10,545 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31011.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 12:04:14,730 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31016.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:04:18,898 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.713e+02 2.586e+02 3.335e+02 4.382e+02 1.879e+03, threshold=6.670e+02, percent-clipped=6.0 2022-12-07 12:04:43,828 INFO [train.py:873] (3/4) Epoch 5, batch 800, loss[loss=0.1669, simple_loss=0.1848, pruned_loss=0.07449, over 14538.00 frames. ], tot_loss[loss=0.1848, simple_loss=0.1946, pruned_loss=0.0875, over 1867807.33 frames. ], batch size: 34, lr: 1.62e-02, grad_scale: 8.0 2022-12-07 12:04:57,135 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31064.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:05:04,541 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31072.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 12:05:47,140 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.226e+02 2.770e+02 3.066e+02 3.847e+02 6.961e+02, threshold=6.133e+02, percent-clipped=2.0 2022-12-07 12:06:11,660 INFO [train.py:873] (3/4) Epoch 5, batch 900, loss[loss=0.166, simple_loss=0.1882, pruned_loss=0.07192, over 14237.00 frames. ], tot_loss[loss=0.1858, simple_loss=0.1958, pruned_loss=0.08786, over 1913838.36 frames. ], batch size: 46, lr: 1.62e-02, grad_scale: 8.0 2022-12-07 12:06:58,283 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.43 vs. limit=5.0 2022-12-07 12:07:14,863 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.620e+02 2.339e+02 3.126e+02 3.777e+02 6.609e+02, threshold=6.252e+02, percent-clipped=2.0 2022-12-07 12:07:24,700 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1758, 1.1037, 1.0146, 1.0201, 0.6525, 0.7347, 0.8278, 0.7791], device='cuda:3'), covar=tensor([0.0345, 0.0299, 0.0452, 0.0357, 0.0536, 0.0428, 0.0250, 0.0529], device='cuda:3'), in_proj_covar=tensor([0.0011, 0.0011, 0.0010, 0.0010, 0.0012, 0.0014, 0.0012, 0.0016], device='cuda:3'), out_proj_covar=tensor([5.1175e-05, 5.2813e-05, 5.0853e-05, 4.8284e-05, 5.4910e-05, 6.7511e-05, 6.1021e-05, 7.1569e-05], device='cuda:3') 2022-12-07 12:07:35,606 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4812, 4.1408, 4.1114, 4.5535, 4.0908, 3.5542, 4.5689, 4.4397], device='cuda:3'), covar=tensor([0.0594, 0.0500, 0.0575, 0.0434, 0.0537, 0.0590, 0.0490, 0.0487], device='cuda:3'), in_proj_covar=tensor([0.0108, 0.0089, 0.0106, 0.0105, 0.0115, 0.0084, 0.0119, 0.0107], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 12:07:39,741 INFO [train.py:873] (3/4) Epoch 5, batch 1000, loss[loss=0.2297, simple_loss=0.1852, pruned_loss=0.1371, over 1198.00 frames. ], tot_loss[loss=0.1845, simple_loss=0.195, pruned_loss=0.087, over 1908308.20 frames. ], batch size: 100, lr: 1.61e-02, grad_scale: 8.0 2022-12-07 12:07:39,900 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1007, 2.0308, 1.5314, 2.0422, 1.7349, 2.0369, 1.8427, 1.7428], device='cuda:3'), covar=tensor([0.0254, 0.0485, 0.1339, 0.0216, 0.0512, 0.0224, 0.0829, 0.0653], device='cuda:3'), in_proj_covar=tensor([0.0228, 0.0300, 0.0315, 0.0183, 0.0242, 0.0242, 0.0262, 0.0304], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 12:07:46,642 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.1374, 4.8250, 4.6560, 5.1621, 4.6505, 4.1698, 5.2169, 5.0089], device='cuda:3'), covar=tensor([0.0746, 0.0552, 0.0667, 0.0583, 0.0658, 0.0501, 0.0649, 0.0771], device='cuda:3'), in_proj_covar=tensor([0.0110, 0.0091, 0.0108, 0.0106, 0.0117, 0.0085, 0.0121, 0.0109], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 12:07:48,346 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31259.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:08:07,802 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31281.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:08:15,271 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31290.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:08:27,873 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2022-12-07 12:08:33,846 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31311.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:08:42,295 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.584e+02 2.699e+02 3.594e+02 4.260e+02 1.045e+03, threshold=7.189e+02, percent-clipped=6.0 2022-12-07 12:09:06,866 INFO [train.py:873] (3/4) Epoch 5, batch 1100, loss[loss=0.1574, simple_loss=0.1799, pruned_loss=0.06747, over 13915.00 frames. ], tot_loss[loss=0.184, simple_loss=0.1945, pruned_loss=0.08672, over 1939328.89 frames. ], batch size: 20, lr: 1.61e-02, grad_scale: 8.0 2022-12-07 12:09:15,219 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31359.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:09:22,312 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31367.0, num_to_drop=1, layers_to_drop={3} 2022-12-07 12:09:29,068 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2763, 1.6904, 2.4207, 2.0424, 2.4155, 1.8720, 1.8920, 2.0319], device='cuda:3'), covar=tensor([0.1024, 0.3167, 0.0195, 0.2139, 0.0332, 0.2126, 0.1180, 0.0577], device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0278, 0.0169, 0.0368, 0.0178, 0.0279, 0.0258, 0.0182], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 12:09:59,559 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31409.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:10:00,396 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8148, 0.6942, 0.8424, 0.7920, 0.7032, 0.5060, 0.7147, 0.5708], device='cuda:3'), covar=tensor([0.0087, 0.0093, 0.0110, 0.0085, 0.0196, 0.0253, 0.0096, 0.0220], device='cuda:3'), in_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0010, 0.0012, 0.0015, 0.0012, 0.0017], device='cuda:3'), out_proj_covar=tensor([5.4990e-05, 5.5548e-05, 5.4808e-05, 4.9794e-05, 5.7767e-05, 7.1315e-05, 6.4380e-05, 7.5467e-05], device='cuda:3') 2022-12-07 12:10:02,801 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.3429, 5.0887, 4.7187, 4.8283, 4.7739, 5.2367, 5.3310, 5.3985], device='cuda:3'), covar=tensor([0.0702, 0.0451, 0.1706, 0.2526, 0.0643, 0.0565, 0.0847, 0.0612], device='cuda:3'), in_proj_covar=tensor([0.0266, 0.0221, 0.0332, 0.0430, 0.0254, 0.0298, 0.0308, 0.0255], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 12:10:09,481 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.418e+02 2.558e+02 3.355e+02 4.456e+02 1.089e+03, threshold=6.709e+02, percent-clipped=2.0 2022-12-07 12:10:34,139 INFO [train.py:873] (3/4) Epoch 5, batch 1200, loss[loss=0.1466, simple_loss=0.1392, pruned_loss=0.07703, over 2648.00 frames. ], tot_loss[loss=0.184, simple_loss=0.1951, pruned_loss=0.08649, over 1982609.66 frames. ], batch size: 100, lr: 1.61e-02, grad_scale: 8.0 2022-12-07 12:10:38,520 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9360, 4.8346, 5.2335, 4.4757, 4.9778, 5.3372, 1.8543, 4.7758], device='cuda:3'), covar=tensor([0.0148, 0.0196, 0.0298, 0.0424, 0.0213, 0.0122, 0.3006, 0.0212], device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0130, 0.0121, 0.0106, 0.0164, 0.0115, 0.0149, 0.0154], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 12:10:53,079 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31470.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:11:10,747 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31491.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 12:11:37,433 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.031e+02 2.496e+02 3.332e+02 4.341e+02 8.542e+02, threshold=6.664e+02, percent-clipped=7.0 2022-12-07 12:11:47,111 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.9872, 0.9161, 1.0258, 0.9154, 0.7464, 0.6821, 0.9436, 0.5202], device='cuda:3'), covar=tensor([0.0393, 0.0214, 0.0252, 0.0183, 0.0616, 0.0495, 0.0310, 0.0770], device='cuda:3'), in_proj_covar=tensor([0.0011, 0.0011, 0.0011, 0.0010, 0.0012, 0.0015, 0.0012, 0.0016], device='cuda:3'), out_proj_covar=tensor([5.4232e-05, 5.5452e-05, 5.4517e-05, 4.9477e-05, 5.7478e-05, 7.0292e-05, 6.4080e-05, 7.5064e-05], device='cuda:3') 2022-12-07 12:12:02,331 INFO [train.py:873] (3/4) Epoch 5, batch 1300, loss[loss=0.1997, simple_loss=0.2066, pruned_loss=0.09641, over 14030.00 frames. ], tot_loss[loss=0.1848, simple_loss=0.1952, pruned_loss=0.08719, over 1935252.08 frames. ], batch size: 22, lr: 1.61e-02, grad_scale: 8.0 2022-12-07 12:12:05,383 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31552.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 12:12:11,376 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31559.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:12:30,376 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31581.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:12:33,700 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9600, 1.9251, 1.8670, 2.0515, 1.9407, 1.6774, 1.1365, 1.7187], device='cuda:3'), covar=tensor([0.0423, 0.0396, 0.0628, 0.0261, 0.0478, 0.0931, 0.1968, 0.0464], device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0129, 0.0120, 0.0104, 0.0162, 0.0110, 0.0147, 0.0152], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 12:12:38,804 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31590.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:12:54,065 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31607.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:12:55,889 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1857, 0.7069, 1.1342, 1.0227, 1.2573, 0.8567, 1.0738, 1.2502], device='cuda:3'), covar=tensor([0.0556, 0.0913, 0.0233, 0.0941, 0.1022, 0.0314, 0.0380, 0.0378], device='cuda:3'), in_proj_covar=tensor([0.0018, 0.0019, 0.0019, 0.0018, 0.0019, 0.0025, 0.0020, 0.0019], device='cuda:3'), out_proj_covar=tensor([6.9035e-05, 7.0937e-05, 6.6096e-05, 6.7984e-05, 6.9405e-05, 9.1114e-05, 7.5548e-05, 6.9696e-05], device='cuda:3') 2022-12-07 12:13:05,512 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.035e+02 2.616e+02 3.313e+02 3.855e+02 7.867e+02, threshold=6.626e+02, percent-clipped=4.0 2022-12-07 12:13:13,012 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31629.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:13:20,926 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31638.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:13:30,391 INFO [train.py:873] (3/4) Epoch 5, batch 1400, loss[loss=0.1826, simple_loss=0.2005, pruned_loss=0.08232, over 14270.00 frames. ], tot_loss[loss=0.1844, simple_loss=0.1951, pruned_loss=0.08687, over 1948213.54 frames. ], batch size: 76, lr: 1.60e-02, grad_scale: 8.0 2022-12-07 12:13:45,875 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31667.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 12:14:18,272 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31704.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:14:28,095 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31715.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 12:14:33,108 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.505e+02 2.718e+02 3.448e+02 4.143e+02 7.580e+02, threshold=6.896e+02, percent-clipped=4.0 2022-12-07 12:14:50,426 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4259, 4.0791, 4.0632, 4.5257, 4.0549, 3.4287, 4.4672, 4.5116], device='cuda:3'), covar=tensor([0.0643, 0.0560, 0.0563, 0.0430, 0.0689, 0.0602, 0.0531, 0.0481], device='cuda:3'), in_proj_covar=tensor([0.0110, 0.0090, 0.0109, 0.0108, 0.0116, 0.0087, 0.0123, 0.0109], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 12:14:57,257 INFO [train.py:873] (3/4) Epoch 5, batch 1500, loss[loss=0.1901, simple_loss=0.1666, pruned_loss=0.1068, over 1223.00 frames. ], tot_loss[loss=0.1851, simple_loss=0.1955, pruned_loss=0.08734, over 1971283.89 frames. ], batch size: 100, lr: 1.60e-02, grad_scale: 8.0 2022-12-07 12:15:11,600 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31765.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:15:11,669 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31765.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:15:21,603 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.70 vs. limit=5.0 2022-12-07 12:15:31,594 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31788.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:16:00,992 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.197e+02 2.523e+02 3.162e+02 3.978e+02 7.805e+02, threshold=6.325e+02, percent-clipped=2.0 2022-12-07 12:16:02,876 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31823.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:16:18,131 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2022-12-07 12:16:24,833 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31847.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 12:16:26,442 INFO [train.py:873] (3/4) Epoch 5, batch 1600, loss[loss=0.1653, simple_loss=0.1523, pruned_loss=0.08915, over 1175.00 frames. ], tot_loss[loss=0.1846, simple_loss=0.1952, pruned_loss=0.08702, over 1969852.71 frames. ], batch size: 100, lr: 1.60e-02, grad_scale: 8.0 2022-12-07 12:16:26,710 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31849.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:16:52,756 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31878.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:16:58,336 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31884.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:17:13,902 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.86 vs. limit=2.0 2022-12-07 12:17:25,442 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31915.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:17:30,354 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.134e+02 2.483e+02 3.139e+02 4.013e+02 6.840e+02, threshold=6.277e+02, percent-clipped=1.0 2022-12-07 12:17:46,190 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31939.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:17:49,607 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4941, 4.2359, 4.1266, 4.5964, 4.1967, 3.7266, 4.5911, 4.4461], device='cuda:3'), covar=tensor([0.0734, 0.0516, 0.0601, 0.0507, 0.0638, 0.0594, 0.0588, 0.0701], device='cuda:3'), in_proj_covar=tensor([0.0108, 0.0089, 0.0108, 0.0104, 0.0113, 0.0085, 0.0121, 0.0107], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 12:17:54,775 INFO [train.py:873] (3/4) Epoch 5, batch 1700, loss[loss=0.1874, simple_loss=0.1951, pruned_loss=0.0898, over 9487.00 frames. ], tot_loss[loss=0.1825, simple_loss=0.1937, pruned_loss=0.08562, over 1936909.17 frames. ], batch size: 100, lr: 1.60e-02, grad_scale: 8.0 2022-12-07 12:17:56,107 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31950.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:18:16,281 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1196, 2.5853, 3.8635, 2.8139, 3.7770, 3.7942, 3.5469, 3.0940], device='cuda:3'), covar=tensor([0.0248, 0.2406, 0.0515, 0.1410, 0.0559, 0.0343, 0.1436, 0.1920], device='cuda:3'), in_proj_covar=tensor([0.0269, 0.0339, 0.0335, 0.0313, 0.0334, 0.0273, 0.0317, 0.0355], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-07 12:18:18,878 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31976.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:18:41,242 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4666, 4.0718, 4.0492, 4.4934, 4.2188, 4.0768, 4.4588, 3.7384], device='cuda:3'), covar=tensor([0.0362, 0.1015, 0.0340, 0.0386, 0.0780, 0.0691, 0.0577, 0.0504], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0213, 0.0143, 0.0135, 0.0144, 0.0113, 0.0212, 0.0136], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 12:18:50,965 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32011.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:18:59,553 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.301e+02 2.546e+02 3.116e+02 4.047e+02 1.395e+03, threshold=6.231e+02, percent-clipped=6.0 2022-12-07 12:19:08,490 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=10.08 vs. limit=5.0 2022-12-07 12:19:25,248 INFO [train.py:873] (3/4) Epoch 5, batch 1800, loss[loss=0.215, simple_loss=0.2077, pruned_loss=0.1111, over 6929.00 frames. ], tot_loss[loss=0.1827, simple_loss=0.194, pruned_loss=0.08568, over 1974402.96 frames. ], batch size: 100, lr: 1.59e-02, grad_scale: 8.0 2022-12-07 12:19:34,902 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32060.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:19:39,398 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32065.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:19:43,739 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.5722, 1.1480, 1.9709, 1.8753, 1.9262, 1.9888, 1.4959, 1.9353], device='cuda:3'), covar=tensor([0.0462, 0.0793, 0.0114, 0.0229, 0.0227, 0.0124, 0.0281, 0.0167], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0157, 0.0100, 0.0137, 0.0115, 0.0118, 0.0089, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-07 12:19:53,215 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8625, 0.8149, 0.7221, 0.8094, 0.9595, 0.4450, 0.5903, 0.8608], device='cuda:3'), covar=tensor([0.0285, 0.0378, 0.0233, 0.0375, 0.0296, 0.0228, 0.0829, 0.0220], device='cuda:3'), in_proj_covar=tensor([0.0017, 0.0018, 0.0018, 0.0018, 0.0019, 0.0024, 0.0020, 0.0019], device='cuda:3'), out_proj_covar=tensor([6.7587e-05, 7.0250e-05, 6.4416e-05, 6.8451e-05, 7.0768e-05, 8.8357e-05, 7.7053e-05, 6.9209e-05], device='cuda:3') 2022-12-07 12:20:20,942 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8122, 1.7761, 2.0239, 1.1910, 1.3615, 1.8254, 1.2070, 1.7231], device='cuda:3'), covar=tensor([0.0759, 0.1409, 0.0417, 0.2047, 0.2371, 0.0594, 0.2980, 0.0868], device='cuda:3'), in_proj_covar=tensor([0.0067, 0.0073, 0.0068, 0.0082, 0.0096, 0.0063, 0.0131, 0.0068], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 12:20:23,223 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32113.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:20:30,237 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.208e+02 2.694e+02 3.195e+02 4.070e+02 6.631e+02, threshold=6.390e+02, percent-clipped=2.0 2022-12-07 12:20:51,115 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32144.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:20:53,820 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32147.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 12:20:55,472 INFO [train.py:873] (3/4) Epoch 5, batch 1900, loss[loss=0.2194, simple_loss=0.1965, pruned_loss=0.1212, over 2619.00 frames. ], tot_loss[loss=0.1835, simple_loss=0.1943, pruned_loss=0.08632, over 1928948.08 frames. ], batch size: 100, lr: 1.59e-02, grad_scale: 8.0 2022-12-07 12:21:07,286 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0103, 2.6635, 4.0322, 4.2649, 4.3083, 2.6565, 4.3503, 3.5506], device='cuda:3'), covar=tensor([0.0119, 0.0304, 0.0439, 0.0134, 0.0082, 0.0457, 0.0077, 0.0289], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0187, 0.0266, 0.0208, 0.0170, 0.0230, 0.0154, 0.0225], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:3') 2022-12-07 12:21:23,390 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32179.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:21:38,187 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32195.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 12:22:01,271 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.008e+02 2.535e+02 3.167e+02 4.268e+02 9.066e+02, threshold=6.334e+02, percent-clipped=2.0 2022-12-07 12:22:11,677 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2022-12-07 12:22:13,225 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32234.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:22:15,985 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1374, 1.2333, 1.1885, 1.2477, 1.2126, 0.8340, 1.3070, 0.8644], device='cuda:3'), covar=tensor([0.0894, 0.0671, 0.1163, 0.0611, 0.0663, 0.0312, 0.0502, 0.0873], device='cuda:3'), in_proj_covar=tensor([0.0012, 0.0012, 0.0012, 0.0010, 0.0012, 0.0016, 0.0013, 0.0017], device='cuda:3'), out_proj_covar=tensor([5.7515e-05, 5.8402e-05, 5.7565e-05, 5.1176e-05, 5.8825e-05, 7.7358e-05, 6.6860e-05, 7.9716e-05], device='cuda:3') 2022-12-07 12:22:25,821 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1787, 1.9746, 4.7847, 4.3319, 4.2665, 4.6825, 4.3167, 4.7859], device='cuda:3'), covar=tensor([0.1104, 0.1048, 0.0051, 0.0096, 0.0102, 0.0082, 0.0082, 0.0068], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0157, 0.0098, 0.0135, 0.0115, 0.0118, 0.0087, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-07 12:22:26,465 INFO [train.py:873] (3/4) Epoch 5, batch 2000, loss[loss=0.1761, simple_loss=0.1976, pruned_loss=0.07732, over 14162.00 frames. ], tot_loss[loss=0.1831, simple_loss=0.1946, pruned_loss=0.08581, over 1926320.85 frames. ], batch size: 35, lr: 1.59e-02, grad_scale: 8.0 2022-12-07 12:22:35,608 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.92 vs. limit=2.0 2022-12-07 12:22:45,448 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32271.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:22:53,527 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2022-12-07 12:23:01,168 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.9680, 0.8724, 0.9310, 0.8281, 1.0766, 0.6409, 0.7750, 1.0965], device='cuda:3'), covar=tensor([0.0526, 0.0682, 0.0204, 0.0423, 0.0446, 0.0280, 0.0584, 0.0278], device='cuda:3'), in_proj_covar=tensor([0.0017, 0.0018, 0.0018, 0.0018, 0.0019, 0.0024, 0.0020, 0.0018], device='cuda:3'), out_proj_covar=tensor([6.8323e-05, 7.0612e-05, 6.3799e-05, 6.8352e-05, 7.0517e-05, 8.9474e-05, 7.7083e-05, 6.7799e-05], device='cuda:3') 2022-12-07 12:23:16,413 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32306.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:23:30,258 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.326e+02 2.510e+02 3.096e+02 3.855e+02 9.073e+02, threshold=6.191e+02, percent-clipped=1.0 2022-12-07 12:23:33,312 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8122, 1.8771, 2.8113, 2.8396, 2.7532, 2.0251, 2.9483, 2.2241], device='cuda:3'), covar=tensor([0.0136, 0.0320, 0.0251, 0.0142, 0.0130, 0.0466, 0.0087, 0.0378], device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0190, 0.0271, 0.0214, 0.0174, 0.0234, 0.0157, 0.0228], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:3') 2022-12-07 12:23:54,566 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2074, 2.8879, 2.6297, 2.7869, 2.0412, 3.0255, 2.9972, 1.2437], device='cuda:3'), covar=tensor([0.4099, 0.1262, 0.2123, 0.1657, 0.1505, 0.1130, 0.1389, 0.3850], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0056, 0.0051, 0.0050, 0.0072, 0.0052, 0.0078, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-07 12:23:55,266 INFO [train.py:873] (3/4) Epoch 5, batch 2100, loss[loss=0.1862, simple_loss=0.2079, pruned_loss=0.08221, over 14007.00 frames. ], tot_loss[loss=0.1826, simple_loss=0.1937, pruned_loss=0.08582, over 1872474.02 frames. ], batch size: 22, lr: 1.59e-02, grad_scale: 8.0 2022-12-07 12:23:57,424 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.62 vs. limit=2.0 2022-12-07 12:24:05,722 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32360.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:24:30,931 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.7601, 3.1050, 4.3525, 3.1818, 4.3415, 4.3828, 4.0331, 3.7920], device='cuda:3'), covar=tensor([0.0237, 0.2251, 0.0530, 0.1428, 0.0457, 0.0410, 0.1708, 0.1734], device='cuda:3'), in_proj_covar=tensor([0.0271, 0.0345, 0.0337, 0.0323, 0.0335, 0.0278, 0.0327, 0.0363], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004], device='cuda:3') 2022-12-07 12:24:34,709 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.6945, 2.6537, 2.6755, 2.6890, 2.6185, 2.4040, 1.2642, 2.4994], device='cuda:3'), covar=tensor([0.0227, 0.0283, 0.0432, 0.0308, 0.0277, 0.0776, 0.2369, 0.0269], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0133, 0.0126, 0.0108, 0.0166, 0.0115, 0.0150, 0.0152], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 12:24:50,024 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32408.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:25:01,675 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.555e+02 2.695e+02 3.113e+02 4.251e+02 1.070e+03, threshold=6.225e+02, percent-clipped=3.0 2022-12-07 12:25:23,248 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32444.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:25:24,214 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8638, 1.6671, 4.2676, 4.0436, 4.0143, 4.2353, 3.8950, 4.2580], device='cuda:3'), covar=tensor([0.1428, 0.1465, 0.0131, 0.0176, 0.0164, 0.0160, 0.0154, 0.0189], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0153, 0.0096, 0.0133, 0.0112, 0.0115, 0.0088, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-07 12:25:27,730 INFO [train.py:873] (3/4) Epoch 5, batch 2200, loss[loss=0.1738, simple_loss=0.1868, pruned_loss=0.08041, over 13919.00 frames. ], tot_loss[loss=0.1827, simple_loss=0.1942, pruned_loss=0.08562, over 1965518.43 frames. ], batch size: 23, lr: 1.58e-02, grad_scale: 8.0 2022-12-07 12:25:54,733 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32479.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:26:05,776 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32492.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:26:05,962 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1016, 2.1886, 2.7972, 2.4566, 2.8111, 2.8380, 2.6549, 2.3152], device='cuda:3'), covar=tensor([0.0354, 0.2182, 0.0599, 0.1411, 0.0454, 0.0506, 0.0798, 0.1863], device='cuda:3'), in_proj_covar=tensor([0.0272, 0.0350, 0.0344, 0.0324, 0.0342, 0.0280, 0.0328, 0.0367], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004], device='cuda:3') 2022-12-07 12:26:22,390 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8862, 1.5225, 2.0104, 1.8098, 2.0765, 1.8443, 1.7745, 1.9106], device='cuda:3'), covar=tensor([0.0262, 0.0996, 0.0110, 0.0218, 0.0122, 0.0255, 0.0117, 0.0186], device='cuda:3'), in_proj_covar=tensor([0.0275, 0.0355, 0.0349, 0.0328, 0.0348, 0.0284, 0.0333, 0.0374], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004], device='cuda:3') 2022-12-07 12:26:31,379 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.306e+02 2.575e+02 3.323e+02 4.480e+02 8.616e+02, threshold=6.645e+02, percent-clipped=4.0 2022-12-07 12:26:36,981 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32527.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:26:43,183 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32534.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:26:56,108 INFO [train.py:873] (3/4) Epoch 5, batch 2300, loss[loss=0.166, simple_loss=0.1647, pruned_loss=0.08362, over 2573.00 frames. ], tot_loss[loss=0.182, simple_loss=0.1941, pruned_loss=0.08496, over 1997179.76 frames. ], batch size: 100, lr: 1.58e-02, grad_scale: 8.0 2022-12-07 12:27:15,959 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32571.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:27:25,537 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32582.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:27:48,223 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32606.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:27:49,561 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2022-12-07 12:27:54,261 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0551, 1.9240, 4.9654, 4.6781, 4.5727, 5.0823, 4.6737, 5.0594], device='cuda:3'), covar=tensor([0.1156, 0.1167, 0.0062, 0.0092, 0.0093, 0.0065, 0.0097, 0.0070], device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0152, 0.0096, 0.0134, 0.0112, 0.0115, 0.0087, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-07 12:27:57,714 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.3127, 1.7712, 1.5333, 1.8220, 1.5861, 1.7475, 1.3974, 1.0810], device='cuda:3'), covar=tensor([0.1573, 0.0553, 0.0616, 0.0318, 0.0844, 0.0377, 0.1201, 0.2179], device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0056, 0.0052, 0.0050, 0.0073, 0.0052, 0.0081, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-07 12:27:59,286 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32619.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:28:00,941 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.405e+02 2.763e+02 3.478e+02 4.434e+02 7.343e+02, threshold=6.957e+02, percent-clipped=3.0 2022-12-07 12:28:26,687 INFO [train.py:873] (3/4) Epoch 5, batch 2400, loss[loss=0.1972, simple_loss=0.206, pruned_loss=0.09417, over 14287.00 frames. ], tot_loss[loss=0.181, simple_loss=0.1935, pruned_loss=0.08425, over 2043735.92 frames. ], batch size: 39, lr: 1.58e-02, grad_scale: 8.0 2022-12-07 12:28:30,509 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=10.66 vs. limit=5.0 2022-12-07 12:28:30,926 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32654.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:28:34,153 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9952, 1.8524, 4.5784, 4.3367, 4.1831, 4.6091, 4.2746, 4.6211], device='cuda:3'), covar=tensor([0.1108, 0.1174, 0.0062, 0.0109, 0.0114, 0.0077, 0.0087, 0.0077], device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0153, 0.0097, 0.0134, 0.0113, 0.0115, 0.0087, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-07 12:29:31,399 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.127e+02 2.520e+02 3.446e+02 4.436e+02 8.213e+02, threshold=6.892e+02, percent-clipped=3.0 2022-12-07 12:29:37,201 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8700, 0.7815, 0.7775, 0.7832, 1.0537, 0.4117, 0.7078, 0.8720], device='cuda:3'), covar=tensor([0.0872, 0.0503, 0.0289, 0.0809, 0.0323, 0.0381, 0.0441, 0.0278], device='cuda:3'), in_proj_covar=tensor([0.0018, 0.0018, 0.0018, 0.0018, 0.0019, 0.0025, 0.0019, 0.0019], device='cuda:3'), out_proj_covar=tensor([6.9739e-05, 7.0570e-05, 6.6224e-05, 7.0714e-05, 7.2036e-05, 9.2448e-05, 7.5973e-05, 6.9852e-05], device='cuda:3') 2022-12-07 12:29:56,008 INFO [train.py:873] (3/4) Epoch 5, batch 2500, loss[loss=0.214, simple_loss=0.2096, pruned_loss=0.1092, over 7814.00 frames. ], tot_loss[loss=0.1816, simple_loss=0.1936, pruned_loss=0.08478, over 2001885.99 frames. ], batch size: 100, lr: 1.58e-02, grad_scale: 8.0 2022-12-07 12:30:35,325 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9407, 1.6258, 2.0505, 1.1812, 1.6403, 1.8712, 2.0483, 1.7353], device='cuda:3'), covar=tensor([0.0573, 0.1374, 0.1210, 0.2312, 0.1199, 0.0671, 0.0403, 0.2172], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0215, 0.0118, 0.0133, 0.0103, 0.0106, 0.0087, 0.0127], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0006, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2022-12-07 12:31:01,908 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.502e+02 2.569e+02 3.218e+02 4.442e+02 9.627e+02, threshold=6.437e+02, percent-clipped=1.0 2022-12-07 12:31:27,904 INFO [train.py:873] (3/4) Epoch 5, batch 2600, loss[loss=0.165, simple_loss=0.1516, pruned_loss=0.08921, over 2629.00 frames. ], tot_loss[loss=0.1802, simple_loss=0.1927, pruned_loss=0.08388, over 1998260.07 frames. ], batch size: 100, lr: 1.57e-02, grad_scale: 16.0 2022-12-07 12:31:33,665 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2022-12-07 12:32:04,675 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1183, 1.9492, 2.0429, 2.1492, 2.0893, 1.9774, 2.2038, 1.8244], device='cuda:3'), covar=tensor([0.0579, 0.1249, 0.0579, 0.0599, 0.0849, 0.0595, 0.0769, 0.0729], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0212, 0.0142, 0.0136, 0.0141, 0.0115, 0.0209, 0.0136], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 12:32:23,127 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.2470, 4.0573, 4.3810, 4.0059, 4.1936, 4.5882, 1.5833, 4.0358], device='cuda:3'), covar=tensor([0.0207, 0.0392, 0.0440, 0.0290, 0.0322, 0.0160, 0.3049, 0.0260], device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0138, 0.0130, 0.0110, 0.0170, 0.0118, 0.0154, 0.0157], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 12:32:33,210 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.371e+02 2.535e+02 3.293e+02 4.332e+02 8.259e+02, threshold=6.586e+02, percent-clipped=6.0 2022-12-07 12:32:57,304 INFO [train.py:873] (3/4) Epoch 5, batch 2700, loss[loss=0.1383, simple_loss=0.164, pruned_loss=0.0563, over 14232.00 frames. ], tot_loss[loss=0.1793, simple_loss=0.1919, pruned_loss=0.08339, over 1978633.09 frames. ], batch size: 37, lr: 1.57e-02, grad_scale: 8.0 2022-12-07 12:33:01,279 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.51 vs. limit=5.0 2022-12-07 12:33:13,461 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32966.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:33:46,012 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.60 vs. limit=5.0 2022-12-07 12:33:58,841 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3551, 2.3009, 2.2257, 2.3754, 1.9244, 2.4940, 2.1054, 1.0477], device='cuda:3'), covar=tensor([0.3159, 0.0820, 0.0873, 0.0785, 0.1117, 0.0453, 0.1835, 0.3674], device='cuda:3'), in_proj_covar=tensor([0.0160, 0.0057, 0.0053, 0.0052, 0.0074, 0.0053, 0.0085, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2022-12-07 12:34:03,192 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.510e+02 2.537e+02 3.172e+02 3.779e+02 6.676e+02, threshold=6.344e+02, percent-clipped=1.0 2022-12-07 12:34:07,325 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33027.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:34:27,430 INFO [train.py:873] (3/4) Epoch 5, batch 2800, loss[loss=0.198, simple_loss=0.1812, pruned_loss=0.1074, over 1168.00 frames. ], tot_loss[loss=0.1796, simple_loss=0.1924, pruned_loss=0.08339, over 2021076.65 frames. ], batch size: 100, lr: 1.57e-02, grad_scale: 8.0 2022-12-07 12:34:45,091 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33069.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:35:11,021 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7565, 1.9844, 1.7894, 2.0889, 1.6426, 1.9143, 1.6404, 0.9485], device='cuda:3'), covar=tensor([0.2009, 0.0773, 0.0905, 0.0472, 0.1082, 0.0486, 0.1893, 0.3340], device='cuda:3'), in_proj_covar=tensor([0.0158, 0.0057, 0.0053, 0.0052, 0.0074, 0.0053, 0.0084, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2022-12-07 12:35:33,622 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.258e+02 2.428e+02 3.052e+02 4.047e+02 7.917e+02, threshold=6.105e+02, percent-clipped=1.0 2022-12-07 12:35:40,375 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33130.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:35:56,744 INFO [train.py:873] (3/4) Epoch 5, batch 2900, loss[loss=0.1885, simple_loss=0.1964, pruned_loss=0.09032, over 13861.00 frames. ], tot_loss[loss=0.1812, simple_loss=0.1933, pruned_loss=0.08452, over 2033273.57 frames. ], batch size: 20, lr: 1.57e-02, grad_scale: 4.0 2022-12-07 12:36:02,333 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33155.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:36:09,610 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2254, 1.6889, 2.2069, 1.9229, 2.2334, 2.0992, 1.9286, 1.9786], device='cuda:3'), covar=tensor([0.0211, 0.1207, 0.0218, 0.0471, 0.0184, 0.0408, 0.0189, 0.0603], device='cuda:3'), in_proj_covar=tensor([0.0265, 0.0346, 0.0341, 0.0322, 0.0345, 0.0274, 0.0320, 0.0359], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-07 12:36:56,616 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33216.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:37:02,203 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.058e+02 2.752e+02 3.414e+02 4.319e+02 7.862e+02, threshold=6.827e+02, percent-clipped=8.0 2022-12-07 12:37:19,186 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.35 vs. limit=2.0 2022-12-07 12:37:26,082 INFO [train.py:873] (3/4) Epoch 5, batch 3000, loss[loss=0.1596, simple_loss=0.1782, pruned_loss=0.07043, over 14406.00 frames. ], tot_loss[loss=0.1804, simple_loss=0.1926, pruned_loss=0.08416, over 1979124.22 frames. ], batch size: 53, lr: 1.56e-02, grad_scale: 4.0 2022-12-07 12:37:26,083 INFO [train.py:896] (3/4) Computing validation loss 2022-12-07 12:37:38,399 INFO [train.py:905] (3/4) Epoch 5, validation: loss=0.1238, simple_loss=0.167, pruned_loss=0.04033, over 857387.00 frames. 2022-12-07 12:37:38,400 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17841MB 2022-12-07 12:38:44,258 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33322.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:38:44,934 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.389e+02 2.483e+02 3.391e+02 4.332e+02 8.491e+02, threshold=6.783e+02, percent-clipped=1.0 2022-12-07 12:39:08,695 INFO [train.py:873] (3/4) Epoch 5, batch 3100, loss[loss=0.206, simple_loss=0.2094, pruned_loss=0.1013, over 8650.00 frames. ], tot_loss[loss=0.1809, simple_loss=0.1924, pruned_loss=0.08466, over 1957387.77 frames. ], batch size: 100, lr: 1.56e-02, grad_scale: 4.0 2022-12-07 12:39:09,814 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8671, 1.5798, 4.2804, 1.9675, 4.3081, 4.6373, 4.3709, 5.1688], device='cuda:3'), covar=tensor([0.0136, 0.2827, 0.0290, 0.2003, 0.0241, 0.0188, 0.0171, 0.0076], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0160, 0.0134, 0.0170, 0.0149, 0.0145, 0.0119, 0.0121], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:3') 2022-12-07 12:39:20,301 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1119, 2.0544, 2.9999, 3.1796, 3.0232, 2.1967, 3.2977, 2.4007], device='cuda:3'), covar=tensor([0.0115, 0.0292, 0.0267, 0.0146, 0.0110, 0.0463, 0.0070, 0.0358], device='cuda:3'), in_proj_covar=tensor([0.0181, 0.0191, 0.0275, 0.0221, 0.0176, 0.0238, 0.0160, 0.0228], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:3') 2022-12-07 12:39:25,665 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 2022-12-07 12:39:46,456 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8476, 1.5187, 2.0655, 1.7461, 2.0410, 1.3419, 1.6820, 1.6974], device='cuda:3'), covar=tensor([0.0747, 0.2549, 0.0161, 0.1323, 0.0296, 0.1259, 0.0759, 0.0428], device='cuda:3'), in_proj_covar=tensor([0.0218, 0.0263, 0.0166, 0.0350, 0.0169, 0.0264, 0.0241, 0.0169], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 12:39:51,597 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1082, 2.0634, 1.9389, 2.1860, 1.7857, 1.9301, 2.1132, 2.1437], device='cuda:3'), covar=tensor([0.0881, 0.0985, 0.1034, 0.0807, 0.1260, 0.0830, 0.0936, 0.0849], device='cuda:3'), in_proj_covar=tensor([0.0108, 0.0093, 0.0108, 0.0107, 0.0114, 0.0085, 0.0122, 0.0108], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 12:40:15,024 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.322e+02 2.236e+02 3.030e+02 3.906e+02 6.925e+02, threshold=6.060e+02, percent-clipped=1.0 2022-12-07 12:40:17,251 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33425.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:40:38,098 INFO [train.py:873] (3/4) Epoch 5, batch 3200, loss[loss=0.1897, simple_loss=0.1708, pruned_loss=0.1043, over 1229.00 frames. ], tot_loss[loss=0.1816, simple_loss=0.1932, pruned_loss=0.085, over 1951288.04 frames. ], batch size: 100, lr: 1.56e-02, grad_scale: 8.0 2022-12-07 12:40:55,035 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1652, 1.9503, 2.1350, 2.1456, 2.0914, 2.1076, 2.2659, 1.8681], device='cuda:3'), covar=tensor([0.0633, 0.1184, 0.0528, 0.0628, 0.0793, 0.0525, 0.0663, 0.0696], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0212, 0.0143, 0.0137, 0.0139, 0.0112, 0.0211, 0.0137], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 12:40:59,550 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33472.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:41:19,510 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9646, 3.8620, 4.1362, 3.5359, 3.8349, 4.0803, 1.5649, 3.6791], device='cuda:3'), covar=tensor([0.0177, 0.0242, 0.0357, 0.0392, 0.0249, 0.0217, 0.2875, 0.0254], device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0133, 0.0125, 0.0106, 0.0162, 0.0114, 0.0148, 0.0153], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 12:41:34,160 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 2022-12-07 12:41:34,516 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33511.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:41:44,536 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.346e+02 2.667e+02 3.351e+02 4.253e+02 1.055e+03, threshold=6.703e+02, percent-clipped=3.0 2022-12-07 12:41:53,192 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33533.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:42:08,400 INFO [train.py:873] (3/4) Epoch 5, batch 3300, loss[loss=0.1712, simple_loss=0.1633, pruned_loss=0.08959, over 1223.00 frames. ], tot_loss[loss=0.1822, simple_loss=0.193, pruned_loss=0.08566, over 1934328.69 frames. ], batch size: 100, lr: 1.56e-02, grad_scale: 8.0 2022-12-07 12:42:41,120 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0 2022-12-07 12:43:12,997 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=33622.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:43:13,763 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.043e+02 2.484e+02 3.236e+02 4.553e+02 1.226e+03, threshold=6.473e+02, percent-clipped=4.0 2022-12-07 12:43:37,537 INFO [train.py:873] (3/4) Epoch 5, batch 3400, loss[loss=0.1616, simple_loss=0.1399, pruned_loss=0.09163, over 2691.00 frames. ], tot_loss[loss=0.1804, simple_loss=0.1922, pruned_loss=0.08431, over 1948056.34 frames. ], batch size: 100, lr: 1.56e-02, grad_scale: 8.0 2022-12-07 12:43:41,353 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8966, 1.5141, 2.0075, 1.7507, 2.1208, 1.7993, 1.6404, 1.9238], device='cuda:3'), covar=tensor([0.0278, 0.0918, 0.0127, 0.0241, 0.0168, 0.0371, 0.0143, 0.0213], device='cuda:3'), in_proj_covar=tensor([0.0264, 0.0341, 0.0343, 0.0309, 0.0346, 0.0274, 0.0321, 0.0358], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-07 12:43:55,236 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.33 vs. limit=5.0 2022-12-07 12:43:55,878 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1091, 1.8431, 4.7864, 4.4986, 4.1558, 4.7828, 4.4852, 4.8554], device='cuda:3'), covar=tensor([0.1171, 0.1249, 0.0081, 0.0114, 0.0142, 0.0089, 0.0092, 0.0098], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0157, 0.0098, 0.0137, 0.0115, 0.0119, 0.0088, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-07 12:43:57,507 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=33670.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:44:03,510 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33677.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:44:36,100 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1424, 1.9188, 2.0038, 2.1204, 2.0956, 2.0348, 2.2020, 1.8511], device='cuda:3'), covar=tensor([0.0563, 0.1107, 0.0532, 0.0582, 0.0708, 0.0511, 0.0670, 0.0623], device='cuda:3'), in_proj_covar=tensor([0.0138, 0.0217, 0.0144, 0.0139, 0.0142, 0.0114, 0.0211, 0.0138], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 12:44:43,755 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.328e+02 2.478e+02 3.354e+02 4.337e+02 1.168e+03, threshold=6.709e+02, percent-clipped=6.0 2022-12-07 12:44:45,690 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=33725.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:44:56,994 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33738.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 12:45:07,674 INFO [train.py:873] (3/4) Epoch 5, batch 3500, loss[loss=0.2056, simple_loss=0.1827, pruned_loss=0.1143, over 3847.00 frames. ], tot_loss[loss=0.1801, simple_loss=0.1923, pruned_loss=0.084, over 1968698.19 frames. ], batch size: 100, lr: 1.55e-02, grad_scale: 8.0 2022-12-07 12:45:18,950 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3087, 2.0126, 3.4650, 2.4615, 3.3215, 1.6590, 2.7476, 3.0894], device='cuda:3'), covar=tensor([0.0459, 0.4790, 0.0340, 0.6557, 0.0371, 0.4526, 0.1103, 0.0330], device='cuda:3'), in_proj_covar=tensor([0.0226, 0.0269, 0.0168, 0.0355, 0.0176, 0.0276, 0.0248, 0.0171], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 12:45:28,351 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=33773.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:45:38,043 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=9.12 vs. limit=5.0 2022-12-07 12:45:49,454 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.54 vs. limit=5.0 2022-12-07 12:46:02,430 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=33811.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:46:13,092 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.895e+01 2.306e+02 2.839e+02 3.644e+02 6.341e+02, threshold=5.678e+02, percent-clipped=0.0 2022-12-07 12:46:18,187 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33828.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:46:36,551 INFO [train.py:873] (3/4) Epoch 5, batch 3600, loss[loss=0.1642, simple_loss=0.148, pruned_loss=0.09016, over 1240.00 frames. ], tot_loss[loss=0.1812, simple_loss=0.1928, pruned_loss=0.08479, over 1910415.32 frames. ], batch size: 100, lr: 1.55e-02, grad_scale: 8.0 2022-12-07 12:46:46,234 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=33859.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:46:50,729 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2022-12-07 12:47:38,618 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 2022-12-07 12:47:44,853 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.078e+02 2.977e+02 3.494e+02 4.554e+02 8.959e+02, threshold=6.988e+02, percent-clipped=9.0 2022-12-07 12:47:59,770 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33939.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:48:09,171 INFO [train.py:873] (3/4) Epoch 5, batch 3700, loss[loss=0.1842, simple_loss=0.1995, pruned_loss=0.08441, over 14250.00 frames. ], tot_loss[loss=0.181, simple_loss=0.1928, pruned_loss=0.08461, over 1952752.77 frames. ], batch size: 44, lr: 1.55e-02, grad_scale: 8.0 2022-12-07 12:48:23,761 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1729, 2.2250, 4.2488, 3.0202, 4.2184, 1.8724, 2.9951, 4.0248], device='cuda:3'), covar=tensor([0.0348, 0.4810, 0.0268, 0.7757, 0.0233, 0.4368, 0.1244, 0.0234], device='cuda:3'), in_proj_covar=tensor([0.0220, 0.0259, 0.0164, 0.0348, 0.0172, 0.0263, 0.0244, 0.0168], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 12:48:36,462 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.64 vs. limit=2.0 2022-12-07 12:48:54,463 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34000.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:49:10,430 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7891, 1.6251, 2.9856, 2.1069, 2.8502, 1.4857, 2.2531, 2.7036], device='cuda:3'), covar=tensor([0.0658, 0.4808, 0.0344, 0.6092, 0.0376, 0.3729, 0.1314, 0.0352], device='cuda:3'), in_proj_covar=tensor([0.0225, 0.0263, 0.0168, 0.0361, 0.0177, 0.0271, 0.0249, 0.0170], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 12:49:15,043 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.339e+02 2.468e+02 3.019e+02 3.918e+02 8.852e+02, threshold=6.038e+02, percent-clipped=4.0 2022-12-07 12:49:24,198 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34033.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 12:49:34,044 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34044.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:49:38,092 INFO [train.py:873] (3/4) Epoch 5, batch 3800, loss[loss=0.1929, simple_loss=0.1749, pruned_loss=0.1055, over 3883.00 frames. ], tot_loss[loss=0.181, simple_loss=0.193, pruned_loss=0.08445, over 2020766.31 frames. ], batch size: 100, lr: 1.55e-02, grad_scale: 8.0 2022-12-07 12:49:41,871 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5218, 3.9796, 3.0821, 4.7945, 4.2628, 4.6524, 3.4657, 3.4067], device='cuda:3'), covar=tensor([0.0496, 0.1152, 0.4249, 0.0253, 0.0758, 0.0553, 0.1514, 0.3632], device='cuda:3'), in_proj_covar=tensor([0.0226, 0.0305, 0.0311, 0.0187, 0.0246, 0.0245, 0.0263, 0.0302], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 12:50:24,519 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2022-12-07 12:50:29,191 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34105.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:50:45,933 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.381e+02 2.542e+02 3.059e+02 3.773e+02 1.009e+03, threshold=6.118e+02, percent-clipped=4.0 2022-12-07 12:50:50,673 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34128.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:51:05,912 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34145.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:51:09,665 INFO [train.py:873] (3/4) Epoch 5, batch 3900, loss[loss=0.1431, simple_loss=0.1366, pruned_loss=0.07478, over 1202.00 frames. ], tot_loss[loss=0.1805, simple_loss=0.1921, pruned_loss=0.08448, over 1923728.17 frames. ], batch size: 100, lr: 1.54e-02, grad_scale: 8.0 2022-12-07 12:51:33,808 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34176.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:51:35,621 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.1548, 4.9267, 4.7744, 5.2455, 4.8314, 4.3646, 5.2874, 5.0571], device='cuda:3'), covar=tensor([0.0843, 0.0495, 0.0549, 0.0568, 0.0746, 0.0490, 0.0620, 0.0915], device='cuda:3'), in_proj_covar=tensor([0.0108, 0.0094, 0.0109, 0.0108, 0.0116, 0.0088, 0.0123, 0.0106], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 12:51:54,872 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4645, 1.3169, 2.7953, 1.4637, 2.7496, 2.6768, 2.0967, 2.7963], device='cuda:3'), covar=tensor([0.0193, 0.2073, 0.0222, 0.1545, 0.0258, 0.0326, 0.0701, 0.0182], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0158, 0.0130, 0.0168, 0.0149, 0.0146, 0.0119, 0.0118], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:3') 2022-12-07 12:51:58,756 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.5801, 5.0556, 5.0325, 5.5220, 5.2798, 4.6158, 5.5664, 4.5746], device='cuda:3'), covar=tensor([0.0259, 0.0789, 0.0245, 0.0434, 0.0584, 0.0287, 0.0358, 0.0403], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0216, 0.0142, 0.0134, 0.0139, 0.0112, 0.0212, 0.0136], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 12:52:02,425 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34206.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:52:15,641 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.91 vs. limit=2.0 2022-12-07 12:52:17,891 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.643e+02 2.452e+02 3.217e+02 4.044e+02 1.041e+03, threshold=6.433e+02, percent-clipped=5.0 2022-12-07 12:52:41,188 INFO [train.py:873] (3/4) Epoch 5, batch 4000, loss[loss=0.2095, simple_loss=0.2102, pruned_loss=0.1044, over 8595.00 frames. ], tot_loss[loss=0.1806, simple_loss=0.1919, pruned_loss=0.08464, over 1847241.15 frames. ], batch size: 100, lr: 1.54e-02, grad_scale: 8.0 2022-12-07 12:53:07,701 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2301, 1.6501, 2.2029, 1.9547, 2.3807, 2.1288, 1.9977, 2.2119], device='cuda:3'), covar=tensor([0.0230, 0.1127, 0.0267, 0.0515, 0.0224, 0.0448, 0.0214, 0.0600], device='cuda:3'), in_proj_covar=tensor([0.0260, 0.0333, 0.0344, 0.0311, 0.0335, 0.0275, 0.0315, 0.0353], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-07 12:53:22,808 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34295.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:53:30,052 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1525, 2.4503, 4.1097, 4.3120, 4.4143, 2.5175, 4.4007, 3.3483], device='cuda:3'), covar=tensor([0.0110, 0.0323, 0.0332, 0.0124, 0.0068, 0.0506, 0.0066, 0.0327], device='cuda:3'), in_proj_covar=tensor([0.0188, 0.0191, 0.0280, 0.0223, 0.0178, 0.0237, 0.0165, 0.0230], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:3') 2022-12-07 12:53:35,751 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.9719, 0.8219, 0.8658, 0.9356, 1.0526, 0.5209, 0.8001, 0.9340], device='cuda:3'), covar=tensor([0.0682, 0.0663, 0.0434, 0.0359, 0.0278, 0.0500, 0.0861, 0.0687], device='cuda:3'), in_proj_covar=tensor([0.0018, 0.0019, 0.0018, 0.0018, 0.0019, 0.0026, 0.0020, 0.0019], device='cuda:3'), out_proj_covar=tensor([7.1946e-05, 7.4619e-05, 6.8254e-05, 7.0510e-05, 7.4782e-05, 9.4462e-05, 7.9836e-05, 7.3464e-05], device='cuda:3') 2022-12-07 12:53:41,254 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7512, 2.6324, 3.5681, 2.1852, 2.2953, 2.8281, 1.3868, 2.8627], device='cuda:3'), covar=tensor([0.1380, 0.0993, 0.0693, 0.2493, 0.2317, 0.1696, 0.5568, 0.0954], device='cuda:3'), in_proj_covar=tensor([0.0070, 0.0076, 0.0073, 0.0081, 0.0103, 0.0065, 0.0136, 0.0070], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-07 12:53:48,674 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.550e+02 2.578e+02 3.207e+02 3.995e+02 8.945e+02, threshold=6.415e+02, percent-clipped=4.0 2022-12-07 12:53:57,308 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7795, 1.3421, 2.0971, 1.3379, 2.0208, 2.0879, 1.7756, 2.0714], device='cuda:3'), covar=tensor([0.0246, 0.1190, 0.0212, 0.1201, 0.0266, 0.0239, 0.0546, 0.0234], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0155, 0.0129, 0.0165, 0.0146, 0.0141, 0.0116, 0.0116], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:3') 2022-12-07 12:53:57,322 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34333.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:54:12,175 INFO [train.py:873] (3/4) Epoch 5, batch 4100, loss[loss=0.1722, simple_loss=0.1799, pruned_loss=0.08225, over 5991.00 frames. ], tot_loss[loss=0.1806, simple_loss=0.1922, pruned_loss=0.08452, over 1926146.02 frames. ], batch size: 100, lr: 1.54e-02, grad_scale: 8.0 2022-12-07 12:54:40,237 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34381.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:54:57,948 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34400.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:55:07,706 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6320, 1.4509, 3.6168, 1.4861, 3.6001, 3.7317, 2.7903, 3.9763], device='cuda:3'), covar=tensor([0.0159, 0.2624, 0.0286, 0.2207, 0.0316, 0.0248, 0.0533, 0.0113], device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0154, 0.0130, 0.0164, 0.0147, 0.0143, 0.0116, 0.0116], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:3') 2022-12-07 12:55:18,133 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.386e+02 2.334e+02 3.145e+02 3.965e+02 8.915e+02, threshold=6.291e+02, percent-clipped=4.0 2022-12-07 12:55:20,528 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.4454, 3.4925, 5.0143, 4.1300, 5.0736, 5.1658, 4.4300, 4.4107], device='cuda:3'), covar=tensor([0.0156, 0.2464, 0.0670, 0.1076, 0.0487, 0.0359, 0.1858, 0.1613], device='cuda:3'), in_proj_covar=tensor([0.0266, 0.0333, 0.0344, 0.0310, 0.0338, 0.0274, 0.0317, 0.0355], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-07 12:55:23,828 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2022-12-07 12:55:41,763 INFO [train.py:873] (3/4) Epoch 5, batch 4200, loss[loss=0.1587, simple_loss=0.1707, pruned_loss=0.07332, over 6905.00 frames. ], tot_loss[loss=0.18, simple_loss=0.1917, pruned_loss=0.08417, over 1916730.07 frames. ], batch size: 100, lr: 1.54e-02, grad_scale: 8.0 2022-12-07 12:55:41,946 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1314, 1.5543, 2.6154, 1.8910, 2.3296, 1.5544, 1.9101, 2.3271], device='cuda:3'), covar=tensor([0.0784, 0.4056, 0.0251, 0.4665, 0.0397, 0.3153, 0.1237, 0.0373], device='cuda:3'), in_proj_covar=tensor([0.0221, 0.0263, 0.0166, 0.0350, 0.0174, 0.0262, 0.0249, 0.0170], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 12:56:28,626 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34501.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:56:46,699 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.78 vs. limit=5.0 2022-12-07 12:56:48,745 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.089e+02 2.566e+02 3.165e+02 3.922e+02 1.587e+03, threshold=6.330e+02, percent-clipped=5.0 2022-12-07 12:57:11,678 INFO [train.py:873] (3/4) Epoch 5, batch 4300, loss[loss=0.1699, simple_loss=0.1846, pruned_loss=0.07759, over 13969.00 frames. ], tot_loss[loss=0.1825, simple_loss=0.1932, pruned_loss=0.08586, over 1855847.15 frames. ], batch size: 19, lr: 1.54e-02, grad_scale: 8.0 2022-12-07 12:57:16,157 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2022-12-07 12:57:34,394 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2847, 3.1418, 3.3529, 3.1302, 3.2090, 2.7516, 1.2789, 3.0167], device='cuda:3'), covar=tensor([0.0195, 0.0304, 0.0337, 0.0307, 0.0286, 0.0690, 0.2865, 0.0251], device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0134, 0.0121, 0.0105, 0.0162, 0.0114, 0.0146, 0.0152], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 12:57:50,015 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.08 vs. limit=2.0 2022-12-07 12:57:51,398 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34593.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:57:53,683 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34595.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:58:18,112 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34622.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:58:18,607 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.121e+02 2.611e+02 3.299e+02 4.173e+02 8.760e+02, threshold=6.598e+02, percent-clipped=4.0 2022-12-07 12:58:21,441 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34626.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:58:36,755 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34643.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:58:41,836 INFO [train.py:873] (3/4) Epoch 5, batch 4400, loss[loss=0.1699, simple_loss=0.1909, pruned_loss=0.07448, over 14265.00 frames. ], tot_loss[loss=0.1803, simple_loss=0.1924, pruned_loss=0.08411, over 1923663.35 frames. ], batch size: 31, lr: 1.53e-02, grad_scale: 8.0 2022-12-07 12:58:46,351 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34654.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:59:12,438 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34683.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 12:59:15,945 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34687.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:59:27,204 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34700.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 12:59:36,226 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34710.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 12:59:48,182 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.105e+02 2.387e+02 3.073e+02 3.868e+02 1.059e+03, threshold=6.147e+02, percent-clipped=1.0 2022-12-07 13:00:10,389 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34748.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:00:11,140 INFO [train.py:873] (3/4) Epoch 5, batch 4500, loss[loss=0.1807, simple_loss=0.1595, pruned_loss=0.1009, over 1243.00 frames. ], tot_loss[loss=0.1789, simple_loss=0.1919, pruned_loss=0.08298, over 1965952.66 frames. ], batch size: 100, lr: 1.53e-02, grad_scale: 8.0 2022-12-07 13:00:31,219 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34771.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 13:00:37,437 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.95 vs. limit=2.0 2022-12-07 13:00:57,982 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34801.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:01:16,933 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.242e+02 2.392e+02 3.325e+02 4.104e+02 6.523e+02, threshold=6.650e+02, percent-clipped=2.0 2022-12-07 13:01:40,484 INFO [train.py:873] (3/4) Epoch 5, batch 4600, loss[loss=0.1778, simple_loss=0.1919, pruned_loss=0.08182, over 14028.00 frames. ], tot_loss[loss=0.1798, simple_loss=0.1926, pruned_loss=0.08351, over 1971200.22 frames. ], batch size: 26, lr: 1.53e-02, grad_scale: 8.0 2022-12-07 13:01:40,575 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34849.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:02:22,055 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4608, 1.8922, 3.5445, 2.5568, 3.3217, 1.9917, 2.8018, 3.3403], device='cuda:3'), covar=tensor([0.0424, 0.4666, 0.0258, 0.6762, 0.0488, 0.3562, 0.1131, 0.0285], device='cuda:3'), in_proj_covar=tensor([0.0219, 0.0260, 0.0166, 0.0351, 0.0175, 0.0265, 0.0240, 0.0169], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 13:02:26,177 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.2713, 1.2909, 1.1650, 1.4169, 1.0616, 0.9235, 0.8588, 0.7412], device='cuda:3'), covar=tensor([0.0269, 0.0367, 0.0455, 0.0244, 0.0440, 0.0281, 0.0178, 0.0388], device='cuda:3'), in_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0010, 0.0013, 0.0015, 0.0012, 0.0017], device='cuda:3'), out_proj_covar=tensor([5.9378e-05, 6.0282e-05, 6.0008e-05, 5.4796e-05, 6.4086e-05, 8.1720e-05, 6.6932e-05, 8.2374e-05], device='cuda:3') 2022-12-07 13:02:44,495 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7277, 2.8145, 2.3248, 2.4914, 2.1342, 2.9566, 2.6301, 1.2074], device='cuda:3'), covar=tensor([0.2994, 0.0624, 0.2186, 0.0873, 0.1157, 0.0422, 0.1294, 0.3794], device='cuda:3'), in_proj_covar=tensor([0.0154, 0.0059, 0.0053, 0.0050, 0.0070, 0.0055, 0.0082, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2022-12-07 13:02:46,770 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.280e+02 2.771e+02 3.403e+02 4.151e+02 6.915e+02, threshold=6.807e+02, percent-clipped=1.0 2022-12-07 13:03:10,250 INFO [train.py:873] (3/4) Epoch 5, batch 4700, loss[loss=0.1916, simple_loss=0.1671, pruned_loss=0.108, over 1249.00 frames. ], tot_loss[loss=0.1791, simple_loss=0.1925, pruned_loss=0.08283, over 1976681.77 frames. ], batch size: 100, lr: 1.53e-02, grad_scale: 8.0 2022-12-07 13:03:10,355 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34949.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:03:24,874 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2022-12-07 13:03:28,927 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2909, 2.4109, 3.3352, 2.2147, 2.3293, 2.3109, 1.3846, 2.5742], device='cuda:3'), covar=tensor([0.1704, 0.1812, 0.0827, 0.2303, 0.2374, 0.1589, 0.5619, 0.1264], device='cuda:3'), in_proj_covar=tensor([0.0070, 0.0079, 0.0075, 0.0083, 0.0105, 0.0064, 0.0136, 0.0072], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-07 13:03:35,715 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34978.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 13:03:39,280 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34982.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:04:21,150 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.362e+02 2.441e+02 3.432e+02 4.388e+02 1.189e+03, threshold=6.863e+02, percent-clipped=8.0 2022-12-07 13:04:45,220 INFO [train.py:873] (3/4) Epoch 5, batch 4800, loss[loss=0.1932, simple_loss=0.1724, pruned_loss=0.1071, over 2636.00 frames. ], tot_loss[loss=0.1774, simple_loss=0.191, pruned_loss=0.08192, over 1965999.98 frames. ], batch size: 100, lr: 1.52e-02, grad_scale: 8.0 2022-12-07 13:04:51,646 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.7450, 1.2556, 1.2825, 1.3093, 1.0636, 1.2851, 1.0370, 0.8188], device='cuda:3'), covar=tensor([0.3106, 0.0806, 0.0355, 0.0385, 0.0792, 0.0647, 0.1051, 0.1374], device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0061, 0.0053, 0.0052, 0.0071, 0.0056, 0.0083, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2022-12-07 13:05:01,179 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35066.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 13:05:24,492 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35092.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:05:53,126 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.345e+02 2.350e+02 3.192e+02 4.148e+02 1.074e+03, threshold=6.385e+02, percent-clipped=3.0 2022-12-07 13:06:16,619 INFO [train.py:873] (3/4) Epoch 5, batch 4900, loss[loss=0.1919, simple_loss=0.2123, pruned_loss=0.08573, over 14376.00 frames. ], tot_loss[loss=0.1779, simple_loss=0.1914, pruned_loss=0.08219, over 2006427.91 frames. ], batch size: 55, lr: 1.52e-02, grad_scale: 16.0 2022-12-07 13:06:20,595 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35153.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:06:55,960 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.73 vs. limit=2.0 2022-12-07 13:07:22,231 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5984, 2.3686, 3.3706, 2.5627, 3.3682, 3.1886, 3.2400, 2.6891], device='cuda:3'), covar=tensor([0.0365, 0.2649, 0.0822, 0.1883, 0.0585, 0.0691, 0.1075, 0.2020], device='cuda:3'), in_proj_covar=tensor([0.0271, 0.0336, 0.0359, 0.0313, 0.0346, 0.0282, 0.0321, 0.0352], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-07 13:07:24,093 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.440e+01 2.537e+02 3.134e+02 4.134e+02 8.105e+02, threshold=6.269e+02, percent-clipped=2.0 2022-12-07 13:07:36,146 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.0218, 0.7131, 0.8466, 1.0351, 1.0121, 0.3953, 0.9587, 1.0031], device='cuda:3'), covar=tensor([0.0362, 0.0530, 0.0328, 0.0263, 0.0408, 0.0312, 0.0368, 0.0442], device='cuda:3'), in_proj_covar=tensor([0.0018, 0.0018, 0.0018, 0.0018, 0.0020, 0.0024, 0.0019, 0.0019], device='cuda:3'), out_proj_covar=tensor([7.1489e-05, 7.3665e-05, 6.8132e-05, 7.2322e-05, 7.6793e-05, 9.1254e-05, 7.8596e-05, 7.3718e-05], device='cuda:3') 2022-12-07 13:07:47,530 INFO [train.py:873] (3/4) Epoch 5, batch 5000, loss[loss=0.2444, simple_loss=0.1988, pruned_loss=0.145, over 1217.00 frames. ], tot_loss[loss=0.179, simple_loss=0.1919, pruned_loss=0.0831, over 1955865.44 frames. ], batch size: 100, lr: 1.52e-02, grad_scale: 16.0 2022-12-07 13:07:47,656 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35249.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:07:53,676 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9673, 1.6274, 4.2149, 3.9218, 3.8847, 4.2570, 3.7590, 4.2923], device='cuda:3'), covar=tensor([0.1084, 0.1236, 0.0073, 0.0117, 0.0123, 0.0079, 0.0133, 0.0079], device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0157, 0.0101, 0.0142, 0.0119, 0.0120, 0.0094, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-07 13:07:58,569 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1772, 1.2944, 1.2222, 1.3682, 1.6609, 0.8343, 1.3606, 0.9095], device='cuda:3'), covar=tensor([0.1077, 0.1096, 0.1460, 0.1316, 0.1342, 0.0495, 0.0486, 0.1305], device='cuda:3'), in_proj_covar=tensor([0.0012, 0.0012, 0.0011, 0.0010, 0.0012, 0.0016, 0.0011, 0.0017], device='cuda:3'), out_proj_covar=tensor([6.0117e-05, 6.1703e-05, 5.9526e-05, 5.6429e-05, 6.3296e-05, 8.4537e-05, 6.6936e-05, 8.4081e-05], device='cuda:3') 2022-12-07 13:08:11,649 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.84 vs. limit=2.0 2022-12-07 13:08:13,635 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35278.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 13:08:17,300 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35282.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:08:30,436 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35297.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:08:46,913 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.9625, 0.7315, 0.6447, 0.9137, 0.9035, 0.3022, 0.8017, 0.9348], device='cuda:3'), covar=tensor([0.0366, 0.0733, 0.0320, 0.0211, 0.0338, 0.0267, 0.0694, 0.0424], device='cuda:3'), in_proj_covar=tensor([0.0018, 0.0019, 0.0019, 0.0018, 0.0020, 0.0025, 0.0020, 0.0020], device='cuda:3'), out_proj_covar=tensor([7.4195e-05, 7.6617e-05, 7.0399e-05, 7.3136e-05, 7.9719e-05, 9.3770e-05, 8.2393e-05, 7.7040e-05], device='cuda:3') 2022-12-07 13:08:53,563 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.157e+02 2.768e+02 3.459e+02 4.682e+02 1.378e+03, threshold=6.919e+02, percent-clipped=12.0 2022-12-07 13:08:56,566 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35326.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:09:00,116 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35330.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:09:13,310 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35345.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 13:09:17,077 INFO [train.py:873] (3/4) Epoch 5, batch 5100, loss[loss=0.1651, simple_loss=0.1868, pruned_loss=0.0717, over 14550.00 frames. ], tot_loss[loss=0.1793, simple_loss=0.1921, pruned_loss=0.08326, over 1960850.74 frames. ], batch size: 43, lr: 1.52e-02, grad_scale: 16.0 2022-12-07 13:09:25,063 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.73 vs. limit=2.0 2022-12-07 13:09:32,916 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35366.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 13:09:38,672 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.65 vs. limit=5.0 2022-12-07 13:09:39,943 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35374.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:10:08,807 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35406.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 13:10:15,196 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35414.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 13:10:23,148 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.147e+02 2.613e+02 3.282e+02 4.367e+02 7.735e+02, threshold=6.564e+02, percent-clipped=1.0 2022-12-07 13:10:34,130 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35435.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:10:44,984 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35448.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:10:45,711 INFO [train.py:873] (3/4) Epoch 5, batch 5200, loss[loss=0.1576, simple_loss=0.1842, pruned_loss=0.06551, over 14457.00 frames. ], tot_loss[loss=0.1796, simple_loss=0.1922, pruned_loss=0.08352, over 1966485.94 frames. ], batch size: 51, lr: 1.52e-02, grad_scale: 8.0 2022-12-07 13:11:16,694 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8234, 4.6162, 4.1495, 4.3635, 4.3345, 4.6645, 4.8223, 4.7692], device='cuda:3'), covar=tensor([0.0739, 0.0391, 0.1954, 0.2553, 0.0816, 0.0586, 0.0762, 0.0787], device='cuda:3'), in_proj_covar=tensor([0.0288, 0.0228, 0.0355, 0.0439, 0.0261, 0.0312, 0.0328, 0.0272], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 13:11:22,237 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2022-12-07 13:11:51,441 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.407e+02 2.356e+02 3.494e+02 4.184e+02 1.202e+03, threshold=6.988e+02, percent-clipped=4.0 2022-12-07 13:11:57,343 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35530.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:12:13,730 INFO [train.py:873] (3/4) Epoch 5, batch 5300, loss[loss=0.2235, simple_loss=0.1868, pruned_loss=0.1301, over 1207.00 frames. ], tot_loss[loss=0.1794, simple_loss=0.192, pruned_loss=0.08337, over 1925438.69 frames. ], batch size: 100, lr: 1.51e-02, grad_scale: 8.0 2022-12-07 13:12:51,638 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35591.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:13:06,525 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.80 vs. limit=2.0 2022-12-07 13:13:20,925 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.526e+02 2.787e+02 3.586e+02 4.482e+02 8.195e+02, threshold=7.172e+02, percent-clipped=2.0 2022-12-07 13:13:36,055 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3927, 2.1750, 3.4110, 3.3905, 3.2951, 2.1736, 3.3594, 2.6539], device='cuda:3'), covar=tensor([0.0121, 0.0326, 0.0320, 0.0144, 0.0109, 0.0522, 0.0096, 0.0392], device='cuda:3'), in_proj_covar=tensor([0.0194, 0.0192, 0.0281, 0.0228, 0.0181, 0.0239, 0.0172, 0.0235], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:3') 2022-12-07 13:13:43,966 INFO [train.py:873] (3/4) Epoch 5, batch 5400, loss[loss=0.1461, simple_loss=0.1387, pruned_loss=0.07678, over 1237.00 frames. ], tot_loss[loss=0.1793, simple_loss=0.1919, pruned_loss=0.0834, over 1928532.64 frames. ], batch size: 100, lr: 1.51e-02, grad_scale: 8.0 2022-12-07 13:13:47,663 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5444, 2.4534, 2.2317, 2.2401, 2.5154, 2.3706, 2.5319, 2.4766], device='cuda:3'), covar=tensor([0.0877, 0.0685, 0.1844, 0.2430, 0.0747, 0.0916, 0.1248, 0.1046], device='cuda:3'), in_proj_covar=tensor([0.0282, 0.0223, 0.0349, 0.0432, 0.0257, 0.0308, 0.0328, 0.0268], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 13:14:15,852 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.10 vs. limit=2.0 2022-12-07 13:14:30,614 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35701.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 13:14:51,484 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.281e+02 2.811e+02 3.690e+02 4.403e+02 1.133e+03, threshold=7.380e+02, percent-clipped=4.0 2022-12-07 13:14:57,143 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35730.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:15:13,038 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35748.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:15:13,721 INFO [train.py:873] (3/4) Epoch 5, batch 5500, loss[loss=0.1622, simple_loss=0.1856, pruned_loss=0.06942, over 14393.00 frames. ], tot_loss[loss=0.1784, simple_loss=0.1915, pruned_loss=0.08262, over 1957585.78 frames. ], batch size: 44, lr: 1.51e-02, grad_scale: 8.0 2022-12-07 13:15:33,539 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35771.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:15:42,979 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.6869, 2.6387, 3.6599, 2.1160, 2.2938, 2.8427, 1.6598, 2.8279], device='cuda:3'), covar=tensor([0.1877, 0.1018, 0.0597, 0.3403, 0.2575, 0.0926, 0.5015, 0.1042], device='cuda:3'), in_proj_covar=tensor([0.0068, 0.0076, 0.0071, 0.0081, 0.0103, 0.0063, 0.0128, 0.0070], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-07 13:15:55,937 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35796.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:16:07,093 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.57 vs. limit=5.0 2022-12-07 13:16:09,230 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35811.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:16:15,216 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4197, 1.1558, 1.3786, 0.9540, 1.2316, 1.3889, 1.1830, 1.3373], device='cuda:3'), covar=tensor([0.0345, 0.0469, 0.0466, 0.0346, 0.0626, 0.0406, 0.0211, 0.0627], device='cuda:3'), in_proj_covar=tensor([0.0106, 0.0209, 0.0116, 0.0125, 0.0100, 0.0107, 0.0087, 0.0128], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0006, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 13:16:19,583 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8538, 2.8808, 2.9985, 2.8853, 2.8420, 2.8236, 1.3402, 2.5992], device='cuda:3'), covar=tensor([0.0280, 0.0348, 0.0428, 0.0342, 0.0298, 0.0573, 0.2738, 0.0307], device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0135, 0.0123, 0.0110, 0.0165, 0.0117, 0.0148, 0.0156], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 13:16:20,154 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.171e+02 2.649e+02 3.424e+02 4.152e+02 8.919e+02, threshold=6.848e+02, percent-clipped=1.0 2022-12-07 13:16:28,195 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35832.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:16:42,804 INFO [train.py:873] (3/4) Epoch 5, batch 5600, loss[loss=0.1706, simple_loss=0.1917, pruned_loss=0.07476, over 14335.00 frames. ], tot_loss[loss=0.1787, simple_loss=0.1918, pruned_loss=0.08285, over 1971676.38 frames. ], batch size: 73, lr: 1.51e-02, grad_scale: 8.0 2022-12-07 13:17:03,259 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35872.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:17:15,338 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35886.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:17:48,145 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.365e+02 2.385e+02 3.015e+02 3.800e+02 5.921e+02, threshold=6.029e+02, percent-clipped=0.0 2022-12-07 13:18:02,942 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0646, 2.9187, 2.8604, 3.2146, 2.6599, 2.6439, 3.1619, 3.1703], device='cuda:3'), covar=tensor([0.0777, 0.0838, 0.0783, 0.0629, 0.1036, 0.0702, 0.0751, 0.0752], device='cuda:3'), in_proj_covar=tensor([0.0113, 0.0098, 0.0109, 0.0113, 0.0116, 0.0087, 0.0124, 0.0110], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 13:18:10,022 INFO [train.py:873] (3/4) Epoch 5, batch 5700, loss[loss=0.1736, simple_loss=0.1977, pruned_loss=0.07477, over 14506.00 frames. ], tot_loss[loss=0.1786, simple_loss=0.1917, pruned_loss=0.08278, over 1979457.17 frames. ], batch size: 49, lr: 1.51e-02, grad_scale: 8.0 2022-12-07 13:18:51,700 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.6284, 4.1954, 4.1116, 4.6671, 4.2809, 4.1340, 4.5985, 3.8169], device='cuda:3'), covar=tensor([0.0406, 0.1267, 0.0359, 0.0419, 0.0887, 0.0644, 0.0567, 0.0603], device='cuda:3'), in_proj_covar=tensor([0.0137, 0.0225, 0.0148, 0.0139, 0.0146, 0.0117, 0.0221, 0.0145], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 13:18:55,484 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36001.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 13:18:56,363 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0893, 3.1492, 4.1978, 2.6833, 2.4836, 2.9419, 1.7207, 3.0077], device='cuda:3'), covar=tensor([0.2665, 0.1175, 0.0493, 0.2151, 0.2555, 0.1330, 0.6006, 0.2786], device='cuda:3'), in_proj_covar=tensor([0.0071, 0.0081, 0.0075, 0.0084, 0.0108, 0.0067, 0.0136, 0.0074], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], device='cuda:3') 2022-12-07 13:19:06,815 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0533, 2.1754, 2.0632, 2.2005, 1.8446, 2.2000, 1.9413, 0.9171], device='cuda:3'), covar=tensor([0.1881, 0.0876, 0.0558, 0.0568, 0.0937, 0.0466, 0.1399, 0.3386], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0057, 0.0051, 0.0050, 0.0070, 0.0055, 0.0080, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004], device='cuda:3') 2022-12-07 13:19:08,877 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.14 vs. limit=5.0 2022-12-07 13:19:15,102 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.306e+02 2.574e+02 3.156e+02 4.047e+02 9.769e+02, threshold=6.311e+02, percent-clipped=9.0 2022-12-07 13:19:20,452 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36030.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:19:35,962 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2022-12-07 13:19:37,072 INFO [train.py:873] (3/4) Epoch 5, batch 5800, loss[loss=0.1596, simple_loss=0.1786, pruned_loss=0.0703, over 14215.00 frames. ], tot_loss[loss=0.1782, simple_loss=0.1911, pruned_loss=0.0826, over 1980603.60 frames. ], batch size: 37, lr: 1.50e-02, grad_scale: 8.0 2022-12-07 13:19:37,140 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36049.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 13:19:40,923 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2115, 2.8710, 2.7007, 1.5247, 2.6607, 2.8612, 3.1778, 2.4173], device='cuda:3'), covar=tensor([0.0535, 0.2596, 0.1103, 0.2887, 0.0879, 0.0470, 0.0893, 0.1793], device='cuda:3'), in_proj_covar=tensor([0.0107, 0.0210, 0.0119, 0.0126, 0.0101, 0.0108, 0.0089, 0.0129], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0006, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005], device='cuda:3') 2022-12-07 13:20:02,749 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36078.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:20:03,349 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.16 vs. limit=2.0 2022-12-07 13:20:42,662 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.345e+02 2.552e+02 3.407e+02 4.382e+02 1.034e+03, threshold=6.813e+02, percent-clipped=3.0 2022-12-07 13:20:45,450 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36127.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:21:00,977 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.73 vs. limit=5.0 2022-12-07 13:21:04,917 INFO [train.py:873] (3/4) Epoch 5, batch 5900, loss[loss=0.1915, simple_loss=0.2088, pruned_loss=0.08709, over 14541.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.19, pruned_loss=0.08092, over 2025224.57 frames. ], batch size: 22, lr: 1.50e-02, grad_scale: 8.0 2022-12-07 13:21:20,677 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36167.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:21:32,749 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36181.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:21:36,997 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36186.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:21:52,119 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 2022-12-07 13:21:55,164 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4603, 1.1779, 1.2362, 1.4885, 1.0067, 1.2387, 1.3287, 0.9497], device='cuda:3'), covar=tensor([0.0871, 0.2930, 0.1031, 0.1469, 0.1255, 0.0667, 0.0634, 0.1796], device='cuda:3'), in_proj_covar=tensor([0.0011, 0.0012, 0.0011, 0.0010, 0.0012, 0.0016, 0.0012, 0.0017], device='cuda:3'), out_proj_covar=tensor([6.0243e-05, 6.2279e-05, 6.0145e-05, 5.6104e-05, 6.5261e-05, 8.6185e-05, 6.9188e-05, 8.3200e-05], device='cuda:3') 2022-12-07 13:22:10,495 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.379e+02 2.756e+02 3.283e+02 4.209e+02 7.399e+02, threshold=6.567e+02, percent-clipped=1.0 2022-12-07 13:22:20,120 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36234.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:22:21,169 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.3150, 1.2962, 1.3258, 1.6321, 1.2764, 1.2572, 1.1172, 0.9927], device='cuda:3'), covar=tensor([0.0559, 0.1472, 0.0777, 0.0698, 0.1182, 0.0365, 0.0555, 0.0852], device='cuda:3'), in_proj_covar=tensor([0.0011, 0.0012, 0.0011, 0.0010, 0.0012, 0.0016, 0.0012, 0.0017], device='cuda:3'), out_proj_covar=tensor([6.0519e-05, 6.2294e-05, 6.0568e-05, 5.6581e-05, 6.5409e-05, 8.6441e-05, 7.0057e-05, 8.3208e-05], device='cuda:3') 2022-12-07 13:22:27,288 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36242.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:22:30,104 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.36 vs. limit=5.0 2022-12-07 13:22:33,228 INFO [train.py:873] (3/4) Epoch 5, batch 6000, loss[loss=0.1547, simple_loss=0.1763, pruned_loss=0.06652, over 14006.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.1897, pruned_loss=0.08103, over 1968688.28 frames. ], batch size: 23, lr: 1.50e-02, grad_scale: 8.0 2022-12-07 13:22:33,229 INFO [train.py:896] (3/4) Computing validation loss 2022-12-07 13:22:45,292 INFO [train.py:905] (3/4) Epoch 5, validation: loss=0.1225, simple_loss=0.1661, pruned_loss=0.03949, over 857387.00 frames. 2022-12-07 13:22:45,293 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-07 13:23:31,390 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.36 vs. limit=5.0 2022-12-07 13:23:51,459 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.588e+02 2.659e+02 3.309e+02 3.881e+02 6.355e+02, threshold=6.617e+02, percent-clipped=0.0 2022-12-07 13:24:13,084 INFO [train.py:873] (3/4) Epoch 5, batch 6100, loss[loss=0.1346, simple_loss=0.1657, pruned_loss=0.05176, over 13951.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.189, pruned_loss=0.07995, over 1986394.92 frames. ], batch size: 19, lr: 1.50e-02, grad_scale: 8.0 2022-12-07 13:24:16,157 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.57 vs. limit=2.0 2022-12-07 13:24:37,518 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5905, 2.5836, 3.4461, 2.1376, 2.1396, 2.5404, 1.4250, 2.4565], device='cuda:3'), covar=tensor([0.1156, 0.1150, 0.0590, 0.2486, 0.2698, 0.1230, 0.5279, 0.1322], device='cuda:3'), in_proj_covar=tensor([0.0068, 0.0076, 0.0072, 0.0082, 0.0105, 0.0066, 0.0133, 0.0071], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-07 13:24:45,915 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.0210, 0.9710, 0.8473, 1.1387, 1.1888, 0.5254, 0.8954, 1.0085], device='cuda:3'), covar=tensor([0.0496, 0.0514, 0.0499, 0.0447, 0.0410, 0.0432, 0.0685, 0.0508], device='cuda:3'), in_proj_covar=tensor([0.0019, 0.0019, 0.0019, 0.0018, 0.0019, 0.0025, 0.0020, 0.0019], device='cuda:3'), out_proj_covar=tensor([7.5575e-05, 7.6641e-05, 7.2549e-05, 7.5163e-05, 7.7331e-05, 9.4769e-05, 8.2830e-05, 7.5136e-05], device='cuda:3') 2022-12-07 13:24:49,736 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.37 vs. limit=5.0 2022-12-07 13:25:01,557 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.1525, 4.7744, 4.6542, 5.1821, 4.8892, 4.3111, 5.1727, 4.3844], device='cuda:3'), covar=tensor([0.0295, 0.0996, 0.0294, 0.0377, 0.0718, 0.0419, 0.0437, 0.0429], device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0220, 0.0145, 0.0136, 0.0140, 0.0113, 0.0213, 0.0142], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 13:25:15,137 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2759, 1.8144, 2.2686, 2.3921, 2.0061, 1.8665, 2.4820, 2.1443], device='cuda:3'), covar=tensor([0.0044, 0.0148, 0.0070, 0.0045, 0.0094, 0.0195, 0.0063, 0.0077], device='cuda:3'), in_proj_covar=tensor([0.0194, 0.0196, 0.0286, 0.0229, 0.0184, 0.0241, 0.0177, 0.0239], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:3') 2022-12-07 13:25:19,185 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.263e+02 2.518e+02 3.418e+02 4.191e+02 1.416e+03, threshold=6.837e+02, percent-clipped=13.0 2022-12-07 13:25:22,043 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36427.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:25:40,682 INFO [train.py:873] (3/4) Epoch 5, batch 6200, loss[loss=0.1786, simple_loss=0.1841, pruned_loss=0.0866, over 6947.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.1899, pruned_loss=0.08076, over 2039686.94 frames. ], batch size: 100, lr: 1.50e-02, grad_scale: 8.0 2022-12-07 13:25:56,789 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36467.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:25:57,998 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 2022-12-07 13:26:02,015 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9156, 1.6012, 1.8844, 2.0746, 1.4773, 1.7431, 2.0353, 1.9690], device='cuda:3'), covar=tensor([0.0032, 0.0073, 0.0034, 0.0021, 0.0084, 0.0110, 0.0039, 0.0033], device='cuda:3'), in_proj_covar=tensor([0.0195, 0.0197, 0.0287, 0.0231, 0.0185, 0.0243, 0.0179, 0.0238], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:3') 2022-12-07 13:26:03,698 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36475.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:26:38,826 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36515.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:26:44,670 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=10.03 vs. limit=5.0 2022-12-07 13:26:46,426 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.239e+02 2.596e+02 3.351e+02 4.378e+02 1.450e+03, threshold=6.701e+02, percent-clipped=3.0 2022-12-07 13:26:58,214 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36537.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:27:08,588 INFO [train.py:873] (3/4) Epoch 5, batch 6300, loss[loss=0.1722, simple_loss=0.1627, pruned_loss=0.09091, over 2644.00 frames. ], tot_loss[loss=0.176, simple_loss=0.1896, pruned_loss=0.08116, over 1966212.16 frames. ], batch size: 100, lr: 1.49e-02, grad_scale: 8.0 2022-12-07 13:27:57,554 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36605.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:28:00,243 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36608.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:28:03,993 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36612.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:28:05,602 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.6648, 4.2061, 4.1038, 4.5557, 4.3448, 4.1057, 4.6269, 3.8933], device='cuda:3'), covar=tensor([0.0310, 0.1092, 0.0320, 0.0499, 0.0761, 0.0547, 0.0498, 0.0457], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0216, 0.0142, 0.0135, 0.0137, 0.0113, 0.0211, 0.0138], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 13:28:14,426 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.237e+02 2.701e+02 3.163e+02 3.734e+02 7.022e+02, threshold=6.325e+02, percent-clipped=2.0 2022-12-07 13:28:35,796 INFO [train.py:873] (3/4) Epoch 5, batch 6400, loss[loss=0.1813, simple_loss=0.1941, pruned_loss=0.08428, over 14232.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.1895, pruned_loss=0.08065, over 1932265.29 frames. ], batch size: 99, lr: 1.49e-02, grad_scale: 8.0 2022-12-07 13:28:51,037 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36666.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:28:53,736 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36669.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:28:57,046 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1515, 1.5328, 4.0390, 1.3952, 3.9207, 4.1443, 3.5233, 4.3349], device='cuda:3'), covar=tensor([0.0219, 0.3131, 0.0362, 0.2679, 0.0394, 0.0354, 0.0454, 0.0203], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0162, 0.0138, 0.0171, 0.0157, 0.0150, 0.0123, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:3') 2022-12-07 13:28:57,128 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36673.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:29:06,362 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.75 vs. limit=2.0 2022-12-07 13:29:20,755 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36700.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 13:29:41,187 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.651e+02 2.894e+02 3.789e+02 4.825e+02 7.842e+02, threshold=7.578e+02, percent-clipped=4.0 2022-12-07 13:29:48,851 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4222, 1.0477, 1.1632, 1.3463, 0.9714, 1.0532, 1.0498, 0.8074], device='cuda:3'), covar=tensor([0.0486, 0.0464, 0.0675, 0.0556, 0.0936, 0.0333, 0.0530, 0.1424], device='cuda:3'), in_proj_covar=tensor([0.0012, 0.0012, 0.0011, 0.0010, 0.0013, 0.0016, 0.0012, 0.0017], device='cuda:3'), out_proj_covar=tensor([6.2065e-05, 6.4403e-05, 6.1260e-05, 5.7148e-05, 6.6181e-05, 8.7138e-05, 7.1830e-05, 8.5948e-05], device='cuda:3') 2022-12-07 13:30:01,050 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([6.1018, 5.4321, 5.6496, 5.9411, 5.7436, 4.7558, 5.9453, 5.1145], device='cuda:3'), covar=tensor([0.0234, 0.1021, 0.0207, 0.0542, 0.0477, 0.0266, 0.0452, 0.0340], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0221, 0.0144, 0.0139, 0.0140, 0.0115, 0.0214, 0.0141], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 13:30:02,670 INFO [train.py:873] (3/4) Epoch 5, batch 6500, loss[loss=0.1761, simple_loss=0.1974, pruned_loss=0.07741, over 14190.00 frames. ], tot_loss[loss=0.1777, simple_loss=0.191, pruned_loss=0.08218, over 1964521.72 frames. ], batch size: 37, lr: 1.49e-02, grad_scale: 8.0 2022-12-07 13:30:13,242 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36761.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 13:30:45,989 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.9907, 0.9160, 1.2721, 1.1647, 1.1878, 0.6509, 1.1510, 1.1184], device='cuda:3'), covar=tensor([0.1519, 0.1160, 0.0786, 0.1365, 0.1265, 0.0595, 0.1005, 0.1295], device='cuda:3'), in_proj_covar=tensor([0.0019, 0.0020, 0.0018, 0.0019, 0.0021, 0.0026, 0.0020, 0.0019], device='cuda:3'), out_proj_covar=tensor([7.7947e-05, 8.1173e-05, 7.2142e-05, 7.7805e-05, 8.3019e-05, 9.9399e-05, 8.5010e-05, 7.8209e-05], device='cuda:3') 2022-12-07 13:30:51,538 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36805.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 13:31:07,675 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.168e+02 2.577e+02 3.368e+02 4.224e+02 7.324e+02, threshold=6.736e+02, percent-clipped=0.0 2022-12-07 13:31:13,718 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1086, 2.0207, 1.8157, 1.8142, 2.0464, 1.9541, 2.0787, 2.0162], device='cuda:3'), covar=tensor([0.1001, 0.1185, 0.2248, 0.2710, 0.0978, 0.1135, 0.1401, 0.1176], device='cuda:3'), in_proj_covar=tensor([0.0289, 0.0223, 0.0357, 0.0443, 0.0257, 0.0315, 0.0329, 0.0274], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 13:31:18,646 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36837.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:31:28,831 INFO [train.py:873] (3/4) Epoch 5, batch 6600, loss[loss=0.181, simple_loss=0.1937, pruned_loss=0.08412, over 14291.00 frames. ], tot_loss[loss=0.176, simple_loss=0.1899, pruned_loss=0.08107, over 1949630.73 frames. ], batch size: 76, lr: 1.49e-02, grad_scale: 8.0 2022-12-07 13:31:44,145 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36866.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 13:32:00,561 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36885.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:32:23,319 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36911.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:32:32,343 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5303, 3.6210, 2.8548, 4.5525, 4.3054, 4.4274, 3.7984, 2.9730], device='cuda:3'), covar=tensor([0.0330, 0.1326, 0.4111, 0.0466, 0.0550, 0.1032, 0.0904, 0.3643], device='cuda:3'), in_proj_covar=tensor([0.0225, 0.0303, 0.0299, 0.0187, 0.0254, 0.0257, 0.0258, 0.0290], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 13:32:34,637 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.557e+01 2.430e+02 3.121e+02 4.365e+02 9.299e+02, threshold=6.241e+02, percent-clipped=3.0 2022-12-07 13:32:56,984 INFO [train.py:873] (3/4) Epoch 5, batch 6700, loss[loss=0.177, simple_loss=0.1855, pruned_loss=0.0842, over 14313.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.1903, pruned_loss=0.08116, over 1961251.38 frames. ], batch size: 46, lr: 1.49e-02, grad_scale: 8.0 2022-12-07 13:33:04,727 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36958.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:33:07,465 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36961.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:33:10,034 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36964.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:33:11,215 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2022-12-07 13:33:13,474 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36968.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:33:17,171 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36972.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:33:45,448 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37004.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:33:59,041 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37019.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:34:03,149 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.645e+02 2.484e+02 3.120e+02 4.102e+02 1.074e+03, threshold=6.239e+02, percent-clipped=7.0 2022-12-07 13:34:24,643 INFO [train.py:873] (3/4) Epoch 5, batch 6800, loss[loss=0.1992, simple_loss=0.2012, pruned_loss=0.09862, over 5996.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.1902, pruned_loss=0.08152, over 1917146.03 frames. ], batch size: 100, lr: 1.48e-02, grad_scale: 8.0 2022-12-07 13:34:31,281 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37056.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 13:34:38,935 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37065.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:35:06,956 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6499, 1.7826, 3.6810, 2.4051, 3.5041, 1.7545, 2.7121, 3.4273], device='cuda:3'), covar=tensor([0.0475, 0.5788, 0.0365, 0.8556, 0.0340, 0.4734, 0.1334, 0.0316], device='cuda:3'), in_proj_covar=tensor([0.0220, 0.0252, 0.0166, 0.0345, 0.0173, 0.0262, 0.0237, 0.0173], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 13:35:30,147 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.366e+02 2.403e+02 2.984e+02 3.959e+02 1.091e+03, threshold=5.968e+02, percent-clipped=8.0 2022-12-07 13:35:52,586 INFO [train.py:873] (3/4) Epoch 5, batch 6900, loss[loss=0.209, simple_loss=0.2069, pruned_loss=0.1056, over 9493.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.19, pruned_loss=0.08166, over 1912000.40 frames. ], batch size: 100, lr: 1.48e-02, grad_scale: 8.0 2022-12-07 13:36:02,893 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37161.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 13:36:04,094 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9813, 3.3280, 4.2300, 2.7809, 2.3962, 3.2466, 1.4831, 3.2401], device='cuda:3'), covar=tensor([0.1689, 0.1412, 0.0750, 0.2643, 0.2846, 0.1634, 0.7182, 0.1649], device='cuda:3'), in_proj_covar=tensor([0.0069, 0.0078, 0.0076, 0.0083, 0.0105, 0.0069, 0.0136, 0.0071], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-07 13:36:19,087 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1632, 2.8541, 2.8546, 3.1528, 3.0837, 3.1079, 3.2172, 2.6446], device='cuda:3'), covar=tensor([0.0536, 0.1439, 0.0550, 0.0605, 0.0774, 0.0517, 0.0700, 0.0675], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0223, 0.0147, 0.0138, 0.0139, 0.0116, 0.0212, 0.0141], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 13:36:56,800 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=9.30 vs. limit=5.0 2022-12-07 13:36:58,808 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.066e+02 2.434e+02 3.159e+02 4.022e+02 7.807e+02, threshold=6.317e+02, percent-clipped=6.0 2022-12-07 13:37:08,975 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7203, 1.3530, 1.7503, 1.2217, 1.3574, 1.6616, 1.5733, 1.4501], device='cuda:3'), covar=tensor([0.0469, 0.1013, 0.0595, 0.1012, 0.1022, 0.1051, 0.0455, 0.1537], device='cuda:3'), in_proj_covar=tensor([0.0111, 0.0205, 0.0118, 0.0126, 0.0104, 0.0113, 0.0090, 0.0130], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0006, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005], device='cuda:3') 2022-12-07 13:37:20,128 INFO [train.py:873] (3/4) Epoch 5, batch 7000, loss[loss=0.1824, simple_loss=0.1841, pruned_loss=0.09042, over 7798.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.1891, pruned_loss=0.08079, over 1915132.26 frames. ], batch size: 100, lr: 1.48e-02, grad_scale: 4.0 2022-12-07 13:37:31,315 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37261.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:37:33,791 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37264.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:37:36,202 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37267.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:37:37,109 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37268.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:37:37,139 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37268.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:37:52,824 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.88 vs. limit=5.0 2022-12-07 13:38:12,642 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37309.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:38:14,985 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37312.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:38:16,719 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37314.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:38:18,493 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37316.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:38:26,222 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.366e+02 2.611e+02 3.349e+02 4.594e+02 9.259e+02, threshold=6.698e+02, percent-clipped=6.0 2022-12-07 13:38:29,796 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37329.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:38:46,858 INFO [train.py:873] (3/4) Epoch 5, batch 7100, loss[loss=0.1772, simple_loss=0.1647, pruned_loss=0.0948, over 1306.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.1899, pruned_loss=0.0812, over 1950368.26 frames. ], batch size: 100, lr: 1.48e-02, grad_scale: 4.0 2022-12-07 13:38:48,525 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.1123, 4.8337, 4.6800, 5.2022, 4.7818, 4.2500, 5.1762, 5.0667], device='cuda:3'), covar=tensor([0.0566, 0.0530, 0.0540, 0.0512, 0.0593, 0.0476, 0.0585, 0.0552], device='cuda:3'), in_proj_covar=tensor([0.0109, 0.0095, 0.0109, 0.0112, 0.0115, 0.0087, 0.0125, 0.0104], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 13:38:52,928 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37356.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 13:38:56,294 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37360.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:38:59,735 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37364.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:39:14,513 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2022-12-07 13:39:34,778 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37404.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 13:39:53,448 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.411e+02 2.303e+02 3.351e+02 4.171e+02 6.616e+02, threshold=6.701e+02, percent-clipped=0.0 2022-12-07 13:39:53,712 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37425.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:40:14,167 INFO [train.py:873] (3/4) Epoch 5, batch 7200, loss[loss=0.1762, simple_loss=0.1816, pruned_loss=0.08537, over 4936.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.1901, pruned_loss=0.08158, over 1964337.81 frames. ], batch size: 100, lr: 1.48e-02, grad_scale: 8.0 2022-12-07 13:40:25,512 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37461.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 13:40:46,660 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.56 vs. limit=5.0 2022-12-07 13:41:01,427 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8619, 4.3965, 4.3540, 4.7829, 4.5355, 4.1903, 4.7440, 4.0050], device='cuda:3'), covar=tensor([0.0315, 0.0970, 0.0305, 0.0391, 0.0666, 0.0489, 0.0543, 0.0505], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0219, 0.0147, 0.0138, 0.0140, 0.0114, 0.0212, 0.0139], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 13:41:08,040 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37509.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 13:41:21,482 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.551e+02 2.612e+02 3.336e+02 4.094e+02 6.871e+02, threshold=6.672e+02, percent-clipped=1.0 2022-12-07 13:41:34,416 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9301, 2.6194, 3.7080, 2.9533, 3.6463, 3.5305, 3.5124, 3.0148], device='cuda:3'), covar=tensor([0.0411, 0.2549, 0.0821, 0.1712, 0.0703, 0.0632, 0.1292, 0.2159], device='cuda:3'), in_proj_covar=tensor([0.0276, 0.0337, 0.0369, 0.0314, 0.0351, 0.0288, 0.0341, 0.0355], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004], device='cuda:3') 2022-12-07 13:41:42,810 INFO [train.py:873] (3/4) Epoch 5, batch 7300, loss[loss=0.1757, simple_loss=0.1884, pruned_loss=0.0815, over 5991.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.1889, pruned_loss=0.08008, over 2033155.06 frames. ], batch size: 100, lr: 1.47e-02, grad_scale: 8.0 2022-12-07 13:41:58,570 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37567.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:42:04,195 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1913, 1.0698, 1.3238, 1.3671, 1.3075, 0.8802, 1.2678, 0.8891], device='cuda:3'), covar=tensor([0.0636, 0.0878, 0.0903, 0.0378, 0.1377, 0.0274, 0.0281, 0.0807], device='cuda:3'), in_proj_covar=tensor([0.0012, 0.0012, 0.0011, 0.0010, 0.0013, 0.0016, 0.0012, 0.0017], device='cuda:3'), out_proj_covar=tensor([6.3347e-05, 6.6212e-05, 6.1991e-05, 5.7911e-05, 6.8459e-05, 8.8435e-05, 7.4076e-05, 8.5812e-05], device='cuda:3') 2022-12-07 13:42:41,415 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37614.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:42:42,190 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37615.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:42:42,775 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.73 vs. limit=5.0 2022-12-07 13:42:50,460 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37624.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:42:51,224 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.575e+02 2.435e+02 3.251e+02 4.059e+02 7.905e+02, threshold=6.501e+02, percent-clipped=2.0 2022-12-07 13:43:12,151 INFO [train.py:873] (3/4) Epoch 5, batch 7400, loss[loss=0.1725, simple_loss=0.1452, pruned_loss=0.09991, over 1234.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.1895, pruned_loss=0.08139, over 1942826.35 frames. ], batch size: 100, lr: 1.47e-02, grad_scale: 8.0 2022-12-07 13:43:22,096 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37660.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:43:24,016 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37662.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:43:56,196 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8737, 4.5332, 4.3078, 4.4747, 4.5544, 4.8006, 4.9353, 4.7713], device='cuda:3'), covar=tensor([0.0847, 0.0721, 0.2155, 0.2878, 0.0788, 0.0577, 0.0963, 0.0999], device='cuda:3'), in_proj_covar=tensor([0.0292, 0.0231, 0.0361, 0.0458, 0.0272, 0.0327, 0.0334, 0.0279], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 13:44:05,021 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37708.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:44:14,198 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.0080, 4.6505, 4.4418, 4.5131, 4.5764, 4.8397, 5.0394, 4.9228], device='cuda:3'), covar=tensor([0.0789, 0.0610, 0.1771, 0.2774, 0.0693, 0.0700, 0.0814, 0.0938], device='cuda:3'), in_proj_covar=tensor([0.0288, 0.0226, 0.0356, 0.0450, 0.0270, 0.0324, 0.0328, 0.0276], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 13:44:15,052 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37720.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:44:17,055 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2022-12-07 13:44:19,313 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.338e+02 2.420e+02 3.401e+02 4.272e+02 7.052e+02, threshold=6.801e+02, percent-clipped=3.0 2022-12-07 13:44:26,002 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7170, 2.0450, 3.7618, 3.8488, 3.8128, 2.3385, 3.7538, 3.0005], device='cuda:3'), covar=tensor([0.0136, 0.0403, 0.0358, 0.0170, 0.0111, 0.0549, 0.0107, 0.0398], device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0196, 0.0290, 0.0231, 0.0184, 0.0240, 0.0179, 0.0235], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:3') 2022-12-07 13:44:31,981 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.00 vs. limit=5.0 2022-12-07 13:44:41,302 INFO [train.py:873] (3/4) Epoch 5, batch 7500, loss[loss=0.1872, simple_loss=0.2031, pruned_loss=0.08568, over 14294.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.1895, pruned_loss=0.08091, over 1962130.63 frames. ], batch size: 76, lr: 1.47e-02, grad_scale: 8.0 2022-12-07 13:46:12,628 INFO [train.py:873] (3/4) Epoch 6, batch 0, loss[loss=0.2437, simple_loss=0.2391, pruned_loss=0.1242, over 14266.00 frames. ], tot_loss[loss=0.2437, simple_loss=0.2391, pruned_loss=0.1242, over 14266.00 frames. ], batch size: 63, lr: 1.37e-02, grad_scale: 8.0 2022-12-07 13:46:12,629 INFO [train.py:896] (3/4) Computing validation loss 2022-12-07 13:46:20,010 INFO [train.py:905] (3/4) Epoch 6, validation: loss=0.1313, simple_loss=0.1749, pruned_loss=0.04388, over 857387.00 frames. 2022-12-07 13:46:20,011 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-07 13:46:31,087 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4575, 2.3500, 2.8668, 1.8353, 1.9471, 2.5707, 1.1754, 2.4529], device='cuda:3'), covar=tensor([0.0822, 0.1036, 0.0536, 0.1898, 0.2173, 0.0695, 0.5311, 0.0849], device='cuda:3'), in_proj_covar=tensor([0.0070, 0.0079, 0.0075, 0.0083, 0.0106, 0.0068, 0.0134, 0.0070], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], device='cuda:3') 2022-12-07 13:46:32,650 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 6.323e+01 1.772e+02 2.799e+02 3.695e+02 8.641e+02, threshold=5.598e+02, percent-clipped=1.0 2022-12-07 13:47:50,542 INFO [train.py:873] (3/4) Epoch 6, batch 100, loss[loss=0.1643, simple_loss=0.1838, pruned_loss=0.07235, over 14189.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.1892, pruned_loss=0.07781, over 980151.18 frames. ], batch size: 84, lr: 1.37e-02, grad_scale: 8.0 2022-12-07 13:48:01,978 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37924.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:48:02,649 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.346e+02 2.495e+02 3.181e+02 4.062e+02 9.836e+02, threshold=6.363e+02, percent-clipped=3.0 2022-12-07 13:48:44,521 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37972.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:48:50,777 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37979.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:49:19,530 INFO [train.py:873] (3/4) Epoch 6, batch 200, loss[loss=0.1724, simple_loss=0.1868, pruned_loss=0.079, over 14096.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.1885, pruned_loss=0.07857, over 1407685.48 frames. ], batch size: 22, lr: 1.37e-02, grad_scale: 8.0 2022-12-07 13:49:21,311 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.6423, 4.4441, 4.8435, 3.7902, 4.5217, 4.9710, 1.7357, 4.3550], device='cuda:3'), covar=tensor([0.0167, 0.0286, 0.0317, 0.0441, 0.0250, 0.0143, 0.3041, 0.0240], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0136, 0.0123, 0.0115, 0.0168, 0.0118, 0.0151, 0.0157], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 13:49:22,134 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.6092, 2.4610, 2.1197, 2.2433, 2.5109, 2.4470, 2.5857, 2.5070], device='cuda:3'), covar=tensor([0.0919, 0.0932, 0.2788, 0.2874, 0.0882, 0.0961, 0.1326, 0.0979], device='cuda:3'), in_proj_covar=tensor([0.0298, 0.0224, 0.0367, 0.0457, 0.0272, 0.0327, 0.0338, 0.0283], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 13:49:27,526 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=38020.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:49:31,828 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.520e+02 2.443e+02 3.000e+02 3.883e+02 6.442e+02, threshold=6.000e+02, percent-clipped=2.0 2022-12-07 13:49:45,632 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=38040.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:50:09,554 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=38068.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:50:38,691 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0106, 4.0262, 4.0394, 3.7913, 3.9550, 4.3668, 1.4200, 3.7087], device='cuda:3'), covar=tensor([0.0248, 0.0279, 0.0667, 0.0499, 0.0364, 0.0191, 0.3834, 0.0360], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0135, 0.0123, 0.0115, 0.0168, 0.0116, 0.0150, 0.0157], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 13:50:47,793 INFO [train.py:873] (3/4) Epoch 6, batch 300, loss[loss=0.1438, simple_loss=0.1704, pruned_loss=0.05861, over 14330.00 frames. ], tot_loss[loss=0.1734, simple_loss=0.1885, pruned_loss=0.07918, over 1663146.03 frames. ], batch size: 55, lr: 1.37e-02, grad_scale: 8.0 2022-12-07 13:50:59,800 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.208e+02 2.428e+02 3.008e+02 3.765e+02 6.349e+02, threshold=6.017e+02, percent-clipped=1.0 2022-12-07 13:51:05,596 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=9.70 vs. limit=5.0 2022-12-07 13:52:16,129 INFO [train.py:873] (3/4) Epoch 6, batch 400, loss[loss=0.1623, simple_loss=0.167, pruned_loss=0.07876, over 3876.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.1872, pruned_loss=0.07829, over 1785784.88 frames. ], batch size: 100, lr: 1.36e-02, grad_scale: 8.0 2022-12-07 13:52:28,766 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.176e+02 2.719e+02 3.194e+02 3.868e+02 8.474e+02, threshold=6.387e+02, percent-clipped=5.0 2022-12-07 13:52:39,029 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8588, 1.2860, 1.1803, 1.2595, 1.5948, 1.1468, 1.2163, 0.8278], device='cuda:3'), covar=tensor([0.1694, 0.1013, 0.1041, 0.1766, 0.1241, 0.0412, 0.0765, 0.2164], device='cuda:3'), in_proj_covar=tensor([0.0012, 0.0012, 0.0010, 0.0010, 0.0012, 0.0015, 0.0011, 0.0016], device='cuda:3'), out_proj_covar=tensor([6.3370e-05, 6.4954e-05, 6.0951e-05, 5.8408e-05, 6.6776e-05, 8.6588e-05, 7.1000e-05, 8.3790e-05], device='cuda:3') 2022-12-07 13:52:41,945 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.07 vs. limit=2.0 2022-12-07 13:53:29,963 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8496, 1.0827, 0.9701, 0.9278, 0.9531, 0.5284, 0.9079, 0.9001], device='cuda:3'), covar=tensor([0.1167, 0.0839, 0.0335, 0.1294, 0.1642, 0.0480, 0.0763, 0.1344], device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0020, 0.0019, 0.0019, 0.0020, 0.0026, 0.0020, 0.0019], device='cuda:3'), out_proj_covar=tensor([8.1456e-05, 8.3195e-05, 7.4744e-05, 8.0423e-05, 8.4647e-05, 1.0071e-04, 8.7758e-05, 8.1028e-05], device='cuda:3') 2022-12-07 13:53:37,629 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.96 vs. limit=5.0 2022-12-07 13:53:45,479 INFO [train.py:873] (3/4) Epoch 6, batch 500, loss[loss=0.2036, simple_loss=0.2013, pruned_loss=0.1029, over 6928.00 frames. ], tot_loss[loss=0.173, simple_loss=0.1881, pruned_loss=0.079, over 1920379.76 frames. ], batch size: 100, lr: 1.36e-02, grad_scale: 8.0 2022-12-07 13:53:57,615 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.672e+02 2.649e+02 3.367e+02 4.608e+02 8.885e+02, threshold=6.735e+02, percent-clipped=8.0 2022-12-07 13:54:06,508 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=38335.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:54:28,739 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=38360.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:55:00,403 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5942, 4.5155, 3.6238, 3.7571, 4.0081, 4.5420, 4.6778, 4.5108], device='cuda:3'), covar=tensor([0.1139, 0.0585, 0.3269, 0.4121, 0.1329, 0.1059, 0.1117, 0.1305], device='cuda:3'), in_proj_covar=tensor([0.0287, 0.0219, 0.0355, 0.0448, 0.0267, 0.0316, 0.0334, 0.0274], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 13:55:12,570 INFO [train.py:873] (3/4) Epoch 6, batch 600, loss[loss=0.1221, simple_loss=0.1544, pruned_loss=0.04492, over 14266.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.1874, pruned_loss=0.0785, over 1926628.47 frames. ], batch size: 18, lr: 1.36e-02, grad_scale: 8.0 2022-12-07 13:55:21,412 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=38421.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:55:24,768 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.074e+02 2.417e+02 3.018e+02 3.905e+02 9.436e+02, threshold=6.036e+02, percent-clipped=5.0 2022-12-07 13:56:24,558 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.79 vs. limit=2.0 2022-12-07 13:56:30,106 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 2022-12-07 13:56:36,455 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.94 vs. limit=2.0 2022-12-07 13:56:41,308 INFO [train.py:873] (3/4) Epoch 6, batch 700, loss[loss=0.1604, simple_loss=0.1685, pruned_loss=0.07611, over 4943.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.1877, pruned_loss=0.07852, over 1974898.66 frames. ], batch size: 100, lr: 1.36e-02, grad_scale: 8.0 2022-12-07 13:56:52,402 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1486, 3.8163, 3.7482, 4.1082, 4.0159, 3.5094, 4.1391, 3.5660], device='cuda:3'), covar=tensor([0.0507, 0.0884, 0.0411, 0.0481, 0.0652, 0.1457, 0.0560, 0.0527], device='cuda:3'), in_proj_covar=tensor([0.0138, 0.0226, 0.0149, 0.0141, 0.0144, 0.0116, 0.0219, 0.0142], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 13:56:53,910 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.144e+02 2.377e+02 3.088e+02 3.950e+02 6.999e+02, threshold=6.177e+02, percent-clipped=2.0 2022-12-07 13:57:12,828 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.79 vs. limit=2.0 2022-12-07 13:57:37,950 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.37 vs. limit=5.0 2022-12-07 13:58:10,603 INFO [train.py:873] (3/4) Epoch 6, batch 800, loss[loss=0.1634, simple_loss=0.1467, pruned_loss=0.09001, over 2704.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.1873, pruned_loss=0.07842, over 2004782.31 frames. ], batch size: 100, lr: 1.36e-02, grad_scale: 8.0 2022-12-07 13:58:23,022 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.240e+02 2.424e+02 2.905e+02 3.924e+02 6.932e+02, threshold=5.809e+02, percent-clipped=2.0 2022-12-07 13:58:23,344 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1877, 2.0813, 3.1917, 3.2603, 3.2095, 2.1286, 3.2558, 2.4700], device='cuda:3'), covar=tensor([0.0149, 0.0342, 0.0317, 0.0169, 0.0107, 0.0558, 0.0093, 0.0386], device='cuda:3'), in_proj_covar=tensor([0.0195, 0.0194, 0.0290, 0.0233, 0.0184, 0.0240, 0.0178, 0.0232], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:3') 2022-12-07 13:58:32,298 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=38635.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:58:39,809 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.74 vs. limit=5.0 2022-12-07 13:58:47,437 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.80 vs. limit=2.0 2022-12-07 13:59:15,089 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=38683.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:59:16,883 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5310, 2.3813, 1.8995, 2.5332, 2.2886, 2.3607, 2.1683, 2.0270], device='cuda:3'), covar=tensor([0.0384, 0.0944, 0.3052, 0.0239, 0.0801, 0.0290, 0.1171, 0.1641], device='cuda:3'), in_proj_covar=tensor([0.0226, 0.0304, 0.0297, 0.0192, 0.0259, 0.0258, 0.0256, 0.0291], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 13:59:39,507 INFO [train.py:873] (3/4) Epoch 6, batch 900, loss[loss=0.1949, simple_loss=0.1974, pruned_loss=0.09624, over 7794.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.1869, pruned_loss=0.07839, over 1950919.75 frames. ], batch size: 100, lr: 1.36e-02, grad_scale: 8.0 2022-12-07 13:59:44,655 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=38716.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 13:59:52,235 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.369e+02 2.457e+02 3.272e+02 3.954e+02 1.451e+03, threshold=6.544e+02, percent-clipped=7.0 2022-12-07 14:00:18,620 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 2022-12-07 14:00:23,548 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2022-12-07 14:01:08,898 INFO [train.py:873] (3/4) Epoch 6, batch 1000, loss[loss=0.1593, simple_loss=0.1805, pruned_loss=0.06908, over 14274.00 frames. ], tot_loss[loss=0.1718, simple_loss=0.1871, pruned_loss=0.07824, over 2005632.35 frames. ], batch size: 63, lr: 1.35e-02, grad_scale: 8.0 2022-12-07 14:01:19,489 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=38823.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:01:21,139 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.119e+02 2.592e+02 3.216e+02 4.248e+02 8.098e+02, threshold=6.432e+02, percent-clipped=2.0 2022-12-07 14:01:43,262 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1196, 2.0521, 1.7025, 1.7856, 2.0499, 2.0012, 2.0965, 2.0090], device='cuda:3'), covar=tensor([0.0870, 0.0881, 0.2623, 0.2751, 0.0952, 0.0963, 0.1449, 0.1059], device='cuda:3'), in_proj_covar=tensor([0.0310, 0.0238, 0.0379, 0.0469, 0.0281, 0.0340, 0.0351, 0.0286], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 14:02:13,877 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=38884.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:02:14,241 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.63 vs. limit=2.0 2022-12-07 14:02:37,249 INFO [train.py:873] (3/4) Epoch 6, batch 1100, loss[loss=0.2207, simple_loss=0.2167, pruned_loss=0.1124, over 7765.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.1873, pruned_loss=0.07881, over 2027146.45 frames. ], batch size: 100, lr: 1.35e-02, grad_scale: 8.0 2022-12-07 14:02:50,186 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.088e+02 2.398e+02 3.098e+02 3.888e+02 7.053e+02, threshold=6.197e+02, percent-clipped=1.0 2022-12-07 14:03:13,266 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=3.28 vs. limit=2.0 2022-12-07 14:04:07,243 INFO [train.py:873] (3/4) Epoch 6, batch 1200, loss[loss=0.1846, simple_loss=0.2002, pruned_loss=0.08444, over 14399.00 frames. ], tot_loss[loss=0.171, simple_loss=0.1862, pruned_loss=0.07785, over 2013828.05 frames. ], batch size: 41, lr: 1.35e-02, grad_scale: 8.0 2022-12-07 14:04:09,070 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7650, 1.2749, 1.6764, 1.1061, 1.3255, 1.7544, 1.6073, 1.4870], device='cuda:3'), covar=tensor([0.0542, 0.1751, 0.0689, 0.1049, 0.1175, 0.0805, 0.0426, 0.1954], device='cuda:3'), in_proj_covar=tensor([0.0109, 0.0203, 0.0119, 0.0127, 0.0108, 0.0112, 0.0088, 0.0127], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0006, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005], device='cuda:3') 2022-12-07 14:04:11,686 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39016.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:04:19,444 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.136e+01 2.396e+02 3.010e+02 3.930e+02 7.519e+02, threshold=6.020e+02, percent-clipped=3.0 2022-12-07 14:04:46,557 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5706, 3.2552, 2.8435, 2.0813, 2.9264, 3.2245, 3.5514, 2.6321], device='cuda:3'), covar=tensor([0.0603, 0.2186, 0.1348, 0.2764, 0.1265, 0.0580, 0.1128, 0.1719], device='cuda:3'), in_proj_covar=tensor([0.0108, 0.0203, 0.0118, 0.0127, 0.0107, 0.0112, 0.0089, 0.0127], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0006, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005], device='cuda:3') 2022-12-07 14:04:54,528 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=39064.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:05:14,161 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.4348, 4.9422, 4.7157, 5.3650, 4.9202, 4.3462, 5.3000, 4.4605], device='cuda:3'), covar=tensor([0.0278, 0.0949, 0.0279, 0.0407, 0.0802, 0.0413, 0.0481, 0.0416], device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0221, 0.0146, 0.0140, 0.0147, 0.0115, 0.0218, 0.0137], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 14:05:35,513 INFO [train.py:873] (3/4) Epoch 6, batch 1300, loss[loss=0.1733, simple_loss=0.1921, pruned_loss=0.0772, over 14292.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.1861, pruned_loss=0.07761, over 2018663.51 frames. ], batch size: 76, lr: 1.35e-02, grad_scale: 8.0 2022-12-07 14:05:46,702 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2022-12-07 14:05:48,564 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.179e+02 2.329e+02 2.773e+02 3.737e+02 7.179e+02, threshold=5.547e+02, percent-clipped=2.0 2022-12-07 14:06:14,118 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0660, 2.0018, 1.7260, 1.7892, 2.0091, 1.9880, 2.0527, 2.0019], device='cuda:3'), covar=tensor([0.0973, 0.0768, 0.2255, 0.2327, 0.1141, 0.1091, 0.1267, 0.1059], device='cuda:3'), in_proj_covar=tensor([0.0304, 0.0225, 0.0368, 0.0460, 0.0272, 0.0334, 0.0336, 0.0281], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 14:06:36,465 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39179.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:06:41,858 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39185.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:06:43,962 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.76 vs. limit=2.0 2022-12-07 14:07:05,237 INFO [train.py:873] (3/4) Epoch 6, batch 1400, loss[loss=0.1686, simple_loss=0.1892, pruned_loss=0.07397, over 14260.00 frames. ], tot_loss[loss=0.1705, simple_loss=0.1861, pruned_loss=0.07746, over 2036666.14 frames. ], batch size: 28, lr: 1.35e-02, grad_scale: 8.0 2022-12-07 14:07:17,497 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.489e+02 2.370e+02 2.964e+02 3.794e+02 7.798e+02, threshold=5.929e+02, percent-clipped=9.0 2022-12-07 14:07:25,666 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4435, 2.7107, 4.4596, 4.5646, 4.8852, 2.9088, 4.6142, 3.7127], device='cuda:3'), covar=tensor([0.0107, 0.0348, 0.0298, 0.0124, 0.0067, 0.0540, 0.0080, 0.0319], device='cuda:3'), in_proj_covar=tensor([0.0203, 0.0202, 0.0299, 0.0237, 0.0189, 0.0247, 0.0185, 0.0237], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:3') 2022-12-07 14:07:36,776 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39246.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:08:10,673 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39284.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:08:24,845 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.00 vs. limit=5.0 2022-12-07 14:08:33,848 INFO [train.py:873] (3/4) Epoch 6, batch 1500, loss[loss=0.2011, simple_loss=0.1729, pruned_loss=0.1147, over 1248.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.1858, pruned_loss=0.07738, over 2019962.21 frames. ], batch size: 100, lr: 1.34e-02, grad_scale: 16.0 2022-12-07 14:08:33,950 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5895, 2.4896, 2.5170, 2.6742, 2.5766, 2.3680, 1.3432, 2.3278], device='cuda:3'), covar=tensor([0.0432, 0.0502, 0.0747, 0.0364, 0.0485, 0.1167, 0.3034, 0.0483], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0137, 0.0123, 0.0116, 0.0167, 0.0118, 0.0149, 0.0156], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 14:08:46,906 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.110e+02 2.464e+02 2.970e+02 3.813e+02 8.133e+02, threshold=5.940e+02, percent-clipped=4.0 2022-12-07 14:09:04,232 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39345.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:09:24,353 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5949, 4.2011, 4.2014, 4.6097, 4.3678, 4.0008, 4.5495, 3.8038], device='cuda:3'), covar=tensor([0.0385, 0.0934, 0.0325, 0.0423, 0.0648, 0.0770, 0.0558, 0.0473], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0218, 0.0145, 0.0140, 0.0145, 0.0115, 0.0215, 0.0137], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 14:09:44,605 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9227, 1.5596, 4.6591, 4.2454, 4.2897, 4.7840, 4.5078, 4.7370], device='cuda:3'), covar=tensor([0.1197, 0.1274, 0.0061, 0.0121, 0.0105, 0.0068, 0.0061, 0.0090], device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0153, 0.0105, 0.0144, 0.0119, 0.0123, 0.0096, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-07 14:09:56,166 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.7138, 0.5640, 0.5604, 0.7811, 0.7659, 0.1246, 0.6265, 0.6431], device='cuda:3'), covar=tensor([0.0253, 0.0323, 0.0155, 0.0228, 0.0134, 0.0069, 0.0267, 0.0351], device='cuda:3'), in_proj_covar=tensor([0.0018, 0.0020, 0.0018, 0.0019, 0.0020, 0.0026, 0.0020, 0.0019], device='cuda:3'), out_proj_covar=tensor([7.8773e-05, 8.3344e-05, 7.4148e-05, 8.0355e-05, 8.3491e-05, 1.0200e-04, 8.7740e-05, 8.0522e-05], device='cuda:3') 2022-12-07 14:10:03,118 INFO [train.py:873] (3/4) Epoch 6, batch 1600, loss[loss=0.1615, simple_loss=0.1777, pruned_loss=0.07263, over 13880.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.1854, pruned_loss=0.07678, over 1988150.61 frames. ], batch size: 20, lr: 1.34e-02, grad_scale: 8.0 2022-12-07 14:10:16,105 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.549e+01 2.367e+02 3.070e+02 3.991e+02 2.269e+03, threshold=6.141e+02, percent-clipped=9.0 2022-12-07 14:10:30,183 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.37 vs. limit=5.0 2022-12-07 14:10:43,918 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.5086, 0.9381, 1.4252, 0.9660, 1.2228, 1.3491, 1.1572, 1.2225], device='cuda:3'), covar=tensor([0.0401, 0.1117, 0.0553, 0.0389, 0.0998, 0.0665, 0.0313, 0.1214], device='cuda:3'), in_proj_covar=tensor([0.0110, 0.0202, 0.0119, 0.0125, 0.0111, 0.0113, 0.0088, 0.0127], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0006, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005], device='cuda:3') 2022-12-07 14:10:49,272 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39463.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:11:03,661 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39479.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:11:13,639 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9156, 4.2212, 4.7430, 4.8593, 4.6084, 4.2541, 4.6907, 3.8197], device='cuda:3'), covar=tensor([0.0482, 0.1368, 0.0448, 0.0687, 0.0773, 0.0526, 0.0847, 0.0790], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0214, 0.0143, 0.0138, 0.0143, 0.0115, 0.0212, 0.0135], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 14:11:25,995 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2022-12-07 14:11:31,578 INFO [train.py:873] (3/4) Epoch 6, batch 1700, loss[loss=0.2073, simple_loss=0.1783, pruned_loss=0.1182, over 2607.00 frames. ], tot_loss[loss=0.1696, simple_loss=0.1856, pruned_loss=0.07674, over 1979712.74 frames. ], batch size: 100, lr: 1.34e-02, grad_scale: 8.0 2022-12-07 14:11:43,982 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39524.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:11:45,580 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.568e+02 2.465e+02 3.198e+02 4.096e+02 7.326e+02, threshold=6.396e+02, percent-clipped=2.0 2022-12-07 14:11:46,567 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=39527.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:11:58,765 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39541.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:12:08,760 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1174, 1.1458, 1.2358, 0.8811, 0.8865, 0.9111, 0.5972, 0.7381], device='cuda:3'), covar=tensor([0.0529, 0.0336, 0.0477, 0.0433, 0.0604, 0.0436, 0.0333, 0.0830], device='cuda:3'), in_proj_covar=tensor([0.0012, 0.0012, 0.0011, 0.0011, 0.0013, 0.0016, 0.0012, 0.0017], device='cuda:3'), out_proj_covar=tensor([6.7039e-05, 6.9704e-05, 6.4474e-05, 6.5481e-05, 7.1367e-05, 9.1747e-05, 7.8579e-05, 8.9472e-05], device='cuda:3') 2022-12-07 14:13:01,689 INFO [train.py:873] (3/4) Epoch 6, batch 1800, loss[loss=0.1686, simple_loss=0.1928, pruned_loss=0.07221, over 13985.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.1861, pruned_loss=0.07716, over 2014343.47 frames. ], batch size: 26, lr: 1.34e-02, grad_scale: 4.0 2022-12-07 14:13:15,824 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.478e+02 2.436e+02 3.001e+02 3.832e+02 8.283e+02, threshold=6.003e+02, percent-clipped=3.0 2022-12-07 14:13:25,780 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.6887, 4.3927, 4.3074, 4.7671, 4.2140, 3.8549, 4.7315, 4.7135], device='cuda:3'), covar=tensor([0.0579, 0.0496, 0.0499, 0.0430, 0.0616, 0.0518, 0.0524, 0.0512], device='cuda:3'), in_proj_covar=tensor([0.0109, 0.0096, 0.0110, 0.0113, 0.0114, 0.0087, 0.0127, 0.0107], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 14:13:26,806 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39639.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 14:13:27,565 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39640.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:14:20,109 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39700.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 14:14:29,789 INFO [train.py:873] (3/4) Epoch 6, batch 1900, loss[loss=0.1589, simple_loss=0.1844, pruned_loss=0.0667, over 14566.00 frames. ], tot_loss[loss=0.171, simple_loss=0.1862, pruned_loss=0.07792, over 1975751.16 frames. ], batch size: 23, lr: 1.34e-02, grad_scale: 4.0 2022-12-07 14:14:44,080 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.151e+02 2.287e+02 3.025e+02 3.865e+02 4.213e+03, threshold=6.050e+02, percent-clipped=8.0 2022-12-07 14:14:56,293 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.00 vs. limit=5.0 2022-12-07 14:15:19,665 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39767.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:15:19,714 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.2299, 3.6714, 2.8445, 4.6176, 4.2763, 4.3713, 3.4248, 3.0558], device='cuda:3'), covar=tensor([0.0982, 0.1283, 0.4722, 0.0390, 0.0715, 0.1258, 0.1545, 0.4210], device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0304, 0.0300, 0.0193, 0.0260, 0.0258, 0.0261, 0.0290], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 14:15:58,844 INFO [train.py:873] (3/4) Epoch 6, batch 2000, loss[loss=0.1744, simple_loss=0.1762, pruned_loss=0.08629, over 3866.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.1868, pruned_loss=0.078, over 2008950.35 frames. ], batch size: 100, lr: 1.34e-02, grad_scale: 8.0 2022-12-07 14:16:02,576 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39815.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:16:05,935 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39819.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:16:12,943 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.379e+02 2.241e+02 2.938e+02 3.748e+02 1.021e+03, threshold=5.875e+02, percent-clipped=4.0 2022-12-07 14:16:14,116 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39828.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:16:20,916 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5172, 1.6711, 2.8188, 2.0386, 2.7705, 1.7176, 2.1974, 2.4568], device='cuda:3'), covar=tensor([0.0870, 0.4600, 0.0361, 0.6251, 0.0494, 0.3680, 0.1275, 0.0393], device='cuda:3'), in_proj_covar=tensor([0.0225, 0.0253, 0.0177, 0.0346, 0.0183, 0.0265, 0.0245, 0.0176], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 14:16:25,408 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39841.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:16:56,538 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39876.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:17:08,345 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=39889.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:17:27,451 INFO [train.py:873] (3/4) Epoch 6, batch 2100, loss[loss=0.1646, simple_loss=0.1821, pruned_loss=0.07359, over 10349.00 frames. ], tot_loss[loss=0.17, simple_loss=0.1855, pruned_loss=0.07728, over 2014111.21 frames. ], batch size: 100, lr: 1.33e-02, grad_scale: 8.0 2022-12-07 14:17:29,344 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3723, 2.2752, 2.2539, 2.3394, 2.3123, 2.0344, 1.3225, 2.0190], device='cuda:3'), covar=tensor([0.0472, 0.0577, 0.0756, 0.0462, 0.0528, 0.1177, 0.2830, 0.0528], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0141, 0.0123, 0.0120, 0.0171, 0.0120, 0.0153, 0.0160], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 14:17:42,101 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.442e+01 2.310e+02 2.903e+02 3.569e+02 8.046e+02, threshold=5.805e+02, percent-clipped=4.0 2022-12-07 14:17:53,528 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39940.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:18:02,197 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8113, 3.3276, 2.6205, 3.9287, 3.7221, 3.7799, 3.1545, 2.6771], device='cuda:3'), covar=tensor([0.0500, 0.1281, 0.3695, 0.0317, 0.0705, 0.1004, 0.1085, 0.3767], device='cuda:3'), in_proj_covar=tensor([0.0224, 0.0297, 0.0295, 0.0190, 0.0254, 0.0256, 0.0255, 0.0281], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 14:18:17,711 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7804, 1.3499, 1.7890, 1.2104, 1.4250, 1.7613, 1.5591, 1.4855], device='cuda:3'), covar=tensor([0.0394, 0.0758, 0.0552, 0.0991, 0.1085, 0.0579, 0.0325, 0.1532], device='cuda:3'), in_proj_covar=tensor([0.0112, 0.0203, 0.0118, 0.0126, 0.0107, 0.0114, 0.0087, 0.0131], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0006, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005], device='cuda:3') 2022-12-07 14:18:21,544 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.74 vs. limit=2.0 2022-12-07 14:18:35,847 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=39988.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:18:41,882 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39995.0, num_to_drop=1, layers_to_drop={3} 2022-12-07 14:18:45,297 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.27 vs. limit=2.0 2022-12-07 14:18:57,535 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2022-12-07 14:19:00,406 INFO [train.py:873] (3/4) Epoch 6, batch 2200, loss[loss=0.1688, simple_loss=0.1682, pruned_loss=0.08465, over 6040.00 frames. ], tot_loss[loss=0.1718, simple_loss=0.186, pruned_loss=0.07877, over 1899544.52 frames. ], batch size: 100, lr: 1.33e-02, grad_scale: 8.0 2022-12-07 14:19:14,065 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.175e+01 2.567e+02 3.218e+02 4.341e+02 1.607e+03, threshold=6.436e+02, percent-clipped=13.0 2022-12-07 14:20:27,392 INFO [train.py:873] (3/4) Epoch 6, batch 2300, loss[loss=0.1788, simple_loss=0.167, pruned_loss=0.09533, over 1221.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.1855, pruned_loss=0.0781, over 1893780.04 frames. ], batch size: 100, lr: 1.33e-02, grad_scale: 8.0 2022-12-07 14:20:35,014 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40119.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:20:38,494 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40123.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:20:41,820 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.659e+01 2.460e+02 3.103e+02 4.082e+02 7.462e+02, threshold=6.206e+02, percent-clipped=2.0 2022-12-07 14:20:44,059 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9844, 1.5826, 1.9393, 1.2926, 1.5649, 1.9668, 1.7529, 1.5893], device='cuda:3'), covar=tensor([0.0534, 0.1059, 0.0785, 0.1510, 0.1134, 0.0651, 0.0507, 0.1837], device='cuda:3'), in_proj_covar=tensor([0.0114, 0.0204, 0.0119, 0.0127, 0.0107, 0.0113, 0.0088, 0.0130], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0006, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005], device='cuda:3') 2022-12-07 14:20:52,097 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.0194, 1.6848, 4.4549, 2.3048, 4.3940, 4.7121, 4.4604, 5.3129], device='cuda:3'), covar=tensor([0.0140, 0.2750, 0.0390, 0.1888, 0.0260, 0.0303, 0.0198, 0.0118], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0157, 0.0139, 0.0170, 0.0154, 0.0151, 0.0125, 0.0126], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:3') 2022-12-07 14:21:18,091 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40167.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:21:21,663 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40171.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:21:42,870 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40195.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:21:57,511 INFO [train.py:873] (3/4) Epoch 6, batch 2400, loss[loss=0.1877, simple_loss=0.1955, pruned_loss=0.08992, over 14291.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.1855, pruned_loss=0.07793, over 1922706.02 frames. ], batch size: 76, lr: 1.33e-02, grad_scale: 8.0 2022-12-07 14:22:03,776 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40218.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:22:11,490 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.489e+02 2.289e+02 3.018e+02 4.008e+02 1.305e+03, threshold=6.036e+02, percent-clipped=3.0 2022-12-07 14:22:14,612 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.6252, 0.4815, 0.4971, 0.7018, 0.6180, 0.1526, 0.4779, 0.5970], device='cuda:3'), covar=tensor([0.0223, 0.0446, 0.0194, 0.0340, 0.0209, 0.0234, 0.0736, 0.0348], device='cuda:3'), in_proj_covar=tensor([0.0018, 0.0020, 0.0018, 0.0019, 0.0019, 0.0026, 0.0020, 0.0019], device='cuda:3'), out_proj_covar=tensor([7.8991e-05, 8.4114e-05, 7.3478e-05, 7.9739e-05, 8.0979e-05, 1.0293e-04, 8.6910e-05, 7.9967e-05], device='cuda:3') 2022-12-07 14:22:31,003 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 2022-12-07 14:22:36,849 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.02 vs. limit=5.0 2022-12-07 14:22:37,332 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40256.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:22:51,146 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40272.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:22:52,125 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6015, 2.1509, 3.5779, 3.6992, 3.6541, 2.3215, 3.6169, 2.9557], device='cuda:3'), covar=tensor([0.0153, 0.0408, 0.0360, 0.0208, 0.0115, 0.0594, 0.0108, 0.0434], device='cuda:3'), in_proj_covar=tensor([0.0204, 0.0206, 0.0302, 0.0241, 0.0191, 0.0251, 0.0190, 0.0242], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:3') 2022-12-07 14:22:57,565 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40279.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:23:11,485 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40295.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 14:23:25,882 INFO [train.py:873] (3/4) Epoch 6, batch 2500, loss[loss=0.1908, simple_loss=0.1666, pruned_loss=0.1075, over 1239.00 frames. ], tot_loss[loss=0.1689, simple_loss=0.185, pruned_loss=0.07643, over 1968254.48 frames. ], batch size: 100, lr: 1.33e-02, grad_scale: 8.0 2022-12-07 14:23:39,898 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.231e+02 2.352e+02 2.913e+02 3.857e+02 7.423e+02, threshold=5.826e+02, percent-clipped=3.0 2022-12-07 14:23:43,163 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40330.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:23:45,755 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40333.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:23:53,914 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40343.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 14:24:36,861 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40391.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:24:54,949 INFO [train.py:873] (3/4) Epoch 6, batch 2600, loss[loss=0.1767, simple_loss=0.193, pruned_loss=0.08017, over 13927.00 frames. ], tot_loss[loss=0.171, simple_loss=0.1857, pruned_loss=0.07815, over 1890392.27 frames. ], batch size: 23, lr: 1.33e-02, grad_scale: 8.0 2022-12-07 14:25:05,691 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40423.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:25:08,967 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.181e+02 2.477e+02 3.292e+02 4.100e+02 6.300e+02, threshold=6.583e+02, percent-clipped=3.0 2022-12-07 14:25:12,221 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.00 vs. limit=5.0 2022-12-07 14:25:42,474 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8554, 1.3258, 2.1138, 1.2940, 2.0140, 2.0517, 1.7604, 2.0831], device='cuda:3'), covar=tensor([0.0260, 0.1417, 0.0257, 0.1270, 0.0305, 0.0347, 0.0531, 0.0204], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0155, 0.0139, 0.0166, 0.0155, 0.0149, 0.0123, 0.0125], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:3') 2022-12-07 14:25:44,438 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2022-12-07 14:25:47,491 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40471.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:25:47,609 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40471.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:26:06,896 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40492.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:26:23,200 INFO [train.py:873] (3/4) Epoch 6, batch 2700, loss[loss=0.1467, simple_loss=0.1706, pruned_loss=0.06139, over 14187.00 frames. ], tot_loss[loss=0.1712, simple_loss=0.1858, pruned_loss=0.07829, over 1884258.91 frames. ], batch size: 84, lr: 1.33e-02, grad_scale: 8.0 2022-12-07 14:26:30,855 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40519.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:26:37,691 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.473e+02 2.565e+02 3.016e+02 3.792e+02 1.283e+03, threshold=6.032e+02, percent-clipped=5.0 2022-12-07 14:26:58,893 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40551.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:27:00,565 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40553.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:27:07,643 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8447, 3.5952, 3.5569, 3.8961, 3.4485, 3.0998, 3.8784, 3.8459], device='cuda:3'), covar=tensor([0.0632, 0.0664, 0.0632, 0.0540, 0.0874, 0.0642, 0.0617, 0.0591], device='cuda:3'), in_proj_covar=tensor([0.0113, 0.0099, 0.0115, 0.0116, 0.0117, 0.0090, 0.0129, 0.0110], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 14:27:19,399 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40574.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:27:34,966 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1117, 2.0346, 1.8040, 1.8100, 2.0573, 2.0443, 2.0925, 2.0477], device='cuda:3'), covar=tensor([0.0952, 0.1265, 0.2574, 0.2917, 0.1051, 0.1064, 0.1438, 0.1088], device='cuda:3'), in_proj_covar=tensor([0.0305, 0.0232, 0.0368, 0.0466, 0.0278, 0.0344, 0.0347, 0.0289], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 14:27:52,084 INFO [train.py:873] (3/4) Epoch 6, batch 2800, loss[loss=0.1614, simple_loss=0.1814, pruned_loss=0.07067, over 14420.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.1855, pruned_loss=0.07734, over 1972346.08 frames. ], batch size: 73, lr: 1.32e-02, grad_scale: 8.0 2022-12-07 14:28:05,950 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.364e+02 2.405e+02 3.321e+02 4.166e+02 7.949e+02, threshold=6.642e+02, percent-clipped=7.0 2022-12-07 14:28:06,164 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8908, 1.6039, 1.9555, 1.8469, 2.1453, 1.8766, 1.6904, 1.8667], device='cuda:3'), covar=tensor([0.0278, 0.0790, 0.0157, 0.0275, 0.0155, 0.0294, 0.0162, 0.0185], device='cuda:3'), in_proj_covar=tensor([0.0286, 0.0337, 0.0385, 0.0314, 0.0358, 0.0294, 0.0347, 0.0350], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 14:28:06,940 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40628.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:28:14,230 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8441, 2.0296, 2.2526, 1.3317, 1.5338, 2.0220, 1.2641, 1.9426], device='cuda:3'), covar=tensor([0.1371, 0.1428, 0.0634, 0.3290, 0.2983, 0.0662, 0.3842, 0.0773], device='cuda:3'), in_proj_covar=tensor([0.0072, 0.0078, 0.0075, 0.0083, 0.0104, 0.0068, 0.0132, 0.0073], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], device='cuda:3') 2022-12-07 14:28:21,205 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2022-12-07 14:28:27,742 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40651.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:28:35,053 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7621, 1.8811, 2.5561, 2.3402, 2.7549, 2.5310, 2.4300, 2.2771], device='cuda:3'), covar=tensor([0.0329, 0.2125, 0.0526, 0.1320, 0.0352, 0.0498, 0.0480, 0.1193], device='cuda:3'), in_proj_covar=tensor([0.0282, 0.0332, 0.0378, 0.0308, 0.0351, 0.0290, 0.0341, 0.0345], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 14:28:40,290 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3316, 4.1721, 3.8294, 3.8929, 4.1165, 4.2399, 4.3892, 4.2855], device='cuda:3'), covar=tensor([0.0892, 0.0519, 0.1883, 0.2869, 0.0729, 0.0666, 0.0805, 0.0919], device='cuda:3'), in_proj_covar=tensor([0.0304, 0.0229, 0.0368, 0.0465, 0.0274, 0.0343, 0.0347, 0.0285], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 14:28:51,131 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.49 vs. limit=5.0 2022-12-07 14:28:58,692 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40686.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:29:20,280 INFO [train.py:873] (3/4) Epoch 6, batch 2900, loss[loss=0.1554, simple_loss=0.1694, pruned_loss=0.07071, over 5987.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.1853, pruned_loss=0.07726, over 1995846.07 frames. ], batch size: 100, lr: 1.32e-02, grad_scale: 8.0 2022-12-07 14:29:21,699 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40712.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:29:27,482 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.63 vs. limit=2.0 2022-12-07 14:29:34,595 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.325e+02 2.437e+02 2.821e+02 3.804e+02 6.190e+02, threshold=5.643e+02, percent-clipped=0.0 2022-12-07 14:29:39,164 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.89 vs. limit=2.0 2022-12-07 14:30:13,376 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3341, 2.4042, 2.2290, 2.4559, 1.8163, 2.6217, 2.3285, 1.1134], device='cuda:3'), covar=tensor([0.1871, 0.0756, 0.1290, 0.0832, 0.1204, 0.0534, 0.1437, 0.3643], device='cuda:3'), in_proj_covar=tensor([0.0159, 0.0062, 0.0051, 0.0054, 0.0077, 0.0057, 0.0084, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0005], device='cuda:3') 2022-12-07 14:30:22,861 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6027, 1.6717, 3.7506, 2.4117, 3.5906, 1.7983, 2.8222, 3.4730], device='cuda:3'), covar=tensor([0.0629, 0.6136, 0.0573, 0.9310, 0.0527, 0.4750, 0.1490, 0.0494], device='cuda:3'), in_proj_covar=tensor([0.0223, 0.0246, 0.0172, 0.0336, 0.0184, 0.0257, 0.0236, 0.0180], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 14:30:31,848 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3890, 1.7023, 2.6633, 2.1473, 2.5603, 1.6377, 2.3023, 2.4264], device='cuda:3'), covar=tensor([0.1020, 0.4624, 0.0280, 0.4229, 0.0526, 0.2821, 0.0857, 0.0501], device='cuda:3'), in_proj_covar=tensor([0.0221, 0.0245, 0.0171, 0.0335, 0.0183, 0.0256, 0.0234, 0.0178], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 14:30:48,712 INFO [train.py:873] (3/4) Epoch 6, batch 3000, loss[loss=0.1606, simple_loss=0.1721, pruned_loss=0.07453, over 5944.00 frames. ], tot_loss[loss=0.1683, simple_loss=0.184, pruned_loss=0.0763, over 1886852.43 frames. ], batch size: 100, lr: 1.32e-02, grad_scale: 8.0 2022-12-07 14:30:48,712 INFO [train.py:896] (3/4) Computing validation loss 2022-12-07 14:31:04,488 INFO [train.py:905] (3/4) Epoch 6, validation: loss=0.1224, simple_loss=0.1659, pruned_loss=0.03945, over 857387.00 frames. 2022-12-07 14:31:04,489 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-07 14:31:15,210 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 2022-12-07 14:31:19,219 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.206e+02 2.431e+02 3.272e+02 4.116e+02 8.676e+02, threshold=6.543e+02, percent-clipped=8.0 2022-12-07 14:31:28,638 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.97 vs. limit=2.0 2022-12-07 14:31:37,468 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40848.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:31:40,148 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40851.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:31:53,080 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40865.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:32:00,802 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40874.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:32:15,588 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.9502, 1.0413, 1.0407, 1.2160, 0.9525, 0.5815, 0.9996, 1.0341], device='cuda:3'), covar=tensor([0.1489, 0.0936, 0.0368, 0.0637, 0.1314, 0.0692, 0.0710, 0.0931], device='cuda:3'), in_proj_covar=tensor([0.0019, 0.0021, 0.0018, 0.0019, 0.0020, 0.0027, 0.0020, 0.0019], device='cuda:3'), out_proj_covar=tensor([8.3383e-05, 8.7623e-05, 7.6428e-05, 8.2282e-05, 8.5344e-05, 1.0676e-04, 9.0110e-05, 8.1833e-05], device='cuda:3') 2022-12-07 14:32:16,766 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0005, 3.9016, 4.2376, 3.5767, 3.9655, 4.2119, 1.5118, 3.7314], device='cuda:3'), covar=tensor([0.0172, 0.0258, 0.0313, 0.0416, 0.0266, 0.0168, 0.2829, 0.0263], device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0139, 0.0122, 0.0117, 0.0169, 0.0117, 0.0150, 0.0158], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 14:32:22,710 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40899.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:32:23,728 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40900.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:32:33,300 INFO [train.py:873] (3/4) Epoch 6, batch 3100, loss[loss=0.1943, simple_loss=0.1978, pruned_loss=0.09534, over 8650.00 frames. ], tot_loss[loss=0.1677, simple_loss=0.184, pruned_loss=0.07568, over 1988563.35 frames. ], batch size: 100, lr: 1.32e-02, grad_scale: 8.0 2022-12-07 14:32:42,996 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40922.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:32:46,537 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40926.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:32:47,205 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.239e+02 2.510e+02 3.195e+02 3.907e+02 1.074e+03, threshold=6.390e+02, percent-clipped=2.0 2022-12-07 14:32:48,230 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40928.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:32:57,912 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.9242, 0.7898, 0.7202, 0.9756, 0.9151, 0.4163, 0.8811, 0.8174], device='cuda:3'), covar=tensor([0.0541, 0.0706, 0.0247, 0.0524, 0.0281, 0.0258, 0.0623, 0.0382], device='cuda:3'), in_proj_covar=tensor([0.0019, 0.0020, 0.0018, 0.0019, 0.0019, 0.0026, 0.0020, 0.0019], device='cuda:3'), out_proj_covar=tensor([8.1878e-05, 8.5380e-05, 7.5101e-05, 8.1518e-05, 8.3317e-05, 1.0479e-04, 8.8188e-05, 8.0211e-05], device='cuda:3') 2022-12-07 14:33:17,787 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40961.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 14:33:30,777 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40976.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:33:40,131 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40986.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:33:49,437 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40997.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:33:58,661 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41007.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:34:01,822 INFO [train.py:873] (3/4) Epoch 6, batch 3200, loss[loss=0.1675, simple_loss=0.1552, pruned_loss=0.0899, over 2721.00 frames. ], tot_loss[loss=0.1683, simple_loss=0.1843, pruned_loss=0.07612, over 1929442.91 frames. ], batch size: 100, lr: 1.32e-02, grad_scale: 8.0 2022-12-07 14:34:16,161 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.950e+01 2.679e+02 3.334e+02 4.339e+02 1.581e+03, threshold=6.667e+02, percent-clipped=5.0 2022-12-07 14:34:20,500 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.62 vs. limit=2.0 2022-12-07 14:34:22,473 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41034.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:34:32,047 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41045.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:34:43,637 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41058.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:35:10,257 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0435, 1.6383, 3.9431, 3.6820, 3.7724, 3.9275, 3.2511, 4.0631], device='cuda:3'), covar=tensor([0.1175, 0.1290, 0.0083, 0.0141, 0.0143, 0.0106, 0.0180, 0.0081], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0156, 0.0107, 0.0147, 0.0120, 0.0126, 0.0097, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-07 14:35:23,142 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41102.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:35:26,800 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41106.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:35:30,960 INFO [train.py:873] (3/4) Epoch 6, batch 3300, loss[loss=0.1867, simple_loss=0.2011, pruned_loss=0.08619, over 14324.00 frames. ], tot_loss[loss=0.168, simple_loss=0.1841, pruned_loss=0.07593, over 1875140.95 frames. ], batch size: 55, lr: 1.32e-02, grad_scale: 8.0 2022-12-07 14:35:45,225 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.442e+02 2.460e+02 3.115e+02 3.768e+02 7.967e+02, threshold=6.230e+02, percent-clipped=3.0 2022-12-07 14:36:04,028 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41148.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:36:09,105 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1193, 1.2559, 1.2874, 1.2050, 1.5791, 1.0451, 0.6838, 0.8579], device='cuda:3'), covar=tensor([0.0611, 0.0702, 0.0600, 0.0422, 0.0505, 0.0201, 0.0262, 0.0889], device='cuda:3'), in_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0010, 0.0012, 0.0015, 0.0012, 0.0016], device='cuda:3'), out_proj_covar=tensor([6.6279e-05, 6.9557e-05, 6.4276e-05, 6.3963e-05, 6.8739e-05, 9.1601e-05, 7.7664e-05, 8.8854e-05], device='cuda:3') 2022-12-07 14:36:17,000 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41163.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:36:19,986 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.94 vs. limit=2.0 2022-12-07 14:36:46,074 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41196.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:36:59,352 INFO [train.py:873] (3/4) Epoch 6, batch 3400, loss[loss=0.1508, simple_loss=0.167, pruned_loss=0.06727, over 9480.00 frames. ], tot_loss[loss=0.1683, simple_loss=0.184, pruned_loss=0.07625, over 1870724.58 frames. ], batch size: 100, lr: 1.31e-02, grad_scale: 8.0 2022-12-07 14:37:08,546 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41221.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:37:13,996 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.559e+02 2.427e+02 3.173e+02 3.892e+02 5.965e+02, threshold=6.346e+02, percent-clipped=0.0 2022-12-07 14:37:39,695 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41256.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 14:38:25,255 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41307.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:38:28,733 INFO [train.py:873] (3/4) Epoch 6, batch 3500, loss[loss=0.1484, simple_loss=0.1777, pruned_loss=0.05951, over 14276.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.1843, pruned_loss=0.07624, over 1965678.33 frames. ], batch size: 80, lr: 1.31e-02, grad_scale: 8.0 2022-12-07 14:38:43,211 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.111e+02 2.629e+02 3.345e+02 4.291e+02 7.270e+02, threshold=6.690e+02, percent-clipped=2.0 2022-12-07 14:39:06,496 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41353.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:39:06,569 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.7435, 0.6216, 0.7021, 0.7224, 0.6106, 0.4796, 0.4989, 0.5151], device='cuda:3'), covar=tensor([0.0106, 0.0125, 0.0105, 0.0109, 0.0261, 0.0310, 0.0153, 0.0449], device='cuda:3'), in_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0010, 0.0011, 0.0015, 0.0012, 0.0016], device='cuda:3'), out_proj_covar=tensor([6.4848e-05, 6.7877e-05, 6.2327e-05, 6.1416e-05, 6.6563e-05, 8.8950e-05, 7.4748e-05, 8.7289e-05], device='cuda:3') 2022-12-07 14:39:08,296 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41355.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:39:48,938 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41401.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:39:57,988 INFO [train.py:873] (3/4) Epoch 6, batch 3600, loss[loss=0.1786, simple_loss=0.1878, pruned_loss=0.08466, over 14217.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.1842, pruned_loss=0.07628, over 1932790.57 frames. ], batch size: 94, lr: 1.31e-02, grad_scale: 8.0 2022-12-07 14:40:11,736 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.067e+02 2.550e+02 3.176e+02 4.214e+02 9.343e+02, threshold=6.353e+02, percent-clipped=2.0 2022-12-07 14:40:35,327 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.16 vs. limit=5.0 2022-12-07 14:40:36,828 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2022-12-07 14:40:39,028 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41458.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:41:05,673 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3346, 2.5467, 2.4678, 2.4844, 2.0646, 2.6987, 2.3007, 1.0791], device='cuda:3'), covar=tensor([0.3073, 0.1029, 0.1185, 0.0757, 0.0939, 0.0378, 0.1198, 0.3971], device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0063, 0.0050, 0.0054, 0.0076, 0.0057, 0.0082, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0005], device='cuda:3') 2022-12-07 14:41:06,113 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.38 vs. limit=5.0 2022-12-07 14:41:25,995 INFO [train.py:873] (3/4) Epoch 6, batch 3700, loss[loss=0.1814, simple_loss=0.1914, pruned_loss=0.08572, over 11184.00 frames. ], tot_loss[loss=0.169, simple_loss=0.1846, pruned_loss=0.07671, over 1931055.31 frames. ], batch size: 100, lr: 1.31e-02, grad_scale: 8.0 2022-12-07 14:41:34,656 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41521.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:41:39,913 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.379e+02 2.533e+02 3.248e+02 4.162e+02 7.233e+02, threshold=6.497e+02, percent-clipped=2.0 2022-12-07 14:41:43,065 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5248, 4.3516, 4.2040, 4.5458, 4.1370, 3.7318, 4.6075, 4.5037], device='cuda:3'), covar=tensor([0.0638, 0.0534, 0.0619, 0.0587, 0.0605, 0.0568, 0.0548, 0.0624], device='cuda:3'), in_proj_covar=tensor([0.0114, 0.0103, 0.0119, 0.0120, 0.0119, 0.0094, 0.0133, 0.0114], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 14:42:06,461 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41556.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 14:42:18,103 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41569.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:42:40,636 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4982, 2.3126, 2.5538, 2.4454, 2.4332, 2.1209, 1.4651, 2.1869], device='cuda:3'), covar=tensor([0.0354, 0.0397, 0.0377, 0.0340, 0.0284, 0.0858, 0.1953, 0.0283], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0140, 0.0124, 0.0117, 0.0173, 0.0118, 0.0150, 0.0159], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 14:42:49,105 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41604.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:42:55,064 INFO [train.py:873] (3/4) Epoch 6, batch 3800, loss[loss=0.2021, simple_loss=0.1813, pruned_loss=0.1114, over 1225.00 frames. ], tot_loss[loss=0.169, simple_loss=0.1849, pruned_loss=0.07658, over 1917928.37 frames. ], batch size: 100, lr: 1.31e-02, grad_scale: 16.0 2022-12-07 14:43:09,831 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.285e+02 2.547e+02 3.241e+02 4.303e+02 1.041e+03, threshold=6.482e+02, percent-clipped=5.0 2022-12-07 14:43:33,376 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41653.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:43:34,384 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41654.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:43:36,767 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6676, 1.5258, 1.7651, 1.9659, 1.4247, 1.6854, 1.9427, 1.8137], device='cuda:3'), covar=tensor([0.0077, 0.0108, 0.0046, 0.0035, 0.0101, 0.0192, 0.0049, 0.0051], device='cuda:3'), in_proj_covar=tensor([0.0209, 0.0206, 0.0307, 0.0242, 0.0194, 0.0253, 0.0193, 0.0243], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:3') 2022-12-07 14:43:43,919 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.09 vs. limit=2.0 2022-12-07 14:44:16,305 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41701.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:44:16,417 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41701.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:44:24,886 INFO [train.py:873] (3/4) Epoch 6, batch 3900, loss[loss=0.1557, simple_loss=0.1759, pruned_loss=0.06779, over 14581.00 frames. ], tot_loss[loss=0.167, simple_loss=0.1835, pruned_loss=0.07522, over 1891230.49 frames. ], batch size: 43, lr: 1.31e-02, grad_scale: 16.0 2022-12-07 14:44:28,615 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41715.0, num_to_drop=1, layers_to_drop={3} 2022-12-07 14:44:38,881 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.336e+02 2.642e+02 3.297e+02 4.041e+02 8.886e+02, threshold=6.594e+02, percent-clipped=2.0 2022-12-07 14:44:58,828 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41749.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:45:06,654 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41758.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:45:38,154 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2279, 1.7915, 3.5030, 2.4657, 3.4200, 1.7386, 2.5807, 3.2015], device='cuda:3'), covar=tensor([0.0762, 0.5109, 0.0378, 0.7928, 0.0409, 0.4078, 0.1403, 0.0518], device='cuda:3'), in_proj_covar=tensor([0.0226, 0.0252, 0.0172, 0.0341, 0.0182, 0.0254, 0.0240, 0.0179], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 14:45:49,541 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41806.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:45:54,444 INFO [train.py:873] (3/4) Epoch 6, batch 4000, loss[loss=0.1675, simple_loss=0.1755, pruned_loss=0.07977, over 6006.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.1837, pruned_loss=0.075, over 1941379.02 frames. ], batch size: 100, lr: 1.30e-02, grad_scale: 16.0 2022-12-07 14:46:08,692 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.360e+02 2.207e+02 2.921e+02 3.758e+02 8.130e+02, threshold=5.842e+02, percent-clipped=2.0 2022-12-07 14:46:33,732 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.92 vs. limit=2.0 2022-12-07 14:47:18,511 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41906.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:47:22,763 INFO [train.py:873] (3/4) Epoch 6, batch 4100, loss[loss=0.2083, simple_loss=0.2105, pruned_loss=0.1031, over 10317.00 frames. ], tot_loss[loss=0.1671, simple_loss=0.1841, pruned_loss=0.07503, over 1986675.60 frames. ], batch size: 100, lr: 1.30e-02, grad_scale: 8.0 2022-12-07 14:47:37,463 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.378e+02 2.509e+02 3.055e+02 4.404e+02 7.346e+02, threshold=6.111e+02, percent-clipped=4.0 2022-12-07 14:47:43,888 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0011, 2.0684, 2.1463, 2.2001, 1.7342, 2.1596, 1.8845, 0.9937], device='cuda:3'), covar=tensor([0.1807, 0.0843, 0.0510, 0.0420, 0.1054, 0.0705, 0.1800, 0.3408], device='cuda:3'), in_proj_covar=tensor([0.0158, 0.0063, 0.0051, 0.0054, 0.0078, 0.0058, 0.0085, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0005], device='cuda:3') 2022-12-07 14:47:55,756 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41948.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 14:48:11,888 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41967.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:48:49,710 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42009.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 14:48:50,449 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42010.0, num_to_drop=1, layers_to_drop={3} 2022-12-07 14:48:51,184 INFO [train.py:873] (3/4) Epoch 6, batch 4200, loss[loss=0.1526, simple_loss=0.1775, pruned_loss=0.06385, over 14266.00 frames. ], tot_loss[loss=0.1673, simple_loss=0.1843, pruned_loss=0.07519, over 2026782.50 frames. ], batch size: 60, lr: 1.30e-02, grad_scale: 8.0 2022-12-07 14:49:04,266 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.6355, 2.5152, 2.7037, 2.6118, 2.5477, 2.4302, 1.3835, 2.2949], device='cuda:3'), covar=tensor([0.0310, 0.0399, 0.0444, 0.0347, 0.0370, 0.0807, 0.2708, 0.0376], device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0144, 0.0128, 0.0120, 0.0176, 0.0122, 0.0155, 0.0163], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 14:49:07,155 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.456e+02 2.446e+02 2.869e+02 3.541e+02 9.255e+02, threshold=5.737e+02, percent-clipped=3.0 2022-12-07 14:49:21,635 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.29 vs. limit=5.0 2022-12-07 14:49:25,531 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5994, 4.5460, 4.8428, 3.9346, 4.6158, 4.8807, 1.9572, 4.3856], device='cuda:3'), covar=tensor([0.0175, 0.0225, 0.0351, 0.0477, 0.0237, 0.0136, 0.2714, 0.0233], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0143, 0.0127, 0.0119, 0.0174, 0.0121, 0.0154, 0.0163], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 14:49:44,566 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4290, 1.0012, 1.3624, 0.7941, 1.1244, 1.2967, 1.0551, 1.1694], device='cuda:3'), covar=tensor([0.0412, 0.0962, 0.0576, 0.0595, 0.1128, 0.0934, 0.0430, 0.1347], device='cuda:3'), in_proj_covar=tensor([0.0114, 0.0202, 0.0120, 0.0127, 0.0109, 0.0112, 0.0092, 0.0130], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0006, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0005], device='cuda:3') 2022-12-07 14:49:59,360 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.0649, 4.8599, 4.5641, 4.6234, 4.7589, 4.9878, 5.1288, 5.1226], device='cuda:3'), covar=tensor([0.0994, 0.0500, 0.2131, 0.3325, 0.0725, 0.0628, 0.1032, 0.0804], device='cuda:3'), in_proj_covar=tensor([0.0307, 0.0221, 0.0370, 0.0462, 0.0266, 0.0337, 0.0337, 0.0282], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 14:50:21,092 INFO [train.py:873] (3/4) Epoch 6, batch 4300, loss[loss=0.1784, simple_loss=0.1962, pruned_loss=0.08025, over 13991.00 frames. ], tot_loss[loss=0.1673, simple_loss=0.1844, pruned_loss=0.07507, over 2038444.98 frames. ], batch size: 22, lr: 1.30e-02, grad_scale: 8.0 2022-12-07 14:50:35,779 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.072e+01 2.450e+02 2.996e+02 3.613e+02 8.545e+02, threshold=5.992e+02, percent-clipped=0.0 2022-12-07 14:51:51,141 INFO [train.py:873] (3/4) Epoch 6, batch 4400, loss[loss=0.1578, simple_loss=0.177, pruned_loss=0.06933, over 14245.00 frames. ], tot_loss[loss=0.1672, simple_loss=0.184, pruned_loss=0.07522, over 1989289.76 frames. ], batch size: 69, lr: 1.30e-02, grad_scale: 8.0 2022-12-07 14:52:06,375 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.325e+02 2.637e+02 3.076e+02 3.933e+02 7.884e+02, threshold=6.152e+02, percent-clipped=2.0 2022-12-07 14:52:18,833 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8846, 2.5432, 4.8493, 3.1650, 4.7064, 2.1658, 3.5959, 4.4964], device='cuda:3'), covar=tensor([0.0478, 0.4277, 0.0677, 0.9771, 0.0322, 0.3657, 0.1133, 0.0259], device='cuda:3'), in_proj_covar=tensor([0.0220, 0.0243, 0.0172, 0.0338, 0.0182, 0.0251, 0.0235, 0.0178], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 14:52:36,697 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42262.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:53:14,097 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42304.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 14:53:19,686 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42310.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 14:53:20,408 INFO [train.py:873] (3/4) Epoch 6, batch 4500, loss[loss=0.1773, simple_loss=0.1817, pruned_loss=0.08647, over 5970.00 frames. ], tot_loss[loss=0.1662, simple_loss=0.1834, pruned_loss=0.07445, over 2024423.59 frames. ], batch size: 100, lr: 1.30e-02, grad_scale: 8.0 2022-12-07 14:53:35,322 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.162e+02 2.450e+02 2.856e+02 3.817e+02 6.882e+02, threshold=5.713e+02, percent-clipped=1.0 2022-12-07 14:54:02,632 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=42358.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:54:22,177 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42379.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:54:47,137 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2022-12-07 14:54:50,835 INFO [train.py:873] (3/4) Epoch 6, batch 4600, loss[loss=0.1685, simple_loss=0.1778, pruned_loss=0.07956, over 5953.00 frames. ], tot_loss[loss=0.1672, simple_loss=0.1844, pruned_loss=0.07497, over 2021276.18 frames. ], batch size: 100, lr: 1.30e-02, grad_scale: 8.0 2022-12-07 14:55:06,686 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.318e+02 2.342e+02 3.312e+02 4.722e+02 1.011e+03, threshold=6.623e+02, percent-clipped=9.0 2022-12-07 14:55:17,208 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42440.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:55:54,465 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0436, 2.6106, 4.2596, 4.3711, 4.5623, 2.7793, 4.3966, 3.5879], device='cuda:3'), covar=tensor([0.0150, 0.0359, 0.0387, 0.0158, 0.0099, 0.0559, 0.0109, 0.0357], device='cuda:3'), in_proj_covar=tensor([0.0210, 0.0206, 0.0311, 0.0244, 0.0196, 0.0252, 0.0200, 0.0240], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:3') 2022-12-07 14:56:00,940 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.05 vs. limit=2.0 2022-12-07 14:56:15,874 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42506.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:56:20,485 INFO [train.py:873] (3/4) Epoch 6, batch 4700, loss[loss=0.195, simple_loss=0.196, pruned_loss=0.09698, over 5025.00 frames. ], tot_loss[loss=0.1672, simple_loss=0.1842, pruned_loss=0.07512, over 1978403.77 frames. ], batch size: 100, lr: 1.29e-02, grad_scale: 8.0 2022-12-07 14:56:35,465 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.003e+02 2.212e+02 2.979e+02 3.531e+02 6.343e+02, threshold=5.959e+02, percent-clipped=0.0 2022-12-07 14:56:39,137 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4768, 1.8097, 1.8847, 1.8244, 1.6403, 1.9286, 1.5014, 0.9786], device='cuda:3'), covar=tensor([0.2342, 0.1152, 0.0457, 0.0337, 0.1036, 0.0539, 0.1607, 0.2736], device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0062, 0.0050, 0.0054, 0.0079, 0.0057, 0.0083, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0005], device='cuda:3') 2022-12-07 14:57:06,018 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42562.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:57:10,574 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42567.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:57:25,701 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.00 vs. limit=2.0 2022-12-07 14:57:25,724 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2022-12-07 14:57:26,164 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3522, 4.2569, 4.6158, 3.7577, 4.3387, 4.5025, 1.6824, 4.0469], device='cuda:3'), covar=tensor([0.0193, 0.0260, 0.0297, 0.0367, 0.0278, 0.0173, 0.2940, 0.0262], device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0143, 0.0124, 0.0118, 0.0173, 0.0120, 0.0151, 0.0163], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 14:57:26,257 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1939, 1.2655, 0.9677, 1.4247, 1.2968, 1.2730, 1.0194, 1.0357], device='cuda:3'), covar=tensor([0.1106, 0.2557, 0.1609, 0.1492, 0.1300, 0.0648, 0.0623, 0.1236], device='cuda:3'), in_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0010, 0.0012, 0.0015, 0.0012, 0.0016], device='cuda:3'), out_proj_covar=tensor([6.6271e-05, 6.9338e-05, 6.3971e-05, 6.2675e-05, 6.8944e-05, 9.1546e-05, 7.6875e-05, 8.7246e-05], device='cuda:3') 2022-12-07 14:57:43,822 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42604.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 14:57:48,833 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=42610.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:57:49,692 INFO [train.py:873] (3/4) Epoch 6, batch 4800, loss[loss=0.1316, simple_loss=0.1664, pruned_loss=0.04843, over 13926.00 frames. ], tot_loss[loss=0.1655, simple_loss=0.1827, pruned_loss=0.07415, over 1907427.77 frames. ], batch size: 23, lr: 1.29e-02, grad_scale: 8.0 2022-12-07 14:58:05,120 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.276e+02 2.352e+02 2.966e+02 3.514e+02 5.979e+02, threshold=5.933e+02, percent-clipped=1.0 2022-12-07 14:58:23,568 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42649.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:58:25,952 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=42652.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 14:58:47,011 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8799, 2.0090, 3.9340, 2.6071, 3.7701, 1.8683, 2.9657, 3.6633], device='cuda:3'), covar=tensor([0.0552, 0.5026, 0.0323, 0.8012, 0.0474, 0.4473, 0.1296, 0.0277], device='cuda:3'), in_proj_covar=tensor([0.0223, 0.0246, 0.0171, 0.0340, 0.0183, 0.0254, 0.0238, 0.0177], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 14:58:53,839 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.23 vs. limit=2.0 2022-12-07 14:59:04,477 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.5667, 1.2918, 1.2256, 1.1712, 1.2644, 0.9738, 1.0665, 0.9096], device='cuda:3'), covar=tensor([0.0453, 0.0857, 0.0846, 0.0807, 0.1078, 0.0305, 0.0315, 0.0773], device='cuda:3'), in_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0010, 0.0012, 0.0015, 0.0012, 0.0016], device='cuda:3'), out_proj_covar=tensor([6.7367e-05, 6.9896e-05, 6.5058e-05, 6.3509e-05, 7.0154e-05, 9.0338e-05, 7.7705e-05, 8.8699e-05], device='cuda:3') 2022-12-07 14:59:14,940 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.78 vs. limit=2.0 2022-12-07 14:59:18,009 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42710.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 14:59:18,817 INFO [train.py:873] (3/4) Epoch 6, batch 4900, loss[loss=0.1948, simple_loss=0.1901, pruned_loss=0.09969, over 7761.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.1843, pruned_loss=0.07546, over 1961163.87 frames. ], batch size: 100, lr: 1.29e-02, grad_scale: 8.0 2022-12-07 14:59:33,470 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.457e+02 2.456e+02 3.213e+02 4.119e+02 7.897e+02, threshold=6.426e+02, percent-clipped=4.0 2022-12-07 14:59:40,190 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42735.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:00:43,650 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0254, 1.9434, 2.0610, 2.0520, 1.9385, 1.8405, 1.2343, 1.7706], device='cuda:3'), covar=tensor([0.0450, 0.0438, 0.0442, 0.0298, 0.0356, 0.0807, 0.1921, 0.0384], device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0145, 0.0125, 0.0120, 0.0175, 0.0119, 0.0154, 0.0165], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 15:00:47,919 INFO [train.py:873] (3/4) Epoch 6, batch 5000, loss[loss=0.1534, simple_loss=0.1833, pruned_loss=0.06171, over 14452.00 frames. ], tot_loss[loss=0.1667, simple_loss=0.1838, pruned_loss=0.07481, over 1920449.67 frames. ], batch size: 49, lr: 1.29e-02, grad_scale: 8.0 2022-12-07 15:01:03,356 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.594e+01 2.343e+02 3.154e+02 4.045e+02 6.991e+02, threshold=6.308e+02, percent-clipped=1.0 2022-12-07 15:01:03,646 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0017, 2.0555, 3.9932, 2.8556, 3.9084, 2.0697, 2.9097, 3.7293], device='cuda:3'), covar=tensor([0.0406, 0.4776, 0.0375, 0.7042, 0.0378, 0.3930, 0.1350, 0.0225], device='cuda:3'), in_proj_covar=tensor([0.0220, 0.0246, 0.0172, 0.0341, 0.0187, 0.0251, 0.0242, 0.0178], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 15:01:26,187 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 2022-12-07 15:01:27,290 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 2022-12-07 15:01:30,750 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42859.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:01:33,653 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42862.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:01:48,848 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42879.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:02:16,481 INFO [train.py:873] (3/4) Epoch 6, batch 5100, loss[loss=0.1358, simple_loss=0.1672, pruned_loss=0.05216, over 14371.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.184, pruned_loss=0.07497, over 1963649.36 frames. ], batch size: 18, lr: 1.29e-02, grad_scale: 8.0 2022-12-07 15:02:24,091 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42920.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:02:30,810 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.301e+02 2.401e+02 2.989e+02 4.009e+02 7.834e+02, threshold=5.978e+02, percent-clipped=3.0 2022-12-07 15:02:42,205 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42940.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:02:58,686 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2022-12-07 15:03:14,029 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2022-12-07 15:03:39,117 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.83 vs. limit=2.0 2022-12-07 15:03:40,339 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43005.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:03:45,519 INFO [train.py:873] (3/4) Epoch 6, batch 5200, loss[loss=0.1747, simple_loss=0.1873, pruned_loss=0.08106, over 14220.00 frames. ], tot_loss[loss=0.1663, simple_loss=0.1834, pruned_loss=0.07454, over 1936180.89 frames. ], batch size: 89, lr: 1.29e-02, grad_scale: 8.0 2022-12-07 15:04:01,022 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.446e+02 2.408e+02 3.132e+02 3.708e+02 6.690e+02, threshold=6.264e+02, percent-clipped=2.0 2022-12-07 15:04:07,533 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43035.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:04:50,238 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43083.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:04:51,708 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0 2022-12-07 15:05:03,736 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7699, 1.4858, 1.8299, 2.0771, 1.3776, 1.7298, 1.9133, 1.9305], device='cuda:3'), covar=tensor([0.0055, 0.0105, 0.0045, 0.0030, 0.0094, 0.0126, 0.0054, 0.0038], device='cuda:3'), in_proj_covar=tensor([0.0211, 0.0209, 0.0311, 0.0249, 0.0198, 0.0251, 0.0200, 0.0241], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:3') 2022-12-07 15:05:15,649 INFO [train.py:873] (3/4) Epoch 6, batch 5300, loss[loss=0.1807, simple_loss=0.1772, pruned_loss=0.09213, over 4957.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.1845, pruned_loss=0.07563, over 1986425.22 frames. ], batch size: 100, lr: 1.29e-02, grad_scale: 8.0 2022-12-07 15:05:21,030 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.72 vs. limit=5.0 2022-12-07 15:05:29,869 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.587e+02 2.504e+02 3.046e+02 3.756e+02 7.408e+02, threshold=6.093e+02, percent-clipped=3.0 2022-12-07 15:06:00,718 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43162.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:06:43,862 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43210.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:06:44,601 INFO [train.py:873] (3/4) Epoch 6, batch 5400, loss[loss=0.1708, simple_loss=0.1682, pruned_loss=0.0867, over 3847.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.184, pruned_loss=0.07488, over 1995449.85 frames. ], batch size: 100, lr: 1.28e-02, grad_scale: 8.0 2022-12-07 15:06:47,960 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43215.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:06:59,606 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.808e+01 2.467e+02 2.944e+02 3.777e+02 9.316e+02, threshold=5.888e+02, percent-clipped=3.0 2022-12-07 15:07:06,183 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43235.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:08:08,981 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43305.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:08:14,526 INFO [train.py:873] (3/4) Epoch 6, batch 5500, loss[loss=0.1595, simple_loss=0.1802, pruned_loss=0.06945, over 13938.00 frames. ], tot_loss[loss=0.1653, simple_loss=0.1828, pruned_loss=0.07387, over 1963717.72 frames. ], batch size: 20, lr: 1.28e-02, grad_scale: 8.0 2022-12-07 15:08:28,979 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.312e+02 2.280e+02 2.901e+02 3.837e+02 7.523e+02, threshold=5.802e+02, percent-clipped=2.0 2022-12-07 15:08:45,851 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4547, 1.0140, 1.3392, 0.8628, 1.2904, 1.3876, 1.1402, 1.1871], device='cuda:3'), covar=tensor([0.0285, 0.0653, 0.0420, 0.0427, 0.0591, 0.0422, 0.0310, 0.0957], device='cuda:3'), in_proj_covar=tensor([0.0114, 0.0198, 0.0121, 0.0129, 0.0111, 0.0113, 0.0093, 0.0132], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0006, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004, 0.0005], device='cuda:3') 2022-12-07 15:08:52,290 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43353.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:08:57,590 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1280, 1.3001, 2.5171, 1.3515, 2.4196, 2.2946, 1.8480, 2.3699], device='cuda:3'), covar=tensor([0.0371, 0.2643, 0.0309, 0.2050, 0.0428, 0.0524, 0.0875, 0.0418], device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0156, 0.0138, 0.0168, 0.0152, 0.0152, 0.0122, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:3') 2022-12-07 15:09:14,519 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0405, 1.3850, 3.9057, 1.6934, 3.9178, 4.0261, 3.0430, 4.4057], device='cuda:3'), covar=tensor([0.0193, 0.2863, 0.0326, 0.2241, 0.0334, 0.0332, 0.0699, 0.0125], device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0157, 0.0139, 0.0169, 0.0153, 0.0153, 0.0123, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:3') 2022-12-07 15:09:22,787 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.41 vs. limit=5.0 2022-12-07 15:09:43,894 INFO [train.py:873] (3/4) Epoch 6, batch 5600, loss[loss=0.2133, simple_loss=0.2065, pruned_loss=0.11, over 8614.00 frames. ], tot_loss[loss=0.165, simple_loss=0.1824, pruned_loss=0.07376, over 1962299.34 frames. ], batch size: 100, lr: 1.28e-02, grad_scale: 8.0 2022-12-07 15:09:59,514 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.410e+02 2.262e+02 2.794e+02 3.488e+02 5.213e+02, threshold=5.588e+02, percent-clipped=0.0 2022-12-07 15:10:49,527 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.6472, 3.9439, 3.2550, 5.0415, 4.4670, 4.7333, 3.9317, 3.3421], device='cuda:3'), covar=tensor([0.0582, 0.1265, 0.4245, 0.0311, 0.0622, 0.1294, 0.1134, 0.3569], device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0300, 0.0290, 0.0201, 0.0262, 0.0266, 0.0252, 0.0275], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 15:11:13,947 INFO [train.py:873] (3/4) Epoch 6, batch 5700, loss[loss=0.1857, simple_loss=0.1987, pruned_loss=0.08637, over 11977.00 frames. ], tot_loss[loss=0.1642, simple_loss=0.1823, pruned_loss=0.07301, over 2050584.87 frames. ], batch size: 100, lr: 1.28e-02, grad_scale: 8.0 2022-12-07 15:11:16,537 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43514.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:11:17,401 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43515.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:11:28,309 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.154e+02 2.452e+02 3.294e+02 4.312e+02 1.114e+03, threshold=6.587e+02, percent-clipped=10.0 2022-12-07 15:11:34,548 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43535.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:11:40,005 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0337, 3.3712, 2.7932, 4.2441, 4.0241, 4.0635, 3.5356, 2.9037], device='cuda:3'), covar=tensor([0.0561, 0.1257, 0.3636, 0.0310, 0.0852, 0.0904, 0.0934, 0.3823], device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0301, 0.0289, 0.0201, 0.0261, 0.0262, 0.0251, 0.0276], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 15:11:59,859 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43563.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:12:10,448 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43575.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:12:17,485 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43583.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:12:42,161 INFO [train.py:873] (3/4) Epoch 6, batch 5800, loss[loss=0.1471, simple_loss=0.1806, pruned_loss=0.05678, over 14312.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.183, pruned_loss=0.07363, over 2056784.48 frames. ], batch size: 28, lr: 1.28e-02, grad_scale: 8.0 2022-12-07 15:12:57,492 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.271e+02 2.484e+02 3.216e+02 3.939e+02 1.250e+03, threshold=6.432e+02, percent-clipped=4.0 2022-12-07 15:13:04,042 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.5291, 5.3062, 5.0005, 5.6297, 5.1395, 4.7712, 5.6467, 5.5123], device='cuda:3'), covar=tensor([0.0749, 0.0588, 0.0602, 0.0612, 0.0708, 0.0389, 0.0535, 0.0619], device='cuda:3'), in_proj_covar=tensor([0.0113, 0.0102, 0.0115, 0.0118, 0.0120, 0.0092, 0.0131, 0.0113], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 15:13:25,718 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0917, 2.0387, 1.8144, 1.7778, 2.0470, 2.0478, 2.0758, 2.0373], device='cuda:3'), covar=tensor([0.0956, 0.0783, 0.2243, 0.2849, 0.1023, 0.0833, 0.1262, 0.0915], device='cuda:3'), in_proj_covar=tensor([0.0306, 0.0224, 0.0375, 0.0472, 0.0270, 0.0346, 0.0340, 0.0287], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 15:14:12,024 INFO [train.py:873] (3/4) Epoch 6, batch 5900, loss[loss=0.1485, simple_loss=0.1771, pruned_loss=0.05994, over 14392.00 frames. ], tot_loss[loss=0.1641, simple_loss=0.1821, pruned_loss=0.07308, over 2034002.71 frames. ], batch size: 41, lr: 1.28e-02, grad_scale: 8.0 2022-12-07 15:14:27,080 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.291e+02 2.477e+02 2.996e+02 3.769e+02 7.369e+02, threshold=5.992e+02, percent-clipped=1.0 2022-12-07 15:14:49,539 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.02 vs. limit=5.0 2022-12-07 15:15:20,313 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43787.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:15:42,279 INFO [train.py:873] (3/4) Epoch 6, batch 6000, loss[loss=0.1729, simple_loss=0.1923, pruned_loss=0.07679, over 14585.00 frames. ], tot_loss[loss=0.1647, simple_loss=0.1823, pruned_loss=0.07359, over 1965245.34 frames. ], batch size: 22, lr: 1.28e-02, grad_scale: 8.0 2022-12-07 15:15:42,279 INFO [train.py:896] (3/4) Computing validation loss 2022-12-07 15:15:47,663 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4103, 1.3629, 1.0943, 1.0216, 1.2631, 0.9818, 1.2848, 0.9415], device='cuda:3'), covar=tensor([0.0423, 0.0753, 0.0683, 0.0920, 0.0422, 0.0380, 0.0343, 0.1502], device='cuda:3'), in_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0010, 0.0012, 0.0016, 0.0012, 0.0017], device='cuda:3'), out_proj_covar=tensor([6.9285e-05, 7.4683e-05, 6.6895e-05, 6.8591e-05, 7.2476e-05, 9.9023e-05, 8.1221e-05, 9.4419e-05], device='cuda:3') 2022-12-07 15:15:51,413 INFO [train.py:905] (3/4) Epoch 6, validation: loss=0.1218, simple_loss=0.1652, pruned_loss=0.03921, over 857387.00 frames. 2022-12-07 15:15:51,413 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-07 15:16:01,777 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.3628, 1.3064, 1.2527, 1.4380, 1.3258, 1.1378, 1.4064, 1.2338], device='cuda:3'), covar=tensor([0.0949, 0.1213, 0.1488, 0.0651, 0.2922, 0.0598, 0.0502, 0.0885], device='cuda:3'), in_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0010, 0.0012, 0.0016, 0.0012, 0.0017], device='cuda:3'), out_proj_covar=tensor([6.8952e-05, 7.4380e-05, 6.6544e-05, 6.8222e-05, 7.2225e-05, 9.8848e-05, 8.0983e-05, 9.3979e-05], device='cuda:3') 2022-12-07 15:16:06,715 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.214e+02 2.267e+02 3.074e+02 3.851e+02 8.509e+02, threshold=6.148e+02, percent-clipped=4.0 2022-12-07 15:16:09,477 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7935, 1.5764, 2.0616, 1.5988, 1.9454, 1.4596, 1.6004, 1.7879], device='cuda:3'), covar=tensor([0.1302, 0.1769, 0.0182, 0.0922, 0.0590, 0.1071, 0.0777, 0.0480], device='cuda:3'), in_proj_covar=tensor([0.0219, 0.0240, 0.0171, 0.0337, 0.0187, 0.0253, 0.0238, 0.0183], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 15:16:24,562 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43848.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:16:44,654 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43870.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:16:46,314 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9670, 3.8501, 3.7304, 4.1226, 3.6625, 3.2978, 4.0429, 3.9266], device='cuda:3'), covar=tensor([0.0923, 0.0826, 0.0721, 0.0673, 0.0872, 0.0761, 0.0752, 0.0883], device='cuda:3'), in_proj_covar=tensor([0.0117, 0.0105, 0.0119, 0.0122, 0.0124, 0.0095, 0.0135, 0.0117], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 15:17:19,421 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43909.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:17:21,001 INFO [train.py:873] (3/4) Epoch 6, batch 6100, loss[loss=0.1591, simple_loss=0.1851, pruned_loss=0.06658, over 14509.00 frames. ], tot_loss[loss=0.1654, simple_loss=0.1825, pruned_loss=0.07411, over 1927521.93 frames. ], batch size: 34, lr: 1.27e-02, grad_scale: 16.0 2022-12-07 15:17:35,710 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.075e+02 2.736e+02 3.463e+02 4.201e+02 1.121e+03, threshold=6.926e+02, percent-clipped=3.0 2022-12-07 15:17:37,063 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8619, 2.2350, 4.3818, 4.0745, 4.0870, 4.4184, 3.8054, 4.4979], device='cuda:3'), covar=tensor([0.1265, 0.0986, 0.0068, 0.0138, 0.0128, 0.0076, 0.0154, 0.0085], device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0156, 0.0106, 0.0148, 0.0122, 0.0123, 0.0097, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-07 15:18:03,901 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43959.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:18:14,079 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43970.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:18:50,022 INFO [train.py:873] (3/4) Epoch 6, batch 6200, loss[loss=0.1685, simple_loss=0.1886, pruned_loss=0.07424, over 14413.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.1818, pruned_loss=0.07386, over 1887669.55 frames. ], batch size: 41, lr: 1.27e-02, grad_scale: 16.0 2022-12-07 15:18:58,952 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44020.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:19:06,617 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.319e+02 2.517e+02 3.185e+02 3.672e+02 8.437e+02, threshold=6.370e+02, percent-clipped=2.0 2022-12-07 15:20:19,413 INFO [train.py:873] (3/4) Epoch 6, batch 6300, loss[loss=0.1634, simple_loss=0.1864, pruned_loss=0.0702, over 13524.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.182, pruned_loss=0.0738, over 1895743.46 frames. ], batch size: 100, lr: 1.27e-02, grad_scale: 8.0 2022-12-07 15:20:30,997 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.84 vs. limit=2.0 2022-12-07 15:20:35,105 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.219e+02 2.319e+02 2.963e+02 3.987e+02 7.339e+02, threshold=5.926e+02, percent-clipped=1.0 2022-12-07 15:20:48,217 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44143.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:21:11,742 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44170.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:21:18,281 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.94 vs. limit=5.0 2022-12-07 15:21:48,489 INFO [train.py:873] (3/4) Epoch 6, batch 6400, loss[loss=0.1515, simple_loss=0.1854, pruned_loss=0.05879, over 13923.00 frames. ], tot_loss[loss=0.1639, simple_loss=0.1816, pruned_loss=0.07308, over 1869261.78 frames. ], batch size: 26, lr: 1.27e-02, grad_scale: 8.0 2022-12-07 15:21:55,395 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=44218.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:22:04,827 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.321e+02 2.640e+02 3.353e+02 4.203e+02 1.187e+03, threshold=6.705e+02, percent-clipped=10.0 2022-12-07 15:22:15,197 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.5643, 0.6338, 0.5818, 0.5807, 0.7495, 0.2898, 0.5366, 0.8016], device='cuda:3'), covar=tensor([0.0565, 0.0400, 0.0216, 0.0322, 0.0248, 0.0174, 0.1011, 0.0262], device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0021, 0.0020, 0.0020, 0.0020, 0.0028, 0.0020, 0.0020], device='cuda:3'), out_proj_covar=tensor([8.9207e-05, 9.2460e-05, 8.7163e-05, 8.7691e-05, 9.0947e-05, 1.1598e-04, 9.4176e-05, 8.8316e-05], device='cuda:3') 2022-12-07 15:22:26,562 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.85 vs. limit=5.0 2022-12-07 15:22:37,035 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44265.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:22:59,711 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.05 vs. limit=2.0 2022-12-07 15:23:11,631 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2022-12-07 15:23:17,596 INFO [train.py:873] (3/4) Epoch 6, batch 6500, loss[loss=0.1736, simple_loss=0.1933, pruned_loss=0.07694, over 14237.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.1829, pruned_loss=0.07489, over 1878501.51 frames. ], batch size: 39, lr: 1.27e-02, grad_scale: 8.0 2022-12-07 15:23:21,314 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44315.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:23:33,243 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.134e+02 2.290e+02 3.052e+02 3.998e+02 7.186e+02, threshold=6.105e+02, percent-clipped=1.0 2022-12-07 15:23:37,059 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.5396, 5.0146, 4.9824, 5.4739, 5.1157, 4.5535, 5.4611, 4.4012], device='cuda:3'), covar=tensor([0.0309, 0.0994, 0.0281, 0.0393, 0.0800, 0.0351, 0.0418, 0.0510], device='cuda:3'), in_proj_covar=tensor([0.0137, 0.0227, 0.0148, 0.0144, 0.0151, 0.0118, 0.0224, 0.0143], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 15:24:46,954 INFO [train.py:873] (3/4) Epoch 6, batch 6600, loss[loss=0.1588, simple_loss=0.1479, pruned_loss=0.08482, over 1231.00 frames. ], tot_loss[loss=0.165, simple_loss=0.1822, pruned_loss=0.07393, over 1947498.09 frames. ], batch size: 100, lr: 1.27e-02, grad_scale: 8.0 2022-12-07 15:25:03,289 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.314e+02 2.475e+02 3.026e+02 3.737e+02 6.916e+02, threshold=6.052e+02, percent-clipped=2.0 2022-12-07 15:25:15,305 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44443.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:25:33,609 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.3268, 1.3896, 1.5143, 1.1276, 0.9515, 0.8693, 0.6028, 0.8378], device='cuda:3'), covar=tensor([0.0291, 0.0434, 0.0356, 0.0359, 0.0402, 0.0253, 0.0295, 0.0679], device='cuda:3'), in_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0010, 0.0011, 0.0016, 0.0013, 0.0017], device='cuda:3'), out_proj_covar=tensor([7.0056e-05, 7.4959e-05, 6.6465e-05, 6.7760e-05, 7.0435e-05, 9.8643e-05, 8.5762e-05, 9.5962e-05], device='cuda:3') 2022-12-07 15:25:58,428 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=44491.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:26:16,139 INFO [train.py:873] (3/4) Epoch 6, batch 6700, loss[loss=0.1973, simple_loss=0.1954, pruned_loss=0.09961, over 3882.00 frames. ], tot_loss[loss=0.1652, simple_loss=0.1825, pruned_loss=0.074, over 1959019.96 frames. ], batch size: 100, lr: 1.27e-02, grad_scale: 8.0 2022-12-07 15:26:30,006 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=44527.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:26:31,879 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.570e+02 2.579e+02 3.217e+02 4.461e+02 7.506e+02, threshold=6.434e+02, percent-clipped=5.0 2022-12-07 15:26:34,725 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1496, 2.2573, 2.8602, 2.4260, 2.9435, 2.8501, 2.8756, 2.5233], device='cuda:3'), covar=tensor([0.0565, 0.2359, 0.0727, 0.1809, 0.0518, 0.0833, 0.0897, 0.1815], device='cuda:3'), in_proj_covar=tensor([0.0295, 0.0332, 0.0376, 0.0313, 0.0360, 0.0306, 0.0347, 0.0346], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 15:26:51,039 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=44550.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:27:02,588 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.69 vs. limit=2.0 2022-12-07 15:27:04,050 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44565.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:27:25,078 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44588.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:27:45,456 INFO [train.py:873] (3/4) Epoch 6, batch 6800, loss[loss=0.1814, simple_loss=0.1943, pruned_loss=0.08423, over 14593.00 frames. ], tot_loss[loss=0.1662, simple_loss=0.1833, pruned_loss=0.07459, over 1941128.22 frames. ], batch size: 23, lr: 1.26e-02, grad_scale: 8.0 2022-12-07 15:27:45,619 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44611.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:27:47,299 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=44613.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:27:49,124 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44615.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:28:01,814 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.238e+02 2.228e+02 2.960e+02 3.530e+02 6.520e+02, threshold=5.920e+02, percent-clipped=1.0 2022-12-07 15:28:24,065 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.7401, 0.6129, 0.5402, 0.7492, 0.7726, 0.2659, 0.6793, 0.8598], device='cuda:3'), covar=tensor([0.0129, 0.0345, 0.0181, 0.0196, 0.0163, 0.0132, 0.0326, 0.0200], device='cuda:3'), in_proj_covar=tensor([0.0019, 0.0020, 0.0019, 0.0019, 0.0020, 0.0027, 0.0019, 0.0019], device='cuda:3'), out_proj_covar=tensor([8.5385e-05, 8.8636e-05, 8.3919e-05, 8.4823e-05, 8.7869e-05, 1.1230e-04, 9.0138e-05, 8.5097e-05], device='cuda:3') 2022-12-07 15:28:32,306 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=44663.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:28:33,205 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1452, 1.4036, 1.8040, 1.6389, 1.2968, 0.8661, 0.9692, 0.8162], device='cuda:3'), covar=tensor([0.0790, 0.1185, 0.0520, 0.0219, 0.0884, 0.0223, 0.0381, 0.0984], device='cuda:3'), in_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0010, 0.0011, 0.0016, 0.0013, 0.0016], device='cuda:3'), out_proj_covar=tensor([6.8641e-05, 7.3135e-05, 6.5364e-05, 6.6029e-05, 6.8110e-05, 9.7012e-05, 8.3328e-05, 9.3471e-05], device='cuda:3') 2022-12-07 15:28:51,221 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.78 vs. limit=2.0 2022-12-07 15:29:15,081 INFO [train.py:873] (3/4) Epoch 6, batch 6900, loss[loss=0.1703, simple_loss=0.1854, pruned_loss=0.07756, over 14403.00 frames. ], tot_loss[loss=0.164, simple_loss=0.1817, pruned_loss=0.07314, over 1967414.65 frames. ], batch size: 53, lr: 1.26e-02, grad_scale: 8.0 2022-12-07 15:29:31,171 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.297e+02 2.434e+02 2.991e+02 3.865e+02 6.867e+02, threshold=5.982e+02, percent-clipped=5.0 2022-12-07 15:30:43,890 INFO [train.py:873] (3/4) Epoch 6, batch 7000, loss[loss=0.1771, simple_loss=0.1886, pruned_loss=0.08282, over 8569.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.1815, pruned_loss=0.07305, over 2002823.73 frames. ], batch size: 100, lr: 1.26e-02, grad_scale: 8.0 2022-12-07 15:31:00,498 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.508e+02 2.539e+02 3.130e+02 3.727e+02 8.173e+02, threshold=6.260e+02, percent-clipped=3.0 2022-12-07 15:31:47,420 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44883.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:31:54,937 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8465, 1.4631, 1.8094, 1.2790, 1.5524, 1.8453, 1.6153, 1.6104], device='cuda:3'), covar=tensor([0.0384, 0.0597, 0.0478, 0.0766, 0.0921, 0.0565, 0.0422, 0.1403], device='cuda:3'), in_proj_covar=tensor([0.0114, 0.0197, 0.0125, 0.0128, 0.0113, 0.0112, 0.0094, 0.0131], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0006, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004, 0.0005], device='cuda:3') 2022-12-07 15:31:57,390 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.6389, 2.1812, 2.4065, 1.3107, 2.1891, 2.3826, 2.7412, 2.0976], device='cuda:3'), covar=tensor([0.0692, 0.2014, 0.1294, 0.3390, 0.1338, 0.0690, 0.0685, 0.2128], device='cuda:3'), in_proj_covar=tensor([0.0114, 0.0197, 0.0125, 0.0128, 0.0114, 0.0112, 0.0094, 0.0131], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0006, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004, 0.0005], device='cuda:3') 2022-12-07 15:32:05,652 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2022-12-07 15:32:08,007 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44906.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:32:12,183 INFO [train.py:873] (3/4) Epoch 6, batch 7100, loss[loss=0.1956, simple_loss=0.1975, pruned_loss=0.09682, over 6946.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.1817, pruned_loss=0.07294, over 1970851.98 frames. ], batch size: 100, lr: 1.26e-02, grad_scale: 8.0 2022-12-07 15:32:13,176 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.6983, 1.3645, 2.8525, 1.4172, 3.0078, 2.8997, 2.1239, 2.9949], device='cuda:3'), covar=tensor([0.0231, 0.2329, 0.0317, 0.1889, 0.0238, 0.0303, 0.0768, 0.0188], device='cuda:3'), in_proj_covar=tensor([0.0159, 0.0156, 0.0144, 0.0168, 0.0155, 0.0156, 0.0127, 0.0129], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:3') 2022-12-07 15:32:27,515 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.394e+02 2.599e+02 3.099e+02 3.894e+02 6.870e+02, threshold=6.198e+02, percent-clipped=1.0 2022-12-07 15:33:15,336 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4027, 2.2134, 4.7249, 4.4680, 4.3194, 4.8541, 4.3414, 4.8706], device='cuda:3'), covar=tensor([0.0997, 0.1028, 0.0085, 0.0132, 0.0136, 0.0073, 0.0112, 0.0084], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0157, 0.0110, 0.0154, 0.0125, 0.0127, 0.0101, 0.0105], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-07 15:33:44,151 INFO [train.py:873] (3/4) Epoch 6, batch 7200, loss[loss=0.1721, simple_loss=0.1882, pruned_loss=0.07802, over 13871.00 frames. ], tot_loss[loss=0.164, simple_loss=0.1818, pruned_loss=0.07309, over 1985503.78 frames. ], batch size: 23, lr: 1.26e-02, grad_scale: 8.0 2022-12-07 15:34:00,689 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.267e+02 2.403e+02 2.886e+02 3.979e+02 7.470e+02, threshold=5.772e+02, percent-clipped=2.0 2022-12-07 15:34:24,815 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45056.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:34:46,883 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45081.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:34:59,578 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 2022-12-07 15:35:13,052 INFO [train.py:873] (3/4) Epoch 6, batch 7300, loss[loss=0.157, simple_loss=0.1811, pruned_loss=0.06644, over 14403.00 frames. ], tot_loss[loss=0.1627, simple_loss=0.1809, pruned_loss=0.07228, over 2034740.78 frames. ], batch size: 53, lr: 1.26e-02, grad_scale: 8.0 2022-12-07 15:35:18,594 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45117.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 15:35:29,218 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.594e+02 2.544e+02 3.092e+02 3.955e+02 6.821e+02, threshold=6.183e+02, percent-clipped=3.0 2022-12-07 15:35:41,383 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45142.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:35:59,655 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0585, 2.3537, 3.8686, 4.2180, 4.1126, 2.5878, 4.0342, 3.1266], device='cuda:3'), covar=tensor([0.0181, 0.0447, 0.0515, 0.0221, 0.0142, 0.0669, 0.0159, 0.0481], device='cuda:3'), in_proj_covar=tensor([0.0223, 0.0213, 0.0326, 0.0258, 0.0205, 0.0259, 0.0213, 0.0250], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:3') 2022-12-07 15:36:17,661 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45183.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:36:37,794 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45206.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:36:41,986 INFO [train.py:873] (3/4) Epoch 6, batch 7400, loss[loss=0.1603, simple_loss=0.1903, pruned_loss=0.06515, over 14439.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.1826, pruned_loss=0.07384, over 2026146.23 frames. ], batch size: 53, lr: 1.26e-02, grad_scale: 8.0 2022-12-07 15:36:58,363 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.666e+02 2.594e+02 3.335e+02 4.109e+02 6.658e+02, threshold=6.669e+02, percent-clipped=6.0 2022-12-07 15:37:00,148 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=45231.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:37:07,632 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1480, 3.2502, 4.1384, 2.7991, 2.4696, 3.2806, 1.9610, 3.1334], device='cuda:3'), covar=tensor([0.1150, 0.1290, 0.0658, 0.2189, 0.2326, 0.1159, 0.4802, 0.2330], device='cuda:3'), in_proj_covar=tensor([0.0070, 0.0082, 0.0078, 0.0084, 0.0107, 0.0067, 0.0133, 0.0076], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2022-12-07 15:37:14,717 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.66 vs. limit=2.0 2022-12-07 15:37:21,139 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=45254.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:37:43,352 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5748, 1.4458, 2.7774, 1.4688, 2.7550, 2.7511, 2.0369, 2.8620], device='cuda:3'), covar=tensor([0.0219, 0.1944, 0.0269, 0.1698, 0.0278, 0.0307, 0.0872, 0.0190], device='cuda:3'), in_proj_covar=tensor([0.0161, 0.0156, 0.0146, 0.0169, 0.0157, 0.0157, 0.0129, 0.0129], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 15:37:54,976 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9842, 1.7118, 5.0200, 4.5845, 4.4760, 5.1560, 4.9371, 5.2618], device='cuda:3'), covar=tensor([0.1172, 0.1186, 0.0052, 0.0102, 0.0116, 0.0054, 0.0045, 0.0058], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0160, 0.0111, 0.0156, 0.0127, 0.0128, 0.0101, 0.0107], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-07 15:38:12,295 INFO [train.py:873] (3/4) Epoch 6, batch 7500, loss[loss=0.1824, simple_loss=0.1948, pruned_loss=0.08501, over 14192.00 frames. ], tot_loss[loss=0.164, simple_loss=0.1821, pruned_loss=0.07299, over 2040339.48 frames. ], batch size: 69, lr: 1.25e-02, grad_scale: 8.0 2022-12-07 15:38:27,612 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.117e+02 2.437e+02 3.096e+02 3.781e+02 7.631e+02, threshold=6.193e+02, percent-clipped=3.0 2022-12-07 15:38:36,267 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4669, 2.8339, 4.1641, 3.3244, 4.2080, 4.0681, 4.0326, 3.6095], device='cuda:3'), covar=tensor([0.0429, 0.2930, 0.0968, 0.1963, 0.0624, 0.0750, 0.1715, 0.2371], device='cuda:3'), in_proj_covar=tensor([0.0301, 0.0327, 0.0379, 0.0314, 0.0355, 0.0303, 0.0345, 0.0343], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 15:38:44,274 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45348.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 15:39:51,270 INFO [train.py:873] (3/4) Epoch 7, batch 0, loss[loss=0.2043, simple_loss=0.2141, pruned_loss=0.09722, over 14207.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.2141, pruned_loss=0.09722, over 14207.00 frames. ], batch size: 89, lr: 1.17e-02, grad_scale: 8.0 2022-12-07 15:39:51,270 INFO [train.py:896] (3/4) Computing validation loss 2022-12-07 15:39:59,859 INFO [train.py:905] (3/4) Epoch 7, validation: loss=0.1305, simple_loss=0.175, pruned_loss=0.04304, over 857387.00 frames. 2022-12-07 15:39:59,860 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-07 15:40:33,526 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2609, 3.0350, 2.7792, 2.9045, 3.1872, 3.1657, 3.2857, 3.2188], device='cuda:3'), covar=tensor([0.0874, 0.0690, 0.2059, 0.2670, 0.0701, 0.0763, 0.0913, 0.0855], device='cuda:3'), in_proj_covar=tensor([0.0319, 0.0240, 0.0388, 0.0494, 0.0279, 0.0365, 0.0361, 0.0305], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 15:40:33,617 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45409.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 15:40:36,170 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45412.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 15:40:51,472 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 5.719e+01 2.285e+02 3.118e+02 4.102e+02 1.207e+03, threshold=6.237e+02, percent-clipped=4.0 2022-12-07 15:40:59,155 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45437.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:41:31,135 INFO [train.py:873] (3/4) Epoch 7, batch 100, loss[loss=0.1483, simple_loss=0.1745, pruned_loss=0.06103, over 14474.00 frames. ], tot_loss[loss=0.1654, simple_loss=0.1831, pruned_loss=0.07382, over 853516.09 frames. ], batch size: 51, lr: 1.17e-02, grad_scale: 8.0 2022-12-07 15:42:07,234 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4651, 2.1076, 3.2282, 2.4697, 3.3002, 2.9813, 3.1501, 2.5882], device='cuda:3'), covar=tensor([0.0521, 0.3713, 0.0978, 0.2310, 0.0766, 0.1070, 0.1512, 0.2975], device='cuda:3'), in_proj_covar=tensor([0.0299, 0.0325, 0.0372, 0.0309, 0.0353, 0.0301, 0.0346, 0.0338], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 15:42:21,346 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.420e+02 2.542e+02 3.020e+02 3.979e+02 7.369e+02, threshold=6.041e+02, percent-clipped=5.0 2022-12-07 15:42:47,851 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2022-12-07 15:43:00,540 INFO [train.py:873] (3/4) Epoch 7, batch 200, loss[loss=0.1984, simple_loss=0.1752, pruned_loss=0.1108, over 1229.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.183, pruned_loss=0.07542, over 1241409.56 frames. ], batch size: 100, lr: 1.17e-02, grad_scale: 8.0 2022-12-07 15:43:50,944 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.430e+02 2.408e+02 2.969e+02 3.747e+02 6.560e+02, threshold=5.938e+02, percent-clipped=2.0 2022-12-07 15:43:55,559 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2022-12-07 15:44:30,682 INFO [train.py:873] (3/4) Epoch 7, batch 300, loss[loss=0.1527, simple_loss=0.1634, pruned_loss=0.07103, over 4951.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.1811, pruned_loss=0.07329, over 1513448.61 frames. ], batch size: 100, lr: 1.17e-02, grad_scale: 8.0 2022-12-07 15:44:31,105 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9544, 4.5237, 4.4718, 4.9448, 4.5790, 4.1839, 4.9012, 4.3646], device='cuda:3'), covar=tensor([0.0300, 0.0903, 0.0250, 0.0312, 0.0676, 0.0561, 0.0459, 0.0361], device='cuda:3'), in_proj_covar=tensor([0.0140, 0.0234, 0.0154, 0.0146, 0.0154, 0.0125, 0.0231, 0.0148], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 15:44:35,648 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.1572, 4.9915, 4.9990, 4.7214, 4.9545, 5.5585, 1.8838, 4.6509], device='cuda:3'), covar=tensor([0.0242, 0.0263, 0.0647, 0.0334, 0.0392, 0.0107, 0.3895, 0.0358], device='cuda:3'), in_proj_covar=tensor([0.0140, 0.0150, 0.0131, 0.0123, 0.0178, 0.0122, 0.0155, 0.0166], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 15:44:56,563 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45702.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:44:58,083 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45704.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 15:44:58,979 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.0090, 0.8512, 0.9649, 0.9956, 1.0465, 0.5873, 1.1254, 0.9238], device='cuda:3'), covar=tensor([0.0683, 0.0667, 0.0488, 0.0568, 0.0503, 0.0628, 0.0474, 0.0601], device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0020, 0.0019, 0.0020, 0.0020, 0.0028, 0.0020, 0.0021], device='cuda:3'), out_proj_covar=tensor([8.9402e-05, 8.9121e-05, 8.6411e-05, 8.9996e-05, 9.0228e-05, 1.1617e-04, 9.5131e-05, 9.2310e-05], device='cuda:3') 2022-12-07 15:45:05,210 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45712.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 15:45:20,483 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.117e+02 2.543e+02 3.312e+02 3.999e+02 6.909e+02, threshold=6.625e+02, percent-clipped=2.0 2022-12-07 15:45:27,790 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45737.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:45:48,364 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=45760.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:45:50,961 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45763.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:45:59,857 INFO [train.py:873] (3/4) Epoch 7, batch 400, loss[loss=0.1727, simple_loss=0.1688, pruned_loss=0.08831, over 2533.00 frames. ], tot_loss[loss=0.1624, simple_loss=0.1806, pruned_loss=0.07212, over 1715718.37 frames. ], batch size: 100, lr: 1.17e-02, grad_scale: 8.0 2022-12-07 15:46:01,067 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45774.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:46:10,766 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=45785.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:46:15,526 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.3983, 5.0826, 4.7860, 5.4771, 4.9087, 4.7206, 5.4194, 5.3274], device='cuda:3'), covar=tensor([0.0621, 0.0529, 0.0801, 0.0535, 0.0714, 0.0480, 0.0510, 0.0670], device='cuda:3'), in_proj_covar=tensor([0.0114, 0.0104, 0.0120, 0.0124, 0.0121, 0.0096, 0.0132, 0.0116], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 15:46:50,563 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.164e+02 2.431e+02 3.067e+02 3.597e+02 6.934e+02, threshold=6.134e+02, percent-clipped=1.0 2022-12-07 15:46:56,434 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45835.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:47:17,590 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45859.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:47:18,458 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1070, 1.3325, 1.3159, 1.0820, 1.2896, 0.9059, 1.2377, 0.8293], device='cuda:3'), covar=tensor([0.0368, 0.0544, 0.0477, 0.0338, 0.0457, 0.0310, 0.0321, 0.0878], device='cuda:3'), in_proj_covar=tensor([0.0011, 0.0013, 0.0010, 0.0011, 0.0011, 0.0017, 0.0013, 0.0017], device='cuda:3'), out_proj_covar=tensor([7.2191e-05, 7.8512e-05, 7.0448e-05, 7.1783e-05, 7.2207e-05, 1.0545e-04, 8.8334e-05, 1.0018e-04], device='cuda:3') 2022-12-07 15:47:30,184 INFO [train.py:873] (3/4) Epoch 7, batch 500, loss[loss=0.182, simple_loss=0.2047, pruned_loss=0.07963, over 14088.00 frames. ], tot_loss[loss=0.1642, simple_loss=0.1822, pruned_loss=0.07309, over 1840034.28 frames. ], batch size: 29, lr: 1.17e-02, grad_scale: 4.0 2022-12-07 15:47:33,678 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.25 vs. limit=5.0 2022-12-07 15:48:12,926 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45920.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:48:14,872 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.65 vs. limit=2.0 2022-12-07 15:48:21,753 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.265e+02 2.455e+02 3.322e+02 4.008e+02 8.135e+02, threshold=6.645e+02, percent-clipped=7.0 2022-12-07 15:48:24,841 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.81 vs. limit=5.0 2022-12-07 15:48:43,250 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9917, 3.1659, 2.9740, 3.0530, 2.2230, 3.0440, 2.9720, 1.3932], device='cuda:3'), covar=tensor([0.2756, 0.0616, 0.0840, 0.0916, 0.1102, 0.0693, 0.1037, 0.3059], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0062, 0.0048, 0.0054, 0.0077, 0.0056, 0.0082, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0004, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0005], device='cuda:3') 2022-12-07 15:48:59,908 INFO [train.py:873] (3/4) Epoch 7, batch 600, loss[loss=0.176, simple_loss=0.1849, pruned_loss=0.08359, over 11179.00 frames. ], tot_loss[loss=0.1639, simple_loss=0.1818, pruned_loss=0.07297, over 1866861.71 frames. ], batch size: 100, lr: 1.17e-02, grad_scale: 4.0 2022-12-07 15:49:07,763 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.13 vs. limit=5.0 2022-12-07 15:49:27,337 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46004.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 15:49:46,452 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3185, 1.3166, 3.3904, 1.4177, 3.1899, 3.3606, 2.2960, 3.6196], device='cuda:3'), covar=tensor([0.0233, 0.2790, 0.0338, 0.2312, 0.0776, 0.0362, 0.0888, 0.0162], device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0159, 0.0149, 0.0170, 0.0158, 0.0160, 0.0131, 0.0131], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 15:49:49,711 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.803e+01 2.347e+02 3.029e+02 4.189e+02 7.825e+02, threshold=6.059e+02, percent-clipped=1.0 2022-12-07 15:49:52,714 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.63 vs. limit=2.0 2022-12-07 15:49:58,322 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8441, 1.2123, 2.5477, 2.3234, 2.4988, 2.4805, 1.9999, 2.5722], device='cuda:3'), covar=tensor([0.0549, 0.0906, 0.0103, 0.0271, 0.0206, 0.0111, 0.0294, 0.0127], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0156, 0.0110, 0.0152, 0.0125, 0.0126, 0.0100, 0.0105], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-07 15:50:09,875 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46052.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 15:50:15,107 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46058.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:50:19,941 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.48 vs. limit=5.0 2022-12-07 15:50:28,497 INFO [train.py:873] (3/4) Epoch 7, batch 700, loss[loss=0.1882, simple_loss=0.1681, pruned_loss=0.1041, over 2565.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.1818, pruned_loss=0.0729, over 1938102.69 frames. ], batch size: 100, lr: 1.17e-02, grad_scale: 4.0 2022-12-07 15:50:31,514 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7218, 2.0704, 1.9161, 2.0802, 1.7672, 2.1302, 1.7648, 1.2050], device='cuda:3'), covar=tensor([0.1874, 0.0651, 0.0528, 0.0436, 0.1093, 0.0521, 0.1672, 0.2547], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0062, 0.0050, 0.0054, 0.0078, 0.0057, 0.0083, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0004, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0005], device='cuda:3') 2022-12-07 15:51:18,729 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.605e+01 2.609e+02 3.339e+02 4.151e+02 6.051e+02, threshold=6.678e+02, percent-clipped=0.0 2022-12-07 15:51:18,857 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46130.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:51:26,001 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1377, 1.7622, 3.2860, 2.1836, 3.2695, 1.6145, 2.4753, 3.1834], device='cuda:3'), covar=tensor([0.0698, 0.5219, 0.0402, 0.7773, 0.0415, 0.4171, 0.1549, 0.0448], device='cuda:3'), in_proj_covar=tensor([0.0217, 0.0235, 0.0165, 0.0321, 0.0183, 0.0241, 0.0230, 0.0179], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 15:51:46,098 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5852, 1.9137, 2.4622, 2.1672, 2.6218, 2.3163, 2.3101, 2.2437], device='cuda:3'), covar=tensor([0.0380, 0.2085, 0.0438, 0.1609, 0.0415, 0.0683, 0.0646, 0.1460], device='cuda:3'), in_proj_covar=tensor([0.0305, 0.0330, 0.0377, 0.0313, 0.0368, 0.0304, 0.0354, 0.0336], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 15:51:56,448 INFO [train.py:873] (3/4) Epoch 7, batch 800, loss[loss=0.1535, simple_loss=0.1566, pruned_loss=0.07517, over 3863.00 frames. ], tot_loss[loss=0.1639, simple_loss=0.1814, pruned_loss=0.07318, over 1905546.91 frames. ], batch size: 100, lr: 1.16e-02, grad_scale: 8.0 2022-12-07 15:52:34,731 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46215.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:52:47,697 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.967e+01 2.522e+02 3.122e+02 4.007e+02 6.928e+02, threshold=6.244e+02, percent-clipped=1.0 2022-12-07 15:53:26,504 INFO [train.py:873] (3/4) Epoch 7, batch 900, loss[loss=0.1728, simple_loss=0.1755, pruned_loss=0.08508, over 4919.00 frames. ], tot_loss[loss=0.1621, simple_loss=0.1801, pruned_loss=0.07206, over 1889489.96 frames. ], batch size: 100, lr: 1.16e-02, grad_scale: 8.0 2022-12-07 15:53:29,273 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46276.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:53:45,468 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.59 vs. limit=5.0 2022-12-07 15:53:50,834 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.23 vs. limit=5.0 2022-12-07 15:54:16,878 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.049e+02 2.538e+02 3.268e+02 4.038e+02 9.582e+02, threshold=6.536e+02, percent-clipped=3.0 2022-12-07 15:54:19,991 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8569, 1.5904, 1.8816, 1.7174, 2.0239, 1.7630, 1.6162, 1.8237], device='cuda:3'), covar=tensor([0.0373, 0.1338, 0.0162, 0.0348, 0.0180, 0.0621, 0.0159, 0.0280], device='cuda:3'), in_proj_covar=tensor([0.0306, 0.0335, 0.0381, 0.0314, 0.0368, 0.0306, 0.0357, 0.0339], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 15:54:23,267 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46337.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:54:42,248 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46358.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:54:49,415 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2022-12-07 15:54:55,203 INFO [train.py:873] (3/4) Epoch 7, batch 1000, loss[loss=0.1623, simple_loss=0.1891, pruned_loss=0.06775, over 14201.00 frames. ], tot_loss[loss=0.1618, simple_loss=0.1803, pruned_loss=0.07161, over 1992869.12 frames. ], batch size: 46, lr: 1.16e-02, grad_scale: 8.0 2022-12-07 15:55:09,830 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2022-12-07 15:55:24,791 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46406.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:55:46,198 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.306e+02 2.372e+02 2.921e+02 3.865e+02 1.419e+03, threshold=5.842e+02, percent-clipped=9.0 2022-12-07 15:55:46,383 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46430.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:56:24,647 INFO [train.py:873] (3/4) Epoch 7, batch 1100, loss[loss=0.1369, simple_loss=0.1711, pruned_loss=0.0514, over 14455.00 frames. ], tot_loss[loss=0.1624, simple_loss=0.1805, pruned_loss=0.07213, over 1963257.40 frames. ], batch size: 41, lr: 1.16e-02, grad_scale: 8.0 2022-12-07 15:56:29,019 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46478.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:56:41,824 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2139, 2.0768, 3.2628, 3.3171, 3.2875, 2.1642, 3.3029, 2.4049], device='cuda:3'), covar=tensor([0.0186, 0.0419, 0.0388, 0.0220, 0.0164, 0.0690, 0.0125, 0.0492], device='cuda:3'), in_proj_covar=tensor([0.0222, 0.0211, 0.0321, 0.0257, 0.0205, 0.0260, 0.0216, 0.0249], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:3') 2022-12-07 15:56:49,650 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2022-12-07 15:56:50,211 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9921, 4.9056, 4.3034, 4.4892, 4.7510, 5.0583, 5.2295, 4.9971], device='cuda:3'), covar=tensor([0.1192, 0.0601, 0.2901, 0.3605, 0.0866, 0.0941, 0.0902, 0.1530], device='cuda:3'), in_proj_covar=tensor([0.0319, 0.0233, 0.0391, 0.0493, 0.0283, 0.0359, 0.0344, 0.0298], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 15:57:01,900 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46515.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:57:15,198 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.295e+02 2.418e+02 3.079e+02 3.635e+02 7.997e+02, threshold=6.158e+02, percent-clipped=1.0 2022-12-07 15:57:28,021 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8807, 1.3987, 3.1166, 1.4000, 3.1943, 3.0210, 2.1951, 3.2063], device='cuda:3'), covar=tensor([0.0284, 0.2833, 0.0351, 0.2387, 0.0332, 0.0474, 0.0798, 0.0242], device='cuda:3'), in_proj_covar=tensor([0.0162, 0.0160, 0.0148, 0.0170, 0.0161, 0.0160, 0.0129, 0.0131], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 15:57:45,028 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46563.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:57:53,828 INFO [train.py:873] (3/4) Epoch 7, batch 1200, loss[loss=0.1408, simple_loss=0.1714, pruned_loss=0.05507, over 14294.00 frames. ], tot_loss[loss=0.1613, simple_loss=0.1801, pruned_loss=0.07128, over 1957946.91 frames. ], batch size: 63, lr: 1.16e-02, grad_scale: 8.0 2022-12-07 15:58:11,722 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4784, 2.3880, 3.2293, 2.6230, 3.3427, 3.1856, 3.1932, 2.7683], device='cuda:3'), covar=tensor([0.0562, 0.2778, 0.0946, 0.2053, 0.0767, 0.0714, 0.1479, 0.1782], device='cuda:3'), in_proj_covar=tensor([0.0302, 0.0328, 0.0375, 0.0310, 0.0358, 0.0298, 0.0351, 0.0334], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 15:58:13,359 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46595.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:58:20,098 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46602.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:58:40,386 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.78 vs. limit=2.0 2022-12-07 15:58:44,673 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.081e+01 2.469e+02 2.984e+02 3.753e+02 8.629e+02, threshold=5.968e+02, percent-clipped=3.0 2022-12-07 15:58:46,474 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46632.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:59:08,458 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46656.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 15:59:14,214 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46663.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 15:59:23,323 INFO [train.py:873] (3/4) Epoch 7, batch 1300, loss[loss=0.2178, simple_loss=0.1974, pruned_loss=0.1191, over 3918.00 frames. ], tot_loss[loss=0.1626, simple_loss=0.1807, pruned_loss=0.07222, over 1897248.61 frames. ], batch size: 100, lr: 1.16e-02, grad_scale: 8.0 2022-12-07 15:59:36,830 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8572, 3.8426, 4.0319, 3.7218, 3.8739, 3.9894, 1.5329, 3.6847], device='cuda:3'), covar=tensor([0.0219, 0.0291, 0.0415, 0.0425, 0.0287, 0.0323, 0.3178, 0.0256], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0145, 0.0126, 0.0118, 0.0173, 0.0120, 0.0148, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 16:00:04,762 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=7.90 vs. limit=5.0 2022-12-07 16:00:14,512 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.436e+02 2.325e+02 3.068e+02 3.769e+02 6.043e+02, threshold=6.136e+02, percent-clipped=2.0 2022-12-07 16:00:16,670 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.12 vs. limit=2.0 2022-12-07 16:00:40,525 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.64 vs. limit=2.0 2022-12-07 16:00:53,006 INFO [train.py:873] (3/4) Epoch 7, batch 1400, loss[loss=0.1508, simple_loss=0.1747, pruned_loss=0.06344, over 14192.00 frames. ], tot_loss[loss=0.1619, simple_loss=0.1802, pruned_loss=0.07176, over 1932582.29 frames. ], batch size: 57, lr: 1.16e-02, grad_scale: 8.0 2022-12-07 16:01:27,155 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46810.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:01:45,246 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.279e+02 2.517e+02 3.054e+02 3.788e+02 7.041e+02, threshold=6.108e+02, percent-clipped=3.0 2022-12-07 16:02:21,948 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46871.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:02:23,386 INFO [train.py:873] (3/4) Epoch 7, batch 1500, loss[loss=0.1822, simple_loss=0.1658, pruned_loss=0.09927, over 1208.00 frames. ], tot_loss[loss=0.1619, simple_loss=0.1801, pruned_loss=0.07188, over 1913681.36 frames. ], batch size: 100, lr: 1.16e-02, grad_scale: 8.0 2022-12-07 16:03:13,535 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.064e+01 2.462e+02 3.053e+02 4.274e+02 1.237e+03, threshold=6.105e+02, percent-clipped=6.0 2022-12-07 16:03:15,391 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46932.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:03:32,146 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46951.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 16:03:38,751 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46958.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:03:43,836 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8343, 0.8907, 0.9377, 0.9496, 0.8570, 0.7689, 0.5274, 0.6497], device='cuda:3'), covar=tensor([0.0178, 0.0207, 0.0130, 0.0148, 0.0328, 0.0402, 0.0283, 0.0480], device='cuda:3'), in_proj_covar=tensor([0.0012, 0.0013, 0.0011, 0.0011, 0.0012, 0.0017, 0.0014, 0.0018], device='cuda:3'), out_proj_covar=tensor([7.4006e-05, 8.0499e-05, 7.2449e-05, 7.2942e-05, 7.4452e-05, 1.0802e-04, 9.2588e-05, 1.0401e-04], device='cuda:3') 2022-12-07 16:03:45,161 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.72 vs. limit=2.0 2022-12-07 16:03:49,280 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9576, 3.2205, 4.6087, 3.4803, 4.5959, 4.4370, 4.2612, 4.0386], device='cuda:3'), covar=tensor([0.0378, 0.2749, 0.0646, 0.1777, 0.0704, 0.0602, 0.2209, 0.1975], device='cuda:3'), in_proj_covar=tensor([0.0305, 0.0333, 0.0381, 0.0312, 0.0364, 0.0305, 0.0357, 0.0336], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 16:03:51,574 INFO [train.py:873] (3/4) Epoch 7, batch 1600, loss[loss=0.1631, simple_loss=0.1453, pruned_loss=0.09045, over 1228.00 frames. ], tot_loss[loss=0.1622, simple_loss=0.1801, pruned_loss=0.07219, over 1912431.23 frames. ], batch size: 100, lr: 1.15e-02, grad_scale: 8.0 2022-12-07 16:03:57,845 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46980.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:04:01,733 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.44 vs. limit=2.0 2022-12-07 16:04:21,278 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.0075, 1.8773, 4.6716, 2.4645, 4.3735, 4.7547, 4.6657, 5.4046], device='cuda:3'), covar=tensor([0.0145, 0.2652, 0.0282, 0.1810, 0.0228, 0.0253, 0.0189, 0.0102], device='cuda:3'), in_proj_covar=tensor([0.0162, 0.0162, 0.0146, 0.0167, 0.0159, 0.0158, 0.0129, 0.0131], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 16:04:32,131 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4822, 4.2143, 3.9970, 4.1088, 4.2591, 4.3756, 4.4874, 4.4554], device='cuda:3'), covar=tensor([0.0809, 0.0593, 0.1937, 0.2750, 0.0745, 0.0712, 0.0987, 0.0873], device='cuda:3'), in_proj_covar=tensor([0.0319, 0.0238, 0.0395, 0.0491, 0.0283, 0.0361, 0.0342, 0.0305], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 16:04:42,987 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.262e+02 2.534e+02 3.085e+02 3.879e+02 6.356e+02, threshold=6.171e+02, percent-clipped=1.0 2022-12-07 16:05:05,425 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.76 vs. limit=2.0 2022-12-07 16:05:21,095 INFO [train.py:873] (3/4) Epoch 7, batch 1700, loss[loss=0.1404, simple_loss=0.1739, pruned_loss=0.05345, over 14191.00 frames. ], tot_loss[loss=0.1629, simple_loss=0.1802, pruned_loss=0.07284, over 1869451.78 frames. ], batch size: 37, lr: 1.15e-02, grad_scale: 8.0 2022-12-07 16:05:28,698 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4951, 1.4966, 4.2363, 2.1049, 4.1798, 4.2863, 4.0364, 4.9906], device='cuda:3'), covar=tensor([0.0192, 0.3245, 0.0394, 0.2191, 0.0314, 0.0396, 0.0396, 0.0126], device='cuda:3'), in_proj_covar=tensor([0.0161, 0.0161, 0.0146, 0.0168, 0.0159, 0.0158, 0.0129, 0.0130], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 16:05:37,309 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1296, 3.1964, 3.3406, 3.1206, 3.2037, 2.6572, 1.3465, 2.9797], device='cuda:3'), covar=tensor([0.0320, 0.0346, 0.0442, 0.0350, 0.0314, 0.1014, 0.3180, 0.0294], device='cuda:3'), in_proj_covar=tensor([0.0138, 0.0150, 0.0130, 0.0120, 0.0175, 0.0124, 0.0151, 0.0166], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 16:05:58,897 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3427, 3.1486, 2.8933, 3.0308, 3.2945, 3.2954, 3.3824, 3.3052], device='cuda:3'), covar=tensor([0.0998, 0.0820, 0.2123, 0.2851, 0.0837, 0.0848, 0.1033, 0.1003], device='cuda:3'), in_proj_covar=tensor([0.0322, 0.0236, 0.0395, 0.0491, 0.0283, 0.0360, 0.0340, 0.0303], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 16:06:11,977 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.285e+02 2.426e+02 3.035e+02 3.683e+02 7.018e+02, threshold=6.070e+02, percent-clipped=2.0 2022-12-07 16:06:22,046 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 2022-12-07 16:06:44,248 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47166.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:06:50,173 INFO [train.py:873] (3/4) Epoch 7, batch 1800, loss[loss=0.1865, simple_loss=0.1623, pruned_loss=0.1054, over 1332.00 frames. ], tot_loss[loss=0.1609, simple_loss=0.1788, pruned_loss=0.07153, over 1863798.56 frames. ], batch size: 100, lr: 1.15e-02, grad_scale: 8.0 2022-12-07 16:07:17,151 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.70 vs. limit=2.0 2022-12-07 16:07:40,891 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.187e+02 2.439e+02 3.276e+02 4.235e+02 9.028e+02, threshold=6.551e+02, percent-clipped=6.0 2022-12-07 16:07:50,937 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47241.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:07:59,637 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47251.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:08:05,776 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47258.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:08:19,485 INFO [train.py:873] (3/4) Epoch 7, batch 1900, loss[loss=0.177, simple_loss=0.1796, pruned_loss=0.08725, over 8608.00 frames. ], tot_loss[loss=0.1618, simple_loss=0.1794, pruned_loss=0.07215, over 1864729.89 frames. ], batch size: 100, lr: 1.15e-02, grad_scale: 8.0 2022-12-07 16:08:40,087 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.00 vs. limit=2.0 2022-12-07 16:08:42,236 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=47299.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:08:44,908 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47302.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:08:48,304 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=47306.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:09:10,046 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.096e+02 2.352e+02 2.909e+02 3.754e+02 8.188e+02, threshold=5.818e+02, percent-clipped=1.0 2022-12-07 16:09:47,745 INFO [train.py:873] (3/4) Epoch 7, batch 2000, loss[loss=0.1648, simple_loss=0.1532, pruned_loss=0.08817, over 1291.00 frames. ], tot_loss[loss=0.1615, simple_loss=0.1795, pruned_loss=0.07174, over 1894078.61 frames. ], batch size: 100, lr: 1.15e-02, grad_scale: 8.0 2022-12-07 16:09:53,100 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47379.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:10:35,434 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.7053, 0.6842, 0.6383, 0.7556, 0.7450, 0.1984, 0.6220, 0.7580], device='cuda:3'), covar=tensor([0.0216, 0.0344, 0.0121, 0.0187, 0.0146, 0.0088, 0.0766, 0.0310], device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0020, 0.0020, 0.0021, 0.0020, 0.0028, 0.0021, 0.0021], device='cuda:3'), out_proj_covar=tensor([9.1037e-05, 9.2699e-05, 8.8956e-05, 9.4515e-05, 9.1976e-05, 1.1868e-04, 9.7618e-05, 9.4359e-05], device='cuda:3') 2022-12-07 16:10:39,231 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([6.0742, 5.5092, 5.4170, 5.9542, 5.4953, 4.6143, 5.9401, 4.8142], device='cuda:3'), covar=tensor([0.0271, 0.0773, 0.0276, 0.0396, 0.0704, 0.0319, 0.0420, 0.0425], device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0234, 0.0153, 0.0148, 0.0155, 0.0124, 0.0229, 0.0146], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 16:10:40,007 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.317e+02 2.312e+02 2.798e+02 3.545e+02 9.094e+02, threshold=5.596e+02, percent-clipped=2.0 2022-12-07 16:10:48,773 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47440.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:11:11,241 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47466.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:11:18,003 INFO [train.py:873] (3/4) Epoch 7, batch 2100, loss[loss=0.2013, simple_loss=0.1996, pruned_loss=0.1015, over 8622.00 frames. ], tot_loss[loss=0.1589, simple_loss=0.1781, pruned_loss=0.06982, over 2004029.55 frames. ], batch size: 100, lr: 1.15e-02, grad_scale: 8.0 2022-12-07 16:11:22,037 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.27 vs. limit=5.0 2022-12-07 16:11:53,886 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=47514.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:12:09,093 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.294e+02 2.308e+02 2.873e+02 3.601e+02 7.245e+02, threshold=5.745e+02, percent-clipped=7.0 2022-12-07 16:12:29,831 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5464, 2.3397, 4.6604, 3.1669, 4.4247, 2.2080, 3.4041, 4.1397], device='cuda:3'), covar=tensor([0.0612, 0.5130, 0.0273, 0.8798, 0.0319, 0.3990, 0.1425, 0.0381], device='cuda:3'), in_proj_covar=tensor([0.0225, 0.0241, 0.0167, 0.0325, 0.0190, 0.0249, 0.0230, 0.0187], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0005, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 16:12:46,255 INFO [train.py:873] (3/4) Epoch 7, batch 2200, loss[loss=0.141, simple_loss=0.1402, pruned_loss=0.07089, over 1284.00 frames. ], tot_loss[loss=0.1608, simple_loss=0.1793, pruned_loss=0.07112, over 1964765.22 frames. ], batch size: 100, lr: 1.15e-02, grad_scale: 8.0 2022-12-07 16:12:53,934 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2022-12-07 16:13:08,035 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47597.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:13:29,505 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2022-12-07 16:13:37,950 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.355e+02 2.347e+02 2.892e+02 3.477e+02 7.491e+02, threshold=5.783e+02, percent-clipped=4.0 2022-12-07 16:14:15,731 INFO [train.py:873] (3/4) Epoch 7, batch 2300, loss[loss=0.1415, simple_loss=0.1748, pruned_loss=0.05412, over 14287.00 frames. ], tot_loss[loss=0.1591, simple_loss=0.178, pruned_loss=0.07011, over 1970830.73 frames. ], batch size: 44, lr: 1.15e-02, grad_scale: 8.0 2022-12-07 16:15:07,553 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.453e+02 2.381e+02 2.948e+02 3.680e+02 8.732e+02, threshold=5.896e+02, percent-clipped=3.0 2022-12-07 16:15:10,926 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47735.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:15:13,637 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47738.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:15:45,315 INFO [train.py:873] (3/4) Epoch 7, batch 2400, loss[loss=0.1413, simple_loss=0.168, pruned_loss=0.05734, over 13971.00 frames. ], tot_loss[loss=0.1603, simple_loss=0.1789, pruned_loss=0.07085, over 1978434.81 frames. ], batch size: 19, lr: 1.15e-02, grad_scale: 8.0 2022-12-07 16:15:48,322 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.69 vs. limit=2.0 2022-12-07 16:15:49,642 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1331, 2.3252, 2.3144, 2.3457, 2.0654, 2.5220, 2.1643, 1.1903], device='cuda:3'), covar=tensor([0.2413, 0.0868, 0.0861, 0.0666, 0.0964, 0.0525, 0.1534, 0.2906], device='cuda:3'), in_proj_covar=tensor([0.0154, 0.0063, 0.0051, 0.0054, 0.0078, 0.0059, 0.0085, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0004, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0005], device='cuda:3') 2022-12-07 16:15:58,203 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47788.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:16:08,465 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47799.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:16:17,307 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47809.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 16:16:36,797 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.330e+02 2.462e+02 3.088e+02 3.941e+02 7.678e+02, threshold=6.176e+02, percent-clipped=7.0 2022-12-07 16:16:53,127 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47849.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:17:07,825 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9091, 1.3230, 3.7322, 1.5112, 3.7958, 3.8905, 2.7337, 4.1631], device='cuda:3'), covar=tensor([0.0176, 0.3099, 0.0412, 0.2399, 0.0384, 0.0340, 0.0622, 0.0137], device='cuda:3'), in_proj_covar=tensor([0.0160, 0.0159, 0.0148, 0.0165, 0.0159, 0.0156, 0.0129, 0.0130], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 16:17:11,985 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47870.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 16:17:14,362 INFO [train.py:873] (3/4) Epoch 7, batch 2500, loss[loss=0.1745, simple_loss=0.1899, pruned_loss=0.07952, over 14499.00 frames. ], tot_loss[loss=0.1599, simple_loss=0.1787, pruned_loss=0.07058, over 1969942.90 frames. ], batch size: 49, lr: 1.14e-02, grad_scale: 8.0 2022-12-07 16:17:19,454 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4488, 3.2373, 3.1774, 3.5905, 3.1194, 2.8896, 3.4549, 3.4888], device='cuda:3'), covar=tensor([0.0736, 0.0885, 0.0876, 0.0593, 0.0990, 0.0784, 0.0894, 0.0873], device='cuda:3'), in_proj_covar=tensor([0.0112, 0.0100, 0.0115, 0.0119, 0.0118, 0.0092, 0.0131, 0.0113], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 16:17:35,623 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47897.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:17:51,381 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.86 vs. limit=2.0 2022-12-07 16:18:06,392 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.397e+02 2.559e+02 3.205e+02 4.003e+02 7.455e+02, threshold=6.409e+02, percent-clipped=4.0 2022-12-07 16:18:18,487 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=47945.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:18:43,889 INFO [train.py:873] (3/4) Epoch 7, batch 2600, loss[loss=0.1928, simple_loss=0.2018, pruned_loss=0.09188, over 6936.00 frames. ], tot_loss[loss=0.159, simple_loss=0.1785, pruned_loss=0.06974, over 2032803.85 frames. ], batch size: 100, lr: 1.14e-02, grad_scale: 8.0 2022-12-07 16:18:52,636 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 2022-12-07 16:19:27,596 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2022-12-07 16:19:36,188 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.203e+02 2.385e+02 3.008e+02 3.884e+02 8.359e+02, threshold=6.016e+02, percent-clipped=2.0 2022-12-07 16:19:40,489 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48035.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:20:13,250 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48072.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:20:13,913 INFO [train.py:873] (3/4) Epoch 7, batch 2700, loss[loss=0.1339, simple_loss=0.1643, pruned_loss=0.05175, over 14653.00 frames. ], tot_loss[loss=0.1581, simple_loss=0.1779, pruned_loss=0.06908, over 2083905.77 frames. ], batch size: 33, lr: 1.14e-02, grad_scale: 8.0 2022-12-07 16:20:23,361 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48083.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:20:33,265 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48094.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:21:03,562 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6983, 3.4061, 3.0804, 2.4762, 3.0245, 3.2639, 3.7295, 2.8945], device='cuda:3'), covar=tensor([0.0503, 0.2011, 0.1214, 0.1800, 0.0985, 0.0970, 0.0600, 0.1297], device='cuda:3'), in_proj_covar=tensor([0.0114, 0.0195, 0.0126, 0.0130, 0.0113, 0.0117, 0.0096, 0.0129], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0006, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004, 0.0005], device='cuda:3') 2022-12-07 16:21:06,887 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.439e+01 2.366e+02 3.097e+02 3.751e+02 6.035e+02, threshold=6.194e+02, percent-clipped=1.0 2022-12-07 16:21:08,809 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48133.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:21:17,950 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48144.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:21:37,445 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48165.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 16:21:44,250 INFO [train.py:873] (3/4) Epoch 7, batch 2800, loss[loss=0.1625, simple_loss=0.1812, pruned_loss=0.07184, over 14172.00 frames. ], tot_loss[loss=0.1579, simple_loss=0.1777, pruned_loss=0.06906, over 1985592.24 frames. ], batch size: 84, lr: 1.14e-02, grad_scale: 8.0 2022-12-07 16:22:20,017 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48212.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 16:22:36,452 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.192e+02 2.244e+02 3.117e+02 3.930e+02 6.165e+02, threshold=6.234e+02, percent-clipped=0.0 2022-12-07 16:23:12,316 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4137, 1.5580, 4.1326, 1.8304, 4.0545, 4.3657, 3.8416, 4.7922], device='cuda:3'), covar=tensor([0.0179, 0.2881, 0.0335, 0.2166, 0.0329, 0.0315, 0.0352, 0.0113], device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0161, 0.0150, 0.0168, 0.0162, 0.0161, 0.0132, 0.0133], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 16:23:14,253 INFO [train.py:873] (3/4) Epoch 7, batch 2900, loss[loss=0.1292, simple_loss=0.1658, pruned_loss=0.04626, over 14486.00 frames. ], tot_loss[loss=0.1572, simple_loss=0.1776, pruned_loss=0.06838, over 2050464.64 frames. ], batch size: 34, lr: 1.14e-02, grad_scale: 8.0 2022-12-07 16:23:14,725 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48273.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 16:23:19,249 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4844, 1.0939, 1.9846, 1.8099, 1.7694, 1.9714, 1.3704, 1.9792], device='cuda:3'), covar=tensor([0.0567, 0.1027, 0.0170, 0.0372, 0.0382, 0.0174, 0.0501, 0.0214], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0156, 0.0112, 0.0155, 0.0128, 0.0127, 0.0103, 0.0107], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-07 16:23:25,367 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.43 vs. limit=5.0 2022-12-07 16:23:45,316 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9254, 4.8078, 5.2433, 4.1943, 4.9567, 5.3685, 1.9786, 4.6685], device='cuda:3'), covar=tensor([0.0189, 0.0226, 0.0287, 0.0383, 0.0266, 0.0102, 0.2963, 0.0223], device='cuda:3'), in_proj_covar=tensor([0.0137, 0.0145, 0.0127, 0.0120, 0.0173, 0.0122, 0.0151, 0.0163], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 16:23:48,974 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8387, 2.3324, 4.7094, 3.0342, 4.3617, 2.1947, 3.4502, 4.3886], device='cuda:3'), covar=tensor([0.0465, 0.5387, 0.0510, 0.9453, 0.0581, 0.4350, 0.1317, 0.0378], device='cuda:3'), in_proj_covar=tensor([0.0223, 0.0239, 0.0173, 0.0324, 0.0190, 0.0243, 0.0223, 0.0185], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0005, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 16:24:01,606 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48324.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:24:07,736 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.587e+02 2.377e+02 3.206e+02 3.983e+02 6.235e+02, threshold=6.412e+02, percent-clipped=1.0 2022-12-07 16:24:46,197 INFO [train.py:873] (3/4) Epoch 7, batch 3000, loss[loss=0.1535, simple_loss=0.1691, pruned_loss=0.069, over 14128.00 frames. ], tot_loss[loss=0.1576, simple_loss=0.1779, pruned_loss=0.06864, over 2040523.17 frames. ], batch size: 19, lr: 1.14e-02, grad_scale: 8.0 2022-12-07 16:24:46,198 INFO [train.py:896] (3/4) Computing validation loss 2022-12-07 16:25:07,554 INFO [train.py:905] (3/4) Epoch 7, validation: loss=0.1228, simple_loss=0.1657, pruned_loss=0.03995, over 857387.00 frames. 2022-12-07 16:25:07,554 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-07 16:25:19,618 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48385.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:25:25,040 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3660, 3.0410, 2.2742, 3.4403, 3.1668, 3.2031, 2.7082, 2.3593], device='cuda:3'), covar=tensor([0.0552, 0.1260, 0.3408, 0.0374, 0.0889, 0.0813, 0.1291, 0.3598], device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0306, 0.0285, 0.0206, 0.0272, 0.0275, 0.0253, 0.0274], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 16:25:27,554 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48394.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:25:30,247 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48397.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:25:33,313 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3901, 1.8436, 3.4587, 2.4392, 3.3713, 1.7641, 2.6214, 3.1633], device='cuda:3'), covar=tensor([0.0578, 0.5212, 0.0395, 0.7067, 0.0569, 0.4210, 0.1427, 0.0511], device='cuda:3'), in_proj_covar=tensor([0.0223, 0.0240, 0.0173, 0.0324, 0.0191, 0.0244, 0.0223, 0.0185], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0005, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 16:25:48,063 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.73 vs. limit=5.0 2022-12-07 16:25:54,464 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48423.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:25:58,983 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48428.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:26:01,391 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.229e+02 2.374e+02 2.963e+02 3.800e+02 1.337e+03, threshold=5.925e+02, percent-clipped=7.0 2022-12-07 16:26:03,157 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9757, 4.4741, 4.4100, 4.8924, 4.5549, 4.2940, 4.8919, 4.1339], device='cuda:3'), covar=tensor([0.0308, 0.1039, 0.0337, 0.0398, 0.0808, 0.0517, 0.0520, 0.0524], device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0233, 0.0154, 0.0147, 0.0157, 0.0123, 0.0231, 0.0148], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 16:26:10,819 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48442.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:26:12,867 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48444.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:26:26,034 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48458.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:26:26,050 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.6982, 2.4253, 4.6204, 3.0383, 4.3716, 2.1568, 3.6201, 4.3492], device='cuda:3'), covar=tensor([0.0333, 0.4972, 0.0361, 0.8448, 0.0400, 0.4115, 0.1055, 0.0273], device='cuda:3'), in_proj_covar=tensor([0.0221, 0.0238, 0.0172, 0.0323, 0.0190, 0.0243, 0.0224, 0.0184], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0005, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 16:26:32,564 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48465.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 16:26:39,376 INFO [train.py:873] (3/4) Epoch 7, batch 3100, loss[loss=0.1507, simple_loss=0.1613, pruned_loss=0.07001, over 5988.00 frames. ], tot_loss[loss=0.1577, simple_loss=0.1777, pruned_loss=0.06891, over 2039782.16 frames. ], batch size: 100, lr: 1.14e-02, grad_scale: 8.0 2022-12-07 16:26:49,101 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48484.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:26:56,192 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48492.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:27:15,584 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48513.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 16:27:28,088 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.90 vs. limit=2.0 2022-12-07 16:27:31,121 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.348e+02 2.509e+02 3.187e+02 4.348e+02 7.597e+02, threshold=6.375e+02, percent-clipped=5.0 2022-12-07 16:28:04,890 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48568.0, num_to_drop=1, layers_to_drop={3} 2022-12-07 16:28:09,138 INFO [train.py:873] (3/4) Epoch 7, batch 3200, loss[loss=0.1475, simple_loss=0.1708, pruned_loss=0.06209, over 14428.00 frames. ], tot_loss[loss=0.1587, simple_loss=0.1783, pruned_loss=0.06954, over 1996248.70 frames. ], batch size: 73, lr: 1.14e-02, grad_scale: 8.0 2022-12-07 16:28:15,083 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.5474, 1.1541, 2.0783, 1.8976, 1.8945, 2.1036, 1.4601, 2.1050], device='cuda:3'), covar=tensor([0.0517, 0.0853, 0.0112, 0.0281, 0.0257, 0.0103, 0.0361, 0.0124], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0156, 0.0111, 0.0154, 0.0126, 0.0127, 0.0103, 0.0108], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-07 16:29:02,495 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.098e+02 2.245e+02 2.769e+02 3.901e+02 6.467e+02, threshold=5.537e+02, percent-clipped=1.0 2022-12-07 16:29:40,602 INFO [train.py:873] (3/4) Epoch 7, batch 3300, loss[loss=0.1558, simple_loss=0.1782, pruned_loss=0.06673, over 14321.00 frames. ], tot_loss[loss=0.1582, simple_loss=0.1776, pruned_loss=0.06937, over 1922825.45 frames. ], batch size: 25, lr: 1.13e-02, grad_scale: 8.0 2022-12-07 16:29:43,951 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.14 vs. limit=5.0 2022-12-07 16:29:46,864 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48680.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:29:51,543 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9069, 2.0095, 2.9221, 3.0057, 2.9320, 2.0749, 2.8529, 2.2302], device='cuda:3'), covar=tensor([0.0183, 0.0412, 0.0274, 0.0188, 0.0173, 0.0609, 0.0157, 0.0457], device='cuda:3'), in_proj_covar=tensor([0.0228, 0.0217, 0.0329, 0.0267, 0.0213, 0.0265, 0.0223, 0.0254], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:3') 2022-12-07 16:30:17,895 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2416, 1.4299, 2.5417, 1.4272, 2.5303, 2.4730, 1.8946, 2.5913], device='cuda:3'), covar=tensor([0.0228, 0.2011, 0.0265, 0.1569, 0.0321, 0.0408, 0.0973, 0.0241], device='cuda:3'), in_proj_covar=tensor([0.0162, 0.0161, 0.0148, 0.0167, 0.0159, 0.0160, 0.0131, 0.0131], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 16:30:30,586 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48728.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:30:32,953 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.098e+02 2.201e+02 2.834e+02 3.394e+02 6.106e+02, threshold=5.668e+02, percent-clipped=1.0 2022-12-07 16:30:53,231 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48753.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:31:10,410 INFO [train.py:873] (3/4) Epoch 7, batch 3400, loss[loss=0.1466, simple_loss=0.1699, pruned_loss=0.06166, over 14294.00 frames. ], tot_loss[loss=0.1591, simple_loss=0.1783, pruned_loss=0.06993, over 1935870.43 frames. ], batch size: 60, lr: 1.13e-02, grad_scale: 8.0 2022-12-07 16:31:13,824 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48776.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:31:16,501 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48779.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:31:50,867 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.76 vs. limit=2.0 2022-12-07 16:31:54,737 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48821.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:32:04,512 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.534e+01 2.454e+02 2.984e+02 3.725e+02 7.522e+02, threshold=5.968e+02, percent-clipped=2.0 2022-12-07 16:32:09,950 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48838.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:32:22,097 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.2639, 1.0472, 1.0549, 1.1943, 1.2130, 0.5636, 1.1874, 1.3115], device='cuda:3'), covar=tensor([0.0950, 0.1158, 0.0739, 0.0927, 0.0634, 0.0584, 0.0786, 0.0567], device='cuda:3'), in_proj_covar=tensor([0.0019, 0.0020, 0.0020, 0.0020, 0.0019, 0.0027, 0.0021, 0.0021], device='cuda:3'), out_proj_covar=tensor([9.1755e-05, 9.2146e-05, 9.2641e-05, 9.2581e-05, 9.0671e-05, 1.1743e-04, 9.8832e-05, 9.4963e-05], device='cuda:3') 2022-12-07 16:32:36,348 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48868.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 16:32:40,438 INFO [train.py:873] (3/4) Epoch 7, batch 3500, loss[loss=0.1887, simple_loss=0.1994, pruned_loss=0.08894, over 14211.00 frames. ], tot_loss[loss=0.159, simple_loss=0.1785, pruned_loss=0.06974, over 1973123.51 frames. ], batch size: 89, lr: 1.13e-02, grad_scale: 4.0 2022-12-07 16:32:48,458 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48882.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:33:03,286 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48899.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:33:18,189 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48916.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 16:33:31,831 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.472e+02 2.610e+02 3.212e+02 4.326e+02 7.674e+02, threshold=6.425e+02, percent-clipped=4.0 2022-12-07 16:33:48,769 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2022-12-07 16:34:08,388 INFO [train.py:873] (3/4) Epoch 7, batch 3600, loss[loss=0.2113, simple_loss=0.204, pruned_loss=0.1093, over 9518.00 frames. ], tot_loss[loss=0.1586, simple_loss=0.1784, pruned_loss=0.0694, over 1959124.33 frames. ], batch size: 100, lr: 1.13e-02, grad_scale: 8.0 2022-12-07 16:34:14,479 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48980.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:34:22,936 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9175, 3.7148, 3.4153, 3.4766, 3.7699, 3.7844, 3.9234, 3.9055], device='cuda:3'), covar=tensor([0.1040, 0.0639, 0.2248, 0.3483, 0.0901, 0.0868, 0.1002, 0.1106], device='cuda:3'), in_proj_covar=tensor([0.0325, 0.0233, 0.0389, 0.0490, 0.0283, 0.0356, 0.0343, 0.0311], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 16:34:58,409 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49028.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:35:01,629 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.848e+01 2.321e+02 2.995e+02 3.518e+02 8.417e+02, threshold=5.990e+02, percent-clipped=3.0 2022-12-07 16:35:05,332 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3193, 3.1164, 3.0940, 3.4385, 2.9716, 2.7246, 3.4363, 3.3730], device='cuda:3'), covar=tensor([0.0852, 0.0770, 0.0794, 0.0610, 0.1011, 0.0774, 0.0663, 0.0773], device='cuda:3'), in_proj_covar=tensor([0.0118, 0.0105, 0.0122, 0.0125, 0.0123, 0.0096, 0.0137, 0.0121], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 16:35:10,615 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7988, 3.4719, 3.4350, 3.7564, 3.5655, 3.7195, 3.8302, 3.1227], device='cuda:3'), covar=tensor([0.0357, 0.1037, 0.0407, 0.0434, 0.0773, 0.0287, 0.0557, 0.0598], device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0235, 0.0156, 0.0148, 0.0157, 0.0124, 0.0234, 0.0148], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 16:35:20,377 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49053.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:35:38,278 INFO [train.py:873] (3/4) Epoch 7, batch 3700, loss[loss=0.1623, simple_loss=0.1842, pruned_loss=0.0702, over 14265.00 frames. ], tot_loss[loss=0.1584, simple_loss=0.1782, pruned_loss=0.06931, over 1997677.24 frames. ], batch size: 57, lr: 1.13e-02, grad_scale: 8.0 2022-12-07 16:35:43,562 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49079.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:35:56,638 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9575, 2.4364, 4.0577, 4.1236, 3.9830, 2.3276, 4.1961, 3.0397], device='cuda:3'), covar=tensor([0.0202, 0.0536, 0.0569, 0.0197, 0.0189, 0.0856, 0.0152, 0.0595], device='cuda:3'), in_proj_covar=tensor([0.0225, 0.0217, 0.0327, 0.0266, 0.0212, 0.0264, 0.0223, 0.0252], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:3') 2022-12-07 16:36:02,766 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49101.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:36:26,231 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49127.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:36:30,493 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.168e+02 2.323e+02 3.026e+02 3.847e+02 8.325e+02, threshold=6.052e+02, percent-clipped=4.0 2022-12-07 16:36:41,972 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2022-12-07 16:36:57,762 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0302, 3.7719, 3.3296, 2.6524, 3.1545, 3.6411, 4.1020, 3.1662], device='cuda:3'), covar=tensor([0.0450, 0.2473, 0.1184, 0.2417, 0.1073, 0.0616, 0.0734, 0.1482], device='cuda:3'), in_proj_covar=tensor([0.0118, 0.0196, 0.0126, 0.0130, 0.0117, 0.0118, 0.0097, 0.0132], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0007, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:3') 2022-12-07 16:37:07,568 INFO [train.py:873] (3/4) Epoch 7, batch 3800, loss[loss=0.1461, simple_loss=0.1722, pruned_loss=0.05997, over 14212.00 frames. ], tot_loss[loss=0.1586, simple_loss=0.178, pruned_loss=0.06963, over 1976327.92 frames. ], batch size: 35, lr: 1.13e-02, grad_scale: 8.0 2022-12-07 16:37:11,415 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49177.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:37:26,358 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49194.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:37:30,441 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.84 vs. limit=2.0 2022-12-07 16:38:01,122 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.147e+01 2.278e+02 2.831e+02 3.417e+02 7.148e+02, threshold=5.663e+02, percent-clipped=1.0 2022-12-07 16:38:04,708 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49236.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:38:37,956 INFO [train.py:873] (3/4) Epoch 7, batch 3900, loss[loss=0.1563, simple_loss=0.1822, pruned_loss=0.06518, over 14254.00 frames. ], tot_loss[loss=0.1575, simple_loss=0.1775, pruned_loss=0.0688, over 2004584.37 frames. ], batch size: 57, lr: 1.13e-02, grad_scale: 8.0 2022-12-07 16:38:59,153 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49297.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:39:00,206 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2022-12-07 16:39:22,587 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7649, 1.4929, 2.0843, 1.6516, 1.9707, 1.4010, 1.5939, 1.7136], device='cuda:3'), covar=tensor([0.1307, 0.1741, 0.0211, 0.1290, 0.0352, 0.1181, 0.0824, 0.0398], device='cuda:3'), in_proj_covar=tensor([0.0225, 0.0240, 0.0172, 0.0326, 0.0194, 0.0246, 0.0227, 0.0184], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0005, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 16:39:30,002 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.098e+02 2.072e+02 2.772e+02 3.353e+02 6.720e+02, threshold=5.544e+02, percent-clipped=1.0 2022-12-07 16:39:34,467 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.67 vs. limit=2.0 2022-12-07 16:40:02,225 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8149, 1.5567, 2.1396, 1.8277, 1.9469, 1.4741, 1.6218, 1.6533], device='cuda:3'), covar=tensor([0.1196, 0.1969, 0.0223, 0.1393, 0.0514, 0.1007, 0.1001, 0.0468], device='cuda:3'), in_proj_covar=tensor([0.0226, 0.0242, 0.0172, 0.0328, 0.0195, 0.0248, 0.0228, 0.0185], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0005, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 16:40:05,914 INFO [train.py:873] (3/4) Epoch 7, batch 4000, loss[loss=0.1667, simple_loss=0.1497, pruned_loss=0.0918, over 1315.00 frames. ], tot_loss[loss=0.1574, simple_loss=0.1772, pruned_loss=0.06881, over 1944154.74 frames. ], batch size: 100, lr: 1.13e-02, grad_scale: 8.0 2022-12-07 16:40:05,999 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8407, 2.5722, 2.7079, 2.8724, 2.7598, 2.7743, 2.9252, 2.4526], device='cuda:3'), covar=tensor([0.0549, 0.1209, 0.0532, 0.0569, 0.0846, 0.0479, 0.0691, 0.0672], device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0240, 0.0160, 0.0151, 0.0158, 0.0126, 0.0240, 0.0151], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 16:40:49,524 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49422.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:40:58,793 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.814e+01 2.339e+02 2.890e+02 3.562e+02 6.238e+02, threshold=5.779e+02, percent-clipped=1.0 2022-12-07 16:40:58,978 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49432.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 16:40:59,828 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.2268, 1.0332, 1.3320, 1.3252, 1.0891, 0.6969, 1.3591, 1.2411], device='cuda:3'), covar=tensor([0.1368, 0.1780, 0.0581, 0.1162, 0.3193, 0.0693, 0.0861, 0.1220], device='cuda:3'), in_proj_covar=tensor([0.0019, 0.0020, 0.0020, 0.0020, 0.0020, 0.0028, 0.0021, 0.0021], device='cuda:3'), out_proj_covar=tensor([9.3222e-05, 9.4680e-05, 9.3639e-05, 9.4656e-05, 9.3701e-05, 1.1955e-04, 1.0074e-04, 9.8061e-05], device='cuda:3') 2022-12-07 16:41:28,640 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4269, 3.1067, 3.1462, 3.4486, 2.9648, 2.8463, 3.4728, 3.3922], device='cuda:3'), covar=tensor([0.0601, 0.0760, 0.0795, 0.0680, 0.1135, 0.0715, 0.0610, 0.0773], device='cuda:3'), in_proj_covar=tensor([0.0116, 0.0105, 0.0120, 0.0124, 0.0122, 0.0095, 0.0134, 0.0117], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 16:41:34,938 INFO [train.py:873] (3/4) Epoch 7, batch 4100, loss[loss=0.2031, simple_loss=0.1947, pruned_loss=0.1057, over 5936.00 frames. ], tot_loss[loss=0.159, simple_loss=0.1779, pruned_loss=0.07007, over 1851505.39 frames. ], batch size: 100, lr: 1.13e-02, grad_scale: 8.0 2022-12-07 16:41:38,599 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49477.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:41:38,673 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7705, 1.8962, 3.8564, 2.7504, 3.7277, 1.9043, 2.8530, 3.5905], device='cuda:3'), covar=tensor([0.0552, 0.4855, 0.0340, 0.6044, 0.0503, 0.3457, 0.1396, 0.0414], device='cuda:3'), in_proj_covar=tensor([0.0223, 0.0238, 0.0170, 0.0322, 0.0192, 0.0244, 0.0227, 0.0184], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0005, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 16:41:40,634 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.68 vs. limit=2.0 2022-12-07 16:41:43,971 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49483.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:41:49,941 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8232, 1.4072, 3.4230, 3.1738, 3.3229, 3.4592, 2.7538, 3.4555], device='cuda:3'), covar=tensor([0.1312, 0.1393, 0.0099, 0.0231, 0.0196, 0.0110, 0.0247, 0.0126], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0162, 0.0114, 0.0159, 0.0130, 0.0129, 0.0107, 0.0112], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 16:41:52,463 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49493.0, num_to_drop=1, layers_to_drop={3} 2022-12-07 16:41:53,490 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49494.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:42:20,804 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49525.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:42:23,784 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0 2022-12-07 16:42:26,657 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.128e+02 2.350e+02 2.777e+02 3.332e+02 5.900e+02, threshold=5.554e+02, percent-clipped=2.0 2022-12-07 16:42:35,643 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49542.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:43:03,203 INFO [train.py:873] (3/4) Epoch 7, batch 4200, loss[loss=0.1193, simple_loss=0.1606, pruned_loss=0.039, over 14281.00 frames. ], tot_loss[loss=0.1577, simple_loss=0.1775, pruned_loss=0.06901, over 1879672.22 frames. ], batch size: 25, lr: 1.12e-02, grad_scale: 4.0 2022-12-07 16:43:18,432 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.81 vs. limit=2.0 2022-12-07 16:43:20,475 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49592.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:43:31,320 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.6246, 2.3714, 3.4011, 2.0358, 2.2852, 2.5286, 1.3303, 2.5603], device='cuda:3'), covar=tensor([0.1367, 0.1473, 0.0426, 0.2834, 0.2150, 0.1103, 0.4562, 0.1281], device='cuda:3'), in_proj_covar=tensor([0.0072, 0.0086, 0.0076, 0.0087, 0.0109, 0.0070, 0.0127, 0.0076], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2022-12-07 16:43:32,348 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2022-12-07 16:43:57,174 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.547e+02 2.513e+02 3.123e+02 4.078e+02 1.360e+03, threshold=6.245e+02, percent-clipped=14.0 2022-12-07 16:44:15,075 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2022-12-07 16:44:32,552 INFO [train.py:873] (3/4) Epoch 7, batch 4300, loss[loss=0.1526, simple_loss=0.179, pruned_loss=0.06311, over 14253.00 frames. ], tot_loss[loss=0.1592, simple_loss=0.1785, pruned_loss=0.06996, over 1890860.61 frames. ], batch size: 63, lr: 1.12e-02, grad_scale: 4.0 2022-12-07 16:45:14,887 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5319, 3.7329, 2.7697, 4.6810, 4.2856, 4.4164, 3.8040, 3.1828], device='cuda:3'), covar=tensor([0.0448, 0.1359, 0.5173, 0.0373, 0.0852, 0.1248, 0.1149, 0.3670], device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0298, 0.0286, 0.0208, 0.0273, 0.0271, 0.0250, 0.0270], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 16:45:15,143 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.93 vs. limit=2.0 2022-12-07 16:45:25,736 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.275e+02 2.563e+02 3.565e+02 4.633e+02 7.506e+02, threshold=7.130e+02, percent-clipped=5.0 2022-12-07 16:45:38,877 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9226, 1.7959, 1.6318, 2.0298, 1.8653, 1.9703, 1.8676, 1.7069], device='cuda:3'), covar=tensor([0.0501, 0.0756, 0.1192, 0.0301, 0.0665, 0.0314, 0.0848, 0.0481], device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0299, 0.0285, 0.0207, 0.0272, 0.0271, 0.0251, 0.0271], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 16:45:56,356 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49766.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:46:02,012 INFO [train.py:873] (3/4) Epoch 7, batch 4400, loss[loss=0.1645, simple_loss=0.1755, pruned_loss=0.07669, over 6006.00 frames. ], tot_loss[loss=0.1587, simple_loss=0.178, pruned_loss=0.06966, over 1929818.52 frames. ], batch size: 100, lr: 1.12e-02, grad_scale: 8.0 2022-12-07 16:46:06,948 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49778.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:46:09,037 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2022-12-07 16:46:15,393 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49788.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 16:46:23,491 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.24 vs. limit=5.0 2022-12-07 16:46:43,278 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2022-12-07 16:46:45,846 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.41 vs. limit=5.0 2022-12-07 16:46:50,916 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49827.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:46:55,944 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.417e+01 2.268e+02 2.875e+02 3.664e+02 5.837e+02, threshold=5.750e+02, percent-clipped=0.0 2022-12-07 16:47:31,486 INFO [train.py:873] (3/4) Epoch 7, batch 4500, loss[loss=0.1571, simple_loss=0.1701, pruned_loss=0.07199, over 5970.00 frames. ], tot_loss[loss=0.1578, simple_loss=0.1777, pruned_loss=0.06897, over 1943162.43 frames. ], batch size: 100, lr: 1.12e-02, grad_scale: 8.0 2022-12-07 16:47:48,498 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49892.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:48:24,074 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.19 vs. limit=5.0 2022-12-07 16:48:25,119 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.380e+02 2.362e+02 2.957e+02 3.544e+02 6.398e+02, threshold=5.915e+02, percent-clipped=1.0 2022-12-07 16:48:31,737 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49940.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:48:50,650 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0036, 2.9156, 2.5693, 2.6672, 2.9454, 2.9143, 3.0117, 2.9611], device='cuda:3'), covar=tensor([0.0909, 0.0653, 0.2129, 0.2867, 0.0822, 0.0940, 0.1111, 0.1012], device='cuda:3'), in_proj_covar=tensor([0.0318, 0.0229, 0.0384, 0.0483, 0.0285, 0.0358, 0.0347, 0.0307], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 16:49:00,840 INFO [train.py:873] (3/4) Epoch 7, batch 4600, loss[loss=0.1511, simple_loss=0.1473, pruned_loss=0.07751, over 2619.00 frames. ], tot_loss[loss=0.159, simple_loss=0.1787, pruned_loss=0.06967, over 2013610.25 frames. ], batch size: 100, lr: 1.12e-02, grad_scale: 8.0 2022-12-07 16:49:31,430 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3128, 3.9840, 4.0889, 4.3335, 3.9425, 3.5501, 4.3479, 4.2405], device='cuda:3'), covar=tensor([0.0604, 0.0601, 0.0592, 0.0625, 0.0675, 0.0603, 0.0608, 0.0698], device='cuda:3'), in_proj_covar=tensor([0.0117, 0.0108, 0.0121, 0.0125, 0.0126, 0.0097, 0.0135, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 16:49:58,507 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.255e+02 2.301e+02 2.897e+02 3.880e+02 6.571e+02, threshold=5.793e+02, percent-clipped=2.0 2022-12-07 16:50:14,909 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0992, 1.7769, 4.6278, 4.3052, 4.1419, 4.7278, 4.2810, 4.7351], device='cuda:3'), covar=tensor([0.1207, 0.1369, 0.0064, 0.0109, 0.0149, 0.0078, 0.0145, 0.0080], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0163, 0.0115, 0.0160, 0.0132, 0.0131, 0.0107, 0.0114], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 16:50:25,766 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5417, 1.8505, 2.3575, 2.1954, 2.4726, 2.4106, 2.2277, 2.2842], device='cuda:3'), covar=tensor([0.0397, 0.2227, 0.0634, 0.1301, 0.0406, 0.0777, 0.0503, 0.1216], device='cuda:3'), in_proj_covar=tensor([0.0314, 0.0325, 0.0382, 0.0307, 0.0364, 0.0306, 0.0361, 0.0339], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 16:50:32,305 INFO [train.py:873] (3/4) Epoch 7, batch 4700, loss[loss=0.17, simple_loss=0.1808, pruned_loss=0.07961, over 6964.00 frames. ], tot_loss[loss=0.1584, simple_loss=0.1779, pruned_loss=0.06944, over 1986478.35 frames. ], batch size: 100, lr: 1.12e-02, grad_scale: 4.0 2022-12-07 16:50:37,280 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50078.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:50:45,872 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50088.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 16:50:48,277 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9567, 2.0675, 1.9101, 2.0619, 1.7244, 1.9156, 2.0191, 2.0362], device='cuda:3'), covar=tensor([0.1029, 0.0938, 0.1076, 0.0967, 0.1409, 0.0915, 0.1120, 0.0968], device='cuda:3'), in_proj_covar=tensor([0.0117, 0.0109, 0.0123, 0.0128, 0.0128, 0.0098, 0.0136, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 16:51:15,732 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50122.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:51:19,174 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50126.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:51:20,981 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50128.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:51:25,948 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.013e+02 2.271e+02 2.971e+02 3.641e+02 8.155e+02, threshold=5.943e+02, percent-clipped=3.0 2022-12-07 16:51:27,665 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50136.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 16:52:00,692 INFO [train.py:873] (3/4) Epoch 7, batch 4800, loss[loss=0.1529, simple_loss=0.1789, pruned_loss=0.06349, over 14298.00 frames. ], tot_loss[loss=0.1567, simple_loss=0.1767, pruned_loss=0.06838, over 1909417.99 frames. ], batch size: 60, lr: 1.12e-02, grad_scale: 8.0 2022-12-07 16:52:14,895 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.2057, 1.2711, 1.1375, 1.0893, 1.3151, 0.3649, 1.2481, 1.3344], device='cuda:3'), covar=tensor([0.0807, 0.1140, 0.1272, 0.2015, 0.1083, 0.0672, 0.2138, 0.2041], device='cuda:3'), in_proj_covar=tensor([0.0019, 0.0020, 0.0020, 0.0020, 0.0020, 0.0029, 0.0022, 0.0021], device='cuda:3'), out_proj_covar=tensor([9.3107e-05, 9.4986e-05, 9.4522e-05, 9.4032e-05, 9.4766e-05, 1.2271e-04, 1.0299e-04, 9.7735e-05], device='cuda:3') 2022-12-07 16:52:14,913 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50189.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:52:22,025 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50197.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 16:52:28,149 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.85 vs. limit=2.0 2022-12-07 16:52:54,460 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.360e+02 2.403e+02 3.099e+02 4.047e+02 7.554e+02, threshold=6.197e+02, percent-clipped=4.0 2022-12-07 16:53:11,395 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3173, 2.9213, 4.0532, 3.1897, 4.1642, 4.0962, 4.0487, 3.4486], device='cuda:3'), covar=tensor([0.0533, 0.2846, 0.1016, 0.1889, 0.0716, 0.0610, 0.1372, 0.2181], device='cuda:3'), in_proj_covar=tensor([0.0308, 0.0320, 0.0375, 0.0303, 0.0360, 0.0300, 0.0350, 0.0331], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 16:53:15,861 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50258.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 16:53:16,756 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9522, 1.7017, 1.9687, 1.7667, 2.0750, 1.8259, 1.6112, 1.9101], device='cuda:3'), covar=tensor([0.0297, 0.0926, 0.0126, 0.0224, 0.0151, 0.0340, 0.0125, 0.0221], device='cuda:3'), in_proj_covar=tensor([0.0308, 0.0321, 0.0375, 0.0304, 0.0361, 0.0300, 0.0350, 0.0331], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 16:53:29,366 INFO [train.py:873] (3/4) Epoch 7, batch 4900, loss[loss=0.1287, simple_loss=0.1646, pruned_loss=0.04638, over 14615.00 frames. ], tot_loss[loss=0.1559, simple_loss=0.1762, pruned_loss=0.06782, over 1910554.31 frames. ], batch size: 22, lr: 1.12e-02, grad_scale: 8.0 2022-12-07 16:54:23,753 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.482e+02 2.354e+02 3.057e+02 3.894e+02 8.460e+02, threshold=6.115e+02, percent-clipped=5.0 2022-12-07 16:54:58,366 INFO [train.py:873] (3/4) Epoch 7, batch 5000, loss[loss=0.1747, simple_loss=0.1893, pruned_loss=0.08003, over 14344.00 frames. ], tot_loss[loss=0.157, simple_loss=0.1771, pruned_loss=0.06839, over 1922453.99 frames. ], batch size: 66, lr: 1.12e-02, grad_scale: 8.0 2022-12-07 16:55:09,296 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1238, 3.1196, 2.9091, 3.2839, 3.0062, 2.7922, 1.1472, 2.8463], device='cuda:3'), covar=tensor([0.0604, 0.0652, 0.1178, 0.0511, 0.0962, 0.1305, 0.5019, 0.0713], device='cuda:3'), in_proj_covar=tensor([0.0138, 0.0145, 0.0128, 0.0122, 0.0176, 0.0122, 0.0150, 0.0166], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 16:55:42,560 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50422.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:55:46,065 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1994, 3.3919, 2.5850, 4.3244, 4.1214, 4.0983, 3.4994, 2.7741], device='cuda:3'), covar=tensor([0.0648, 0.1764, 0.5327, 0.0347, 0.0774, 0.1185, 0.1449, 0.4770], device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0307, 0.0286, 0.0211, 0.0274, 0.0276, 0.0256, 0.0271], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 16:55:52,925 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.022e+02 2.513e+02 3.076e+02 4.134e+02 7.861e+02, threshold=6.153e+02, percent-clipped=5.0 2022-12-07 16:56:03,162 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5773, 2.4353, 2.1923, 2.2445, 2.5013, 2.4800, 2.5647, 2.5210], device='cuda:3'), covar=tensor([0.1076, 0.0847, 0.2469, 0.2920, 0.1037, 0.1126, 0.1330, 0.1243], device='cuda:3'), in_proj_covar=tensor([0.0330, 0.0238, 0.0393, 0.0497, 0.0292, 0.0364, 0.0360, 0.0317], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 16:56:03,959 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1333, 4.2608, 4.5471, 3.7082, 4.3614, 4.5186, 1.5649, 4.1125], device='cuda:3'), covar=tensor([0.0228, 0.0308, 0.0321, 0.0543, 0.0251, 0.0198, 0.3075, 0.0255], device='cuda:3'), in_proj_covar=tensor([0.0138, 0.0145, 0.0127, 0.0122, 0.0175, 0.0121, 0.0150, 0.0164], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 16:56:24,885 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50470.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:56:27,488 INFO [train.py:873] (3/4) Epoch 7, batch 5100, loss[loss=0.1443, simple_loss=0.1689, pruned_loss=0.05986, over 14241.00 frames. ], tot_loss[loss=0.1574, simple_loss=0.1773, pruned_loss=0.06878, over 1929314.98 frames. ], batch size: 46, lr: 1.11e-02, grad_scale: 8.0 2022-12-07 16:56:37,635 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50484.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:57:20,713 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.225e+02 2.452e+02 3.012e+02 3.839e+02 6.174e+02, threshold=6.024e+02, percent-clipped=1.0 2022-12-07 16:57:32,001 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50546.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:57:38,432 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50553.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 16:57:49,808 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2032, 2.2531, 3.0542, 2.3984, 3.0044, 3.0068, 2.8669, 2.3861], device='cuda:3'), covar=tensor([0.0585, 0.2958, 0.0888, 0.2134, 0.0963, 0.0821, 0.1302, 0.2439], device='cuda:3'), in_proj_covar=tensor([0.0306, 0.0320, 0.0375, 0.0307, 0.0362, 0.0300, 0.0352, 0.0331], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 16:57:55,459 INFO [train.py:873] (3/4) Epoch 7, batch 5200, loss[loss=0.1703, simple_loss=0.1799, pruned_loss=0.08031, over 4979.00 frames. ], tot_loss[loss=0.1575, simple_loss=0.1774, pruned_loss=0.06886, over 1908921.58 frames. ], batch size: 100, lr: 1.11e-02, grad_scale: 8.0 2022-12-07 16:58:16,270 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7183, 1.5028, 3.6629, 1.6583, 3.7265, 3.8371, 2.8462, 4.0582], device='cuda:3'), covar=tensor([0.0200, 0.3000, 0.0333, 0.2344, 0.0439, 0.0295, 0.0699, 0.0160], device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0161, 0.0149, 0.0170, 0.0162, 0.0162, 0.0132, 0.0133], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 16:58:26,299 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50607.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 16:58:49,801 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.178e+02 2.234e+02 2.881e+02 3.626e+02 9.172e+02, threshold=5.763e+02, percent-clipped=1.0 2022-12-07 16:59:24,244 INFO [train.py:873] (3/4) Epoch 7, batch 5300, loss[loss=0.1279, simple_loss=0.1615, pruned_loss=0.04719, over 14250.00 frames. ], tot_loss[loss=0.1561, simple_loss=0.1764, pruned_loss=0.06787, over 1914920.86 frames. ], batch size: 39, lr: 1.11e-02, grad_scale: 8.0 2022-12-07 16:59:36,828 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.80 vs. limit=5.0 2022-12-07 16:59:39,807 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.9565, 0.9339, 0.7802, 0.9194, 1.0020, 0.5040, 0.9649, 0.8755], device='cuda:3'), covar=tensor([0.0438, 0.1031, 0.0720, 0.0780, 0.0642, 0.0809, 0.0506, 0.0604], device='cuda:3'), in_proj_covar=tensor([0.0019, 0.0021, 0.0021, 0.0021, 0.0020, 0.0029, 0.0020, 0.0021], device='cuda:3'), out_proj_covar=tensor([9.2804e-05, 9.6674e-05, 9.7224e-05, 9.6799e-05, 9.6593e-05, 1.2452e-04, 1.0028e-04, 9.9267e-05], device='cuda:3') 2022-12-07 17:00:16,768 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.527e+02 2.441e+02 3.172e+02 3.955e+02 6.877e+02, threshold=6.344e+02, percent-clipped=8.0 2022-12-07 17:00:49,951 INFO [train.py:873] (3/4) Epoch 7, batch 5400, loss[loss=0.1439, simple_loss=0.1361, pruned_loss=0.07588, over 1237.00 frames. ], tot_loss[loss=0.1564, simple_loss=0.1766, pruned_loss=0.06808, over 1920995.12 frames. ], batch size: 100, lr: 1.11e-02, grad_scale: 8.0 2022-12-07 17:00:58,377 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50782.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:00:59,896 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50784.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:01:18,488 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6523, 3.4866, 3.4438, 3.8169, 3.2974, 3.1890, 3.7948, 3.7457], device='cuda:3'), covar=tensor([0.0741, 0.0762, 0.0758, 0.0586, 0.0903, 0.0614, 0.0641, 0.0629], device='cuda:3'), in_proj_covar=tensor([0.0119, 0.0108, 0.0124, 0.0126, 0.0125, 0.0096, 0.0137, 0.0119], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 17:01:20,933 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50808.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 17:01:41,868 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50832.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:01:43,630 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.515e+02 2.469e+02 3.143e+02 3.850e+02 7.689e+02, threshold=6.287e+02, percent-clipped=2.0 2022-12-07 17:01:47,879 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.9144, 0.7798, 0.8218, 0.9719, 0.9264, 0.5423, 0.8876, 0.8622], device='cuda:3'), covar=tensor([0.0577, 0.1072, 0.0505, 0.0596, 0.0648, 0.0789, 0.0688, 0.0887], device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0021, 0.0021, 0.0021, 0.0021, 0.0030, 0.0021, 0.0021], device='cuda:3'), out_proj_covar=tensor([9.4349e-05, 9.8784e-05, 9.6949e-05, 9.7403e-05, 9.8959e-05, 1.2699e-04, 1.0191e-04, 9.9595e-05], device='cuda:3') 2022-12-07 17:01:51,279 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50843.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:01:59,984 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50853.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 17:02:01,326 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2022-12-07 17:02:14,146 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50869.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 17:02:17,735 INFO [train.py:873] (3/4) Epoch 7, batch 5500, loss[loss=0.1434, simple_loss=0.1353, pruned_loss=0.07577, over 2568.00 frames. ], tot_loss[loss=0.1564, simple_loss=0.1767, pruned_loss=0.06807, over 1976874.00 frames. ], batch size: 100, lr: 1.11e-02, grad_scale: 8.0 2022-12-07 17:02:42,269 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50901.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 17:02:43,095 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50902.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:03:10,733 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.245e+02 2.288e+02 2.905e+02 3.483e+02 6.290e+02, threshold=5.809e+02, percent-clipped=1.0 2022-12-07 17:03:21,395 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50946.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:03:44,819 INFO [train.py:873] (3/4) Epoch 7, batch 5600, loss[loss=0.1403, simple_loss=0.1354, pruned_loss=0.07259, over 2658.00 frames. ], tot_loss[loss=0.156, simple_loss=0.1763, pruned_loss=0.06783, over 1994857.43 frames. ], batch size: 100, lr: 1.11e-02, grad_scale: 8.0 2022-12-07 17:03:53,530 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2022-12-07 17:04:15,907 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51007.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:04:38,633 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.538e+02 2.390e+02 3.229e+02 4.054e+02 7.897e+02, threshold=6.457e+02, percent-clipped=5.0 2022-12-07 17:04:46,733 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7343, 1.3913, 2.9480, 1.6096, 3.0089, 2.9424, 2.0714, 3.0516], device='cuda:3'), covar=tensor([0.0251, 0.2397, 0.0280, 0.1707, 0.0246, 0.0293, 0.0906, 0.0171], device='cuda:3'), in_proj_covar=tensor([0.0160, 0.0158, 0.0146, 0.0167, 0.0158, 0.0158, 0.0127, 0.0130], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:3') 2022-12-07 17:05:03,874 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51062.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:05:13,367 INFO [train.py:873] (3/4) Epoch 7, batch 5700, loss[loss=0.1559, simple_loss=0.1497, pruned_loss=0.08109, over 1225.00 frames. ], tot_loss[loss=0.1569, simple_loss=0.1768, pruned_loss=0.0685, over 1993886.90 frames. ], batch size: 100, lr: 1.11e-02, grad_scale: 8.0 2022-12-07 17:05:56,753 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51123.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:06:05,990 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.357e+02 2.301e+02 2.940e+02 3.681e+02 9.913e+02, threshold=5.881e+02, percent-clipped=3.0 2022-12-07 17:06:07,392 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2022-12-07 17:06:07,819 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0088, 3.7573, 3.5680, 3.6816, 3.8627, 3.8943, 3.9783, 3.9500], device='cuda:3'), covar=tensor([0.0777, 0.0659, 0.2117, 0.2729, 0.0691, 0.0718, 0.1072, 0.0952], device='cuda:3'), in_proj_covar=tensor([0.0327, 0.0234, 0.0389, 0.0492, 0.0290, 0.0354, 0.0350, 0.0305], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 17:06:09,837 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51138.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:06:31,996 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51164.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 17:06:32,878 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1105, 0.8327, 1.0728, 1.1674, 1.2029, 0.6994, 1.1067, 1.0508], device='cuda:3'), covar=tensor([0.1492, 0.1438, 0.0867, 0.0847, 0.1121, 0.0841, 0.0766, 0.0630], device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0021, 0.0021, 0.0020, 0.0020, 0.0030, 0.0021, 0.0021], device='cuda:3'), out_proj_covar=tensor([9.5065e-05, 9.7156e-05, 9.6938e-05, 9.5947e-05, 9.6900e-05, 1.2702e-04, 1.0122e-04, 9.8557e-05], device='cuda:3') 2022-12-07 17:06:39,100 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51172.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 17:06:39,724 INFO [train.py:873] (3/4) Epoch 7, batch 5800, loss[loss=0.149, simple_loss=0.175, pruned_loss=0.06144, over 14232.00 frames. ], tot_loss[loss=0.157, simple_loss=0.1771, pruned_loss=0.0684, over 2015602.46 frames. ], batch size: 32, lr: 1.11e-02, grad_scale: 8.0 2022-12-07 17:07:05,756 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51202.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:07:32,504 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5552, 2.2896, 4.5883, 3.0620, 4.3381, 2.2943, 3.4971, 4.2442], device='cuda:3'), covar=tensor([0.0315, 0.4579, 0.0270, 0.8298, 0.0315, 0.3281, 0.1054, 0.0384], device='cuda:3'), in_proj_covar=tensor([0.0223, 0.0232, 0.0174, 0.0318, 0.0190, 0.0239, 0.0220, 0.0181], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0005, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 17:07:33,330 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51233.0, num_to_drop=1, layers_to_drop={3} 2022-12-07 17:07:33,900 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.270e+02 2.265e+02 2.888e+02 3.967e+02 8.048e+02, threshold=5.777e+02, percent-clipped=2.0 2022-12-07 17:07:48,090 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51250.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:08:08,703 INFO [train.py:873] (3/4) Epoch 7, batch 5900, loss[loss=0.1567, simple_loss=0.1508, pruned_loss=0.08134, over 2655.00 frames. ], tot_loss[loss=0.1573, simple_loss=0.1773, pruned_loss=0.06865, over 1990235.90 frames. ], batch size: 100, lr: 1.11e-02, grad_scale: 8.0 2022-12-07 17:08:34,080 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51302.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:09:01,666 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.784e+01 2.421e+02 2.907e+02 3.542e+02 5.776e+02, threshold=5.815e+02, percent-clipped=0.0 2022-12-07 17:09:27,036 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.0952, 0.9561, 1.1020, 1.1327, 0.9956, 0.7083, 1.1674, 0.9848], device='cuda:3'), covar=tensor([0.0992, 0.1103, 0.0767, 0.0806, 0.1320, 0.0965, 0.0682, 0.0677], device='cuda:3'), in_proj_covar=tensor([0.0019, 0.0020, 0.0020, 0.0020, 0.0020, 0.0029, 0.0020, 0.0020], device='cuda:3'), out_proj_covar=tensor([9.3267e-05, 9.5548e-05, 9.5299e-05, 9.4083e-05, 9.6860e-05, 1.2469e-04, 9.9198e-05, 9.7176e-05], device='cuda:3') 2022-12-07 17:09:35,646 INFO [train.py:873] (3/4) Epoch 7, batch 6000, loss[loss=0.2072, simple_loss=0.2094, pruned_loss=0.1025, over 8651.00 frames. ], tot_loss[loss=0.1572, simple_loss=0.1769, pruned_loss=0.06878, over 1934569.19 frames. ], batch size: 100, lr: 1.10e-02, grad_scale: 8.0 2022-12-07 17:09:35,646 INFO [train.py:896] (3/4) Computing validation loss 2022-12-07 17:09:56,742 INFO [train.py:905] (3/4) Epoch 7, validation: loss=0.1227, simple_loss=0.1653, pruned_loss=0.04007, over 857387.00 frames. 2022-12-07 17:09:56,743 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-07 17:10:04,638 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51382.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:10:13,323 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51392.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:10:36,276 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51418.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:10:49,690 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.039e+02 2.553e+02 3.284e+02 4.114e+02 1.252e+03, threshold=6.568e+02, percent-clipped=8.0 2022-12-07 17:10:53,285 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51438.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:10:57,557 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51443.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:11:06,379 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51453.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:11:15,600 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2022-12-07 17:11:15,937 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51464.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 17:11:23,461 INFO [train.py:873] (3/4) Epoch 7, batch 6100, loss[loss=0.19, simple_loss=0.1923, pruned_loss=0.0938, over 4999.00 frames. ], tot_loss[loss=0.1572, simple_loss=0.1774, pruned_loss=0.06846, over 1998157.25 frames. ], batch size: 100, lr: 1.10e-02, grad_scale: 8.0 2022-12-07 17:11:34,630 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51486.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:11:42,553 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51495.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:11:56,669 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8252, 3.0869, 4.5041, 3.5388, 4.7830, 4.5077, 4.4154, 3.8856], device='cuda:3'), covar=tensor([0.0485, 0.3064, 0.0925, 0.1741, 0.0516, 0.0671, 0.1535, 0.2260], device='cuda:3'), in_proj_covar=tensor([0.0301, 0.0324, 0.0375, 0.0302, 0.0362, 0.0296, 0.0351, 0.0327], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 17:11:57,343 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51512.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 17:12:10,075 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1024, 2.3686, 2.2577, 2.4793, 1.9452, 2.4334, 2.2162, 1.1229], device='cuda:3'), covar=tensor([0.2786, 0.1223, 0.1220, 0.0712, 0.1337, 0.0580, 0.1306, 0.3774], device='cuda:3'), in_proj_covar=tensor([0.0157, 0.0065, 0.0054, 0.0056, 0.0082, 0.0061, 0.0088, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005], device='cuda:3') 2022-12-07 17:12:10,894 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51528.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 17:12:16,320 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.479e+02 2.483e+02 3.170e+02 3.695e+02 8.405e+02, threshold=6.340e+02, percent-clipped=2.0 2022-12-07 17:12:35,820 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51556.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:12:50,522 INFO [train.py:873] (3/4) Epoch 7, batch 6200, loss[loss=0.1368, simple_loss=0.1346, pruned_loss=0.06947, over 2570.00 frames. ], tot_loss[loss=0.1581, simple_loss=0.1777, pruned_loss=0.0692, over 1973723.50 frames. ], batch size: 100, lr: 1.10e-02, grad_scale: 8.0 2022-12-07 17:13:01,286 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6292, 2.2344, 3.5188, 3.6326, 3.5850, 2.2527, 3.3951, 2.7842], device='cuda:3'), covar=tensor([0.0211, 0.0531, 0.0583, 0.0298, 0.0210, 0.0856, 0.0187, 0.0606], device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0222, 0.0330, 0.0272, 0.0217, 0.0266, 0.0230, 0.0254], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 17:13:16,636 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51602.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:13:44,670 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.366e+02 2.329e+02 2.979e+02 3.715e+02 8.886e+02, threshold=5.958e+02, percent-clipped=2.0 2022-12-07 17:13:56,258 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0306, 2.3907, 2.3027, 2.4574, 1.9353, 2.3941, 2.1294, 1.1071], device='cuda:3'), covar=tensor([0.1966, 0.0989, 0.0943, 0.0549, 0.1200, 0.0531, 0.1297, 0.3328], device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0065, 0.0054, 0.0056, 0.0082, 0.0061, 0.0087, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005], device='cuda:3') 2022-12-07 17:13:58,961 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51650.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:14:16,728 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2022-12-07 17:14:19,070 INFO [train.py:873] (3/4) Epoch 7, batch 6300, loss[loss=0.1529, simple_loss=0.1718, pruned_loss=0.06698, over 14250.00 frames. ], tot_loss[loss=0.1562, simple_loss=0.1765, pruned_loss=0.06793, over 1986310.47 frames. ], batch size: 80, lr: 1.10e-02, grad_scale: 8.0 2022-12-07 17:14:58,622 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51718.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:15:04,893 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2057, 2.1937, 3.2191, 3.3549, 3.2698, 2.1557, 3.2662, 2.6548], device='cuda:3'), covar=tensor([0.0202, 0.0433, 0.0407, 0.0251, 0.0179, 0.0659, 0.0147, 0.0462], device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0222, 0.0328, 0.0271, 0.0215, 0.0263, 0.0230, 0.0254], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 17:15:05,545 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0099, 1.9993, 2.0392, 2.0421, 1.9876, 1.7620, 1.1640, 1.7344], device='cuda:3'), covar=tensor([0.0485, 0.0420, 0.0499, 0.0310, 0.0390, 0.0899, 0.2212, 0.0372], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0149, 0.0129, 0.0121, 0.0178, 0.0122, 0.0152, 0.0167], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 17:15:12,417 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.367e+02 2.206e+02 2.744e+02 3.241e+02 7.232e+02, threshold=5.488e+02, percent-clipped=1.0 2022-12-07 17:15:16,193 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2975, 3.0303, 3.1079, 3.2836, 3.1326, 3.2399, 3.3663, 2.7716], device='cuda:3'), covar=tensor([0.0416, 0.1085, 0.0421, 0.0540, 0.0826, 0.0362, 0.0601, 0.0633], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0236, 0.0163, 0.0154, 0.0159, 0.0127, 0.0238, 0.0147], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 17:15:16,199 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51738.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:15:24,000 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51747.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:15:24,662 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51748.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:15:40,165 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51766.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:15:46,559 INFO [train.py:873] (3/4) Epoch 7, batch 6400, loss[loss=0.15, simple_loss=0.1616, pruned_loss=0.06923, over 4936.00 frames. ], tot_loss[loss=0.1567, simple_loss=0.177, pruned_loss=0.06817, over 1996379.58 frames. ], batch size: 100, lr: 1.10e-02, grad_scale: 8.0 2022-12-07 17:16:02,268 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8379, 2.3826, 3.1641, 2.2463, 2.1468, 2.7386, 1.5095, 2.6643], device='cuda:3'), covar=tensor([0.1244, 0.1700, 0.1038, 0.2998, 0.2901, 0.1219, 0.5472, 0.1321], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0086, 0.0082, 0.0089, 0.0109, 0.0072, 0.0130, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2022-12-07 17:16:10,615 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2873, 1.3769, 2.5235, 1.4502, 2.4918, 2.4720, 2.0591, 2.5925], device='cuda:3'), covar=tensor([0.0282, 0.2246, 0.0308, 0.1758, 0.0345, 0.0441, 0.0845, 0.0250], device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0163, 0.0150, 0.0171, 0.0164, 0.0163, 0.0130, 0.0133], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 17:16:17,620 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51808.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:16:34,997 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51828.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 17:16:39,968 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.142e+02 2.371e+02 3.096e+02 4.074e+02 1.030e+03, threshold=6.192e+02, percent-clipped=3.0 2022-12-07 17:16:55,642 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51851.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:16:57,358 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2719, 1.8706, 2.2823, 2.3721, 1.9647, 1.9108, 2.2364, 2.1843], device='cuda:3'), covar=tensor([0.0107, 0.0247, 0.0099, 0.0071, 0.0156, 0.0301, 0.0127, 0.0137], device='cuda:3'), in_proj_covar=tensor([0.0235, 0.0223, 0.0330, 0.0272, 0.0217, 0.0265, 0.0230, 0.0255], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 17:17:14,511 INFO [train.py:873] (3/4) Epoch 7, batch 6500, loss[loss=0.1485, simple_loss=0.1784, pruned_loss=0.05929, over 14279.00 frames. ], tot_loss[loss=0.1575, simple_loss=0.1773, pruned_loss=0.06889, over 1889855.98 frames. ], batch size: 31, lr: 1.10e-02, grad_scale: 8.0 2022-12-07 17:17:17,362 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51876.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 17:18:07,971 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.565e+02 2.306e+02 2.865e+02 3.658e+02 7.175e+02, threshold=5.730e+02, percent-clipped=1.0 2022-12-07 17:18:16,916 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2942, 1.4485, 2.5258, 1.3943, 2.5381, 2.4643, 2.0373, 2.6032], device='cuda:3'), covar=tensor([0.0329, 0.1964, 0.0309, 0.1815, 0.0340, 0.0459, 0.0816, 0.0213], device='cuda:3'), in_proj_covar=tensor([0.0162, 0.0160, 0.0148, 0.0170, 0.0161, 0.0161, 0.0128, 0.0130], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 17:18:42,429 INFO [train.py:873] (3/4) Epoch 7, batch 6600, loss[loss=0.1722, simple_loss=0.1886, pruned_loss=0.07785, over 6932.00 frames. ], tot_loss[loss=0.1592, simple_loss=0.178, pruned_loss=0.07021, over 1882194.60 frames. ], batch size: 100, lr: 1.10e-02, grad_scale: 8.0 2022-12-07 17:19:02,820 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4532, 1.8488, 1.8774, 1.9597, 1.7439, 1.9220, 1.4713, 1.1400], device='cuda:3'), covar=tensor([0.1631, 0.0853, 0.0367, 0.0314, 0.1079, 0.0373, 0.1976, 0.2364], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0063, 0.0053, 0.0055, 0.0081, 0.0059, 0.0086, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0005], device='cuda:3') 2022-12-07 17:19:07,539 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1967, 2.1996, 3.1870, 3.2828, 3.1979, 2.1907, 3.2377, 2.5081], device='cuda:3'), covar=tensor([0.0186, 0.0434, 0.0358, 0.0202, 0.0182, 0.0673, 0.0156, 0.0469], device='cuda:3'), in_proj_covar=tensor([0.0236, 0.0221, 0.0331, 0.0276, 0.0217, 0.0266, 0.0230, 0.0255], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:3') 2022-12-07 17:19:10,533 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2022-12-07 17:19:10,629 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.65 vs. limit=5.0 2022-12-07 17:19:15,708 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1592, 4.1218, 4.4953, 3.8437, 4.2142, 4.4842, 1.7440, 3.9510], device='cuda:3'), covar=tensor([0.0233, 0.0299, 0.0334, 0.0425, 0.0301, 0.0202, 0.3047, 0.0267], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0145, 0.0128, 0.0121, 0.0176, 0.0118, 0.0150, 0.0165], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 17:19:35,920 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.186e+02 2.181e+02 2.663e+02 3.409e+02 8.463e+02, threshold=5.326e+02, percent-clipped=3.0 2022-12-07 17:19:39,836 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52038.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:19:41,502 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52040.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:19:48,547 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52048.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:20:10,105 INFO [train.py:873] (3/4) Epoch 7, batch 6700, loss[loss=0.1644, simple_loss=0.1487, pruned_loss=0.0901, over 1272.00 frames. ], tot_loss[loss=0.1575, simple_loss=0.1771, pruned_loss=0.06893, over 1927679.03 frames. ], batch size: 100, lr: 1.10e-02, grad_scale: 16.0 2022-12-07 17:20:12,799 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8289, 1.7002, 1.9709, 1.7897, 2.0855, 1.7824, 1.7403, 1.8937], device='cuda:3'), covar=tensor([0.0483, 0.1139, 0.0218, 0.0357, 0.0292, 0.0605, 0.0219, 0.0344], device='cuda:3'), in_proj_covar=tensor([0.0313, 0.0333, 0.0385, 0.0309, 0.0371, 0.0303, 0.0356, 0.0331], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 17:20:21,601 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52086.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:20:30,299 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52096.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:20:35,003 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52101.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:20:36,546 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52103.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:20:42,663 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9855, 3.6369, 3.1875, 2.7384, 3.2557, 3.6086, 4.0590, 3.3032], device='cuda:3'), covar=tensor([0.0455, 0.2091, 0.0963, 0.1643, 0.0737, 0.0485, 0.0527, 0.1066], device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0193, 0.0122, 0.0127, 0.0116, 0.0117, 0.0097, 0.0130], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0007, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:3') 2022-12-07 17:21:03,676 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.373e+02 2.435e+02 3.169e+02 4.506e+02 1.155e+03, threshold=6.339e+02, percent-clipped=9.0 2022-12-07 17:21:07,758 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8586, 1.5903, 2.1212, 1.6691, 1.9141, 1.5052, 1.6685, 1.7920], device='cuda:3'), covar=tensor([0.1457, 0.1761, 0.0229, 0.1108, 0.0552, 0.1073, 0.0891, 0.0476], device='cuda:3'), in_proj_covar=tensor([0.0228, 0.0240, 0.0179, 0.0321, 0.0200, 0.0247, 0.0232, 0.0185], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 17:21:18,794 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52151.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:21:38,017 INFO [train.py:873] (3/4) Epoch 7, batch 6800, loss[loss=0.1598, simple_loss=0.1811, pruned_loss=0.0692, over 12726.00 frames. ], tot_loss[loss=0.1571, simple_loss=0.1771, pruned_loss=0.06853, over 1946349.46 frames. ], batch size: 100, lr: 1.10e-02, grad_scale: 16.0 2022-12-07 17:21:55,470 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52193.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:22:00,769 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52199.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:22:01,389 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.89 vs. limit=5.0 2022-12-07 17:22:22,601 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9825, 1.6677, 3.7606, 3.5433, 3.6922, 3.8739, 3.1485, 3.8312], device='cuda:3'), covar=tensor([0.1131, 0.1182, 0.0083, 0.0177, 0.0139, 0.0093, 0.0179, 0.0107], device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0152, 0.0112, 0.0154, 0.0127, 0.0126, 0.0103, 0.0109], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-07 17:22:26,309 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.57 vs. limit=5.0 2022-12-07 17:22:28,640 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4496, 2.3034, 3.1704, 2.4139, 3.1468, 3.1004, 3.0763, 2.6280], device='cuda:3'), covar=tensor([0.0642, 0.3118, 0.1074, 0.2406, 0.0941, 0.0872, 0.1380, 0.2137], device='cuda:3'), in_proj_covar=tensor([0.0316, 0.0333, 0.0386, 0.0311, 0.0374, 0.0304, 0.0358, 0.0333], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 17:22:32,532 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.294e+02 2.590e+02 3.090e+02 3.890e+02 7.235e+02, threshold=6.180e+02, percent-clipped=2.0 2022-12-07 17:22:41,113 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0684, 1.8789, 2.0203, 2.1022, 2.0717, 1.9924, 2.1668, 1.7540], device='cuda:3'), covar=tensor([0.0808, 0.1311, 0.0642, 0.0754, 0.0810, 0.0720, 0.0825, 0.0762], device='cuda:3'), in_proj_covar=tensor([0.0154, 0.0239, 0.0166, 0.0157, 0.0160, 0.0132, 0.0244, 0.0150], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 17:22:49,413 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52254.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:23:03,176 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.6301, 1.5661, 2.9093, 1.5347, 2.9783, 2.8522, 2.0135, 3.0316], device='cuda:3'), covar=tensor([0.0242, 0.2330, 0.0362, 0.1854, 0.0289, 0.0432, 0.0965, 0.0212], device='cuda:3'), in_proj_covar=tensor([0.0162, 0.0159, 0.0149, 0.0170, 0.0164, 0.0162, 0.0130, 0.0130], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 17:23:05,763 INFO [train.py:873] (3/4) Epoch 7, batch 6900, loss[loss=0.1711, simple_loss=0.186, pruned_loss=0.07812, over 14403.00 frames. ], tot_loss[loss=0.1564, simple_loss=0.177, pruned_loss=0.06796, over 1999526.78 frames. ], batch size: 53, lr: 1.10e-02, grad_scale: 8.0 2022-12-07 17:23:12,937 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9242, 1.5599, 1.8407, 2.0035, 1.4739, 1.7463, 1.9113, 1.9536], device='cuda:3'), covar=tensor([0.0050, 0.0101, 0.0046, 0.0030, 0.0093, 0.0127, 0.0058, 0.0044], device='cuda:3'), in_proj_covar=tensor([0.0234, 0.0219, 0.0327, 0.0272, 0.0216, 0.0264, 0.0230, 0.0254], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:3') 2022-12-07 17:23:40,120 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52312.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:24:00,341 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.245e+02 2.451e+02 3.077e+02 3.642e+02 5.911e+02, threshold=6.154e+02, percent-clipped=0.0 2022-12-07 17:24:33,253 INFO [train.py:873] (3/4) Epoch 7, batch 7000, loss[loss=0.1215, simple_loss=0.1576, pruned_loss=0.04268, over 14045.00 frames. ], tot_loss[loss=0.1575, simple_loss=0.1775, pruned_loss=0.06877, over 1988389.30 frames. ], batch size: 19, lr: 1.09e-02, grad_scale: 8.0 2022-12-07 17:24:33,459 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52373.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:24:38,072 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.6086, 1.6474, 4.5246, 1.8250, 4.2313, 4.6131, 4.1134, 4.9222], device='cuda:3'), covar=tensor([0.0169, 0.2799, 0.0323, 0.2263, 0.0287, 0.0272, 0.0321, 0.0113], device='cuda:3'), in_proj_covar=tensor([0.0160, 0.0158, 0.0146, 0.0167, 0.0162, 0.0159, 0.0128, 0.0129], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:3') 2022-12-07 17:24:53,464 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52396.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:25:00,140 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52403.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:25:27,918 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.282e+02 2.412e+02 3.030e+02 3.769e+02 8.602e+02, threshold=6.059e+02, percent-clipped=2.0 2022-12-07 17:25:41,995 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52451.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:26:01,216 INFO [train.py:873] (3/4) Epoch 7, batch 7100, loss[loss=0.1671, simple_loss=0.1809, pruned_loss=0.07669, over 14217.00 frames. ], tot_loss[loss=0.1559, simple_loss=0.1761, pruned_loss=0.06781, over 1932467.49 frames. ], batch size: 35, lr: 1.09e-02, grad_scale: 8.0 2022-12-07 17:26:56,578 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.032e+02 2.463e+02 3.010e+02 3.918e+02 9.338e+02, threshold=6.019e+02, percent-clipped=5.0 2022-12-07 17:26:58,398 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5694, 2.4494, 2.1951, 2.2342, 2.5166, 2.4879, 2.5904, 2.5215], device='cuda:3'), covar=tensor([0.1098, 0.0786, 0.2373, 0.3118, 0.1010, 0.1089, 0.1270, 0.0981], device='cuda:3'), in_proj_covar=tensor([0.0327, 0.0234, 0.0392, 0.0489, 0.0282, 0.0359, 0.0355, 0.0308], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 17:27:08,062 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52549.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:27:13,350 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52555.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 17:27:29,109 INFO [train.py:873] (3/4) Epoch 7, batch 7200, loss[loss=0.1635, simple_loss=0.172, pruned_loss=0.07756, over 4936.00 frames. ], tot_loss[loss=0.1563, simple_loss=0.1763, pruned_loss=0.06814, over 1982842.37 frames. ], batch size: 100, lr: 1.09e-02, grad_scale: 8.0 2022-12-07 17:27:50,835 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0798, 3.5124, 2.7081, 4.2806, 4.0683, 4.0518, 3.4397, 2.9227], device='cuda:3'), covar=tensor([0.0624, 0.1458, 0.4308, 0.0337, 0.0615, 0.1126, 0.1217, 0.3870], device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0299, 0.0281, 0.0212, 0.0273, 0.0268, 0.0252, 0.0270], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 17:28:05,263 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8362, 3.5453, 3.4684, 3.7869, 3.6205, 3.8145, 3.8461, 3.1428], device='cuda:3'), covar=tensor([0.0350, 0.0901, 0.0376, 0.0433, 0.0747, 0.0294, 0.0497, 0.0568], device='cuda:3'), in_proj_covar=tensor([0.0157, 0.0244, 0.0168, 0.0159, 0.0163, 0.0133, 0.0246, 0.0153], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-07 17:28:07,095 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52616.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 17:28:24,357 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.010e+02 2.308e+02 2.919e+02 3.605e+02 5.197e+02, threshold=5.838e+02, percent-clipped=0.0 2022-12-07 17:28:29,701 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2022-12-07 17:28:52,720 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52668.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:28:55,358 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8861, 1.5245, 2.0957, 1.7437, 1.9184, 1.4815, 1.7561, 1.7093], device='cuda:3'), covar=tensor([0.1457, 0.2383, 0.0234, 0.1491, 0.0529, 0.1339, 0.0725, 0.0658], device='cuda:3'), in_proj_covar=tensor([0.0227, 0.0236, 0.0178, 0.0319, 0.0194, 0.0243, 0.0225, 0.0186], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 17:28:56,838 INFO [train.py:873] (3/4) Epoch 7, batch 7300, loss[loss=0.1719, simple_loss=0.1684, pruned_loss=0.08775, over 3918.00 frames. ], tot_loss[loss=0.1543, simple_loss=0.1745, pruned_loss=0.06701, over 1963085.78 frames. ], batch size: 100, lr: 1.09e-02, grad_scale: 8.0 2022-12-07 17:29:16,721 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52696.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:29:50,134 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 2022-12-07 17:29:51,339 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.524e+02 2.341e+02 3.110e+02 4.071e+02 1.233e+03, threshold=6.219e+02, percent-clipped=5.0 2022-12-07 17:29:58,705 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52744.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:30:05,540 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52752.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:30:07,443 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.28 vs. limit=2.0 2022-12-07 17:30:24,392 INFO [train.py:873] (3/4) Epoch 7, batch 7400, loss[loss=0.1429, simple_loss=0.1731, pruned_loss=0.05641, over 14281.00 frames. ], tot_loss[loss=0.1555, simple_loss=0.1756, pruned_loss=0.06775, over 1935965.53 frames. ], batch size: 46, lr: 1.09e-02, grad_scale: 8.0 2022-12-07 17:30:38,545 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=12.46 vs. limit=5.0 2022-12-07 17:31:00,179 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52813.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:31:19,607 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.081e+01 2.208e+02 2.890e+02 3.954e+02 8.452e+02, threshold=5.779e+02, percent-clipped=3.0 2022-12-07 17:31:31,432 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52849.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:31:52,182 INFO [train.py:873] (3/4) Epoch 7, batch 7500, loss[loss=0.178, simple_loss=0.1744, pruned_loss=0.09082, over 3921.00 frames. ], tot_loss[loss=0.1556, simple_loss=0.1756, pruned_loss=0.06785, over 1925182.42 frames. ], batch size: 100, lr: 1.09e-02, grad_scale: 8.0 2022-12-07 17:32:00,741 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3424, 3.0393, 2.9219, 2.1842, 2.7694, 3.0213, 3.3985, 2.6979], device='cuda:3'), covar=tensor([0.0699, 0.1922, 0.1283, 0.2121, 0.1152, 0.0606, 0.0607, 0.1595], device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0194, 0.0125, 0.0128, 0.0118, 0.0121, 0.0101, 0.0133], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0007, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:3') 2022-12-07 17:32:13,109 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52897.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:32:24,641 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52911.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 17:33:18,935 INFO [train.py:873] (3/4) Epoch 8, batch 0, loss[loss=0.17, simple_loss=0.1872, pruned_loss=0.07641, over 14222.00 frames. ], tot_loss[loss=0.17, simple_loss=0.1872, pruned_loss=0.07641, over 14222.00 frames. ], batch size: 25, lr: 1.03e-02, grad_scale: 8.0 2022-12-07 17:33:18,935 INFO [train.py:896] (3/4) Computing validation loss 2022-12-07 17:33:26,264 INFO [train.py:905] (3/4) Epoch 8, validation: loss=0.1282, simple_loss=0.1716, pruned_loss=0.04242, over 857387.00 frames. 2022-12-07 17:33:26,264 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-07 17:33:27,146 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 6.973e+01 1.673e+02 2.697e+02 3.774e+02 7.847e+02, threshold=5.395e+02, percent-clipped=3.0 2022-12-07 17:33:36,527 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5048, 3.4417, 3.6805, 3.2945, 3.5659, 3.4567, 1.4335, 3.3337], device='cuda:3'), covar=tensor([0.0256, 0.0279, 0.0329, 0.0470, 0.0294, 0.0457, 0.2874, 0.0262], device='cuda:3'), in_proj_covar=tensor([0.0138, 0.0146, 0.0128, 0.0124, 0.0179, 0.0119, 0.0151, 0.0169], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 17:33:56,355 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52968.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:34:02,649 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2022-12-07 17:34:38,738 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53016.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:34:55,992 INFO [train.py:873] (3/4) Epoch 8, batch 100, loss[loss=0.2016, simple_loss=0.2038, pruned_loss=0.09969, over 8639.00 frames. ], tot_loss[loss=0.1529, simple_loss=0.1756, pruned_loss=0.06513, over 912299.98 frames. ], batch size: 100, lr: 1.02e-02, grad_scale: 8.0 2022-12-07 17:34:56,727 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.667e+02 2.496e+02 3.096e+02 4.127e+02 8.461e+02, threshold=6.192e+02, percent-clipped=10.0 2022-12-07 17:35:06,694 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9257, 3.6345, 3.5457, 3.9615, 3.7207, 3.4660, 4.0020, 3.3611], device='cuda:3'), covar=tensor([0.0487, 0.1020, 0.0430, 0.0414, 0.0766, 0.1211, 0.0582, 0.0576], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0240, 0.0164, 0.0155, 0.0158, 0.0128, 0.0243, 0.0149], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 17:35:13,535 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53056.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:35:59,522 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53108.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:36:02,767 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.58 vs. limit=2.0 2022-12-07 17:36:07,753 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53117.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:36:09,806 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.36 vs. limit=2.0 2022-12-07 17:36:23,090 INFO [train.py:873] (3/4) Epoch 8, batch 200, loss[loss=0.1517, simple_loss=0.1785, pruned_loss=0.06241, over 14138.00 frames. ], tot_loss[loss=0.1534, simple_loss=0.1748, pruned_loss=0.06604, over 1300766.94 frames. ], batch size: 84, lr: 1.02e-02, grad_scale: 8.0 2022-12-07 17:36:23,900 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.126e+02 2.183e+02 2.918e+02 3.622e+02 1.019e+03, threshold=5.836e+02, percent-clipped=2.0 2022-12-07 17:36:34,658 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2022-12-07 17:36:38,336 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2022-12-07 17:37:12,901 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4760, 2.1403, 2.2406, 1.3874, 2.0997, 2.2169, 2.4700, 2.0740], device='cuda:3'), covar=tensor([0.0682, 0.1200, 0.1006, 0.2227, 0.1099, 0.0596, 0.0614, 0.1447], device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0191, 0.0125, 0.0127, 0.0117, 0.0119, 0.0100, 0.0131], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0007, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:3') 2022-12-07 17:37:30,584 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53211.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 17:37:52,052 INFO [train.py:873] (3/4) Epoch 8, batch 300, loss[loss=0.1939, simple_loss=0.1723, pruned_loss=0.1078, over 1178.00 frames. ], tot_loss[loss=0.1525, simple_loss=0.174, pruned_loss=0.06551, over 1538197.72 frames. ], batch size: 100, lr: 1.02e-02, grad_scale: 8.0 2022-12-07 17:37:52,977 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.915e+01 2.253e+02 2.827e+02 3.791e+02 8.401e+02, threshold=5.654e+02, percent-clipped=6.0 2022-12-07 17:37:57,650 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53241.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:38:00,202 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2595, 3.5437, 4.3974, 3.0864, 2.6069, 3.4550, 1.7742, 3.5588], device='cuda:3'), covar=tensor([0.0889, 0.1126, 0.0750, 0.1647, 0.2452, 0.0813, 0.5095, 0.1170], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0089, 0.0080, 0.0090, 0.0112, 0.0074, 0.0131, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2022-12-07 17:38:12,813 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53259.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 17:38:23,942 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.29 vs. limit=5.0 2022-12-07 17:38:50,067 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53302.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:38:51,773 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53304.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:39:02,117 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53315.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:39:18,814 INFO [train.py:873] (3/4) Epoch 8, batch 400, loss[loss=0.1596, simple_loss=0.1527, pruned_loss=0.08323, over 2601.00 frames. ], tot_loss[loss=0.1535, simple_loss=0.1744, pruned_loss=0.06624, over 1677719.71 frames. ], batch size: 100, lr: 1.02e-02, grad_scale: 8.0 2022-12-07 17:39:19,679 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.217e+02 2.324e+02 2.854e+02 3.563e+02 7.977e+02, threshold=5.709e+02, percent-clipped=8.0 2022-12-07 17:39:43,838 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6061, 2.3426, 3.2806, 2.5281, 3.3946, 3.3482, 3.2168, 2.7881], device='cuda:3'), covar=tensor([0.0590, 0.3036, 0.0934, 0.1993, 0.0789, 0.0754, 0.1427, 0.2125], device='cuda:3'), in_proj_covar=tensor([0.0316, 0.0326, 0.0385, 0.0312, 0.0372, 0.0303, 0.0355, 0.0330], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 17:39:45,416 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53365.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:39:47,056 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53367.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:39:54,829 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53376.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:40:23,185 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53408.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:40:26,515 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53412.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:40:40,776 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53428.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 17:40:47,041 INFO [train.py:873] (3/4) Epoch 8, batch 500, loss[loss=0.1269, simple_loss=0.1593, pruned_loss=0.04727, over 14651.00 frames. ], tot_loss[loss=0.1545, simple_loss=0.175, pruned_loss=0.06704, over 1821030.71 frames. ], batch size: 33, lr: 1.02e-02, grad_scale: 4.0 2022-12-07 17:40:48,831 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.378e+02 2.493e+02 3.208e+02 4.073e+02 9.438e+02, threshold=6.416e+02, percent-clipped=6.0 2022-12-07 17:40:49,065 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7521, 3.4634, 2.9950, 2.3777, 3.1321, 3.3962, 3.6603, 2.6573], device='cuda:3'), covar=tensor([0.0558, 0.2524, 0.1333, 0.2443, 0.0963, 0.0679, 0.1016, 0.2054], device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0187, 0.0124, 0.0124, 0.0116, 0.0120, 0.0099, 0.0131], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:3') 2022-12-07 17:41:05,155 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53456.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:41:38,650 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2022-12-07 17:42:13,825 INFO [train.py:873] (3/4) Epoch 8, batch 600, loss[loss=0.1409, simple_loss=0.1688, pruned_loss=0.05648, over 14306.00 frames. ], tot_loss[loss=0.1529, simple_loss=0.1741, pruned_loss=0.06586, over 1928346.29 frames. ], batch size: 46, lr: 1.02e-02, grad_scale: 4.0 2022-12-07 17:42:15,675 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 6.499e+01 2.137e+02 2.538e+02 3.360e+02 7.193e+02, threshold=5.075e+02, percent-clipped=1.0 2022-12-07 17:42:52,698 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1648, 3.2568, 4.0850, 2.9570, 2.5339, 3.3846, 1.6905, 3.6204], device='cuda:3'), covar=tensor([0.1261, 0.0999, 0.0581, 0.1635, 0.2199, 0.0798, 0.3987, 0.0770], device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0084, 0.0077, 0.0086, 0.0107, 0.0072, 0.0125, 0.0076], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2022-12-07 17:43:08,725 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53597.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:43:42,309 INFO [train.py:873] (3/4) Epoch 8, batch 700, loss[loss=0.2158, simple_loss=0.2062, pruned_loss=0.1127, over 8620.00 frames. ], tot_loss[loss=0.1531, simple_loss=0.1746, pruned_loss=0.06581, over 1981803.22 frames. ], batch size: 100, lr: 1.02e-02, grad_scale: 4.0 2022-12-07 17:43:44,307 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.331e+02 2.400e+02 3.018e+02 4.076e+02 1.039e+03, threshold=6.035e+02, percent-clipped=12.0 2022-12-07 17:44:04,167 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53660.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:44:08,133 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.02 vs. limit=2.0 2022-12-07 17:44:08,407 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2042, 1.8601, 4.4981, 4.1890, 4.2057, 4.6780, 4.2402, 4.6312], device='cuda:3'), covar=tensor([0.1304, 0.1321, 0.0100, 0.0184, 0.0153, 0.0090, 0.0126, 0.0111], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0157, 0.0116, 0.0161, 0.0133, 0.0130, 0.0108, 0.0114], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 17:44:13,806 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53671.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:44:49,921 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53712.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:44:59,978 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53723.0, num_to_drop=1, layers_to_drop={3} 2022-12-07 17:45:08,675 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0145, 1.3924, 3.9570, 3.6894, 3.8452, 4.0372, 3.2158, 4.0608], device='cuda:3'), covar=tensor([0.1281, 0.1486, 0.0090, 0.0179, 0.0130, 0.0083, 0.0238, 0.0101], device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0156, 0.0115, 0.0159, 0.0131, 0.0128, 0.0106, 0.0113], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 17:45:10,197 INFO [train.py:873] (3/4) Epoch 8, batch 800, loss[loss=0.1558, simple_loss=0.1789, pruned_loss=0.06633, over 14422.00 frames. ], tot_loss[loss=0.1531, simple_loss=0.1741, pruned_loss=0.06607, over 1913967.60 frames. ], batch size: 51, lr: 1.02e-02, grad_scale: 8.0 2022-12-07 17:45:11,876 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.469e+01 2.335e+02 2.764e+02 3.237e+02 6.593e+02, threshold=5.529e+02, percent-clipped=1.0 2022-12-07 17:45:32,806 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53760.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:45:36,611 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.04 vs. limit=2.0 2022-12-07 17:46:05,263 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.85 vs. limit=5.0 2022-12-07 17:46:11,821 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.7820, 4.6405, 4.4098, 4.8545, 4.3608, 4.0045, 4.8515, 4.7729], device='cuda:3'), covar=tensor([0.0571, 0.0499, 0.0570, 0.0520, 0.0530, 0.0506, 0.0505, 0.0608], device='cuda:3'), in_proj_covar=tensor([0.0117, 0.0108, 0.0119, 0.0124, 0.0124, 0.0095, 0.0136, 0.0118], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 17:46:38,790 INFO [train.py:873] (3/4) Epoch 8, batch 900, loss[loss=0.1823, simple_loss=0.1886, pruned_loss=0.08796, over 8611.00 frames. ], tot_loss[loss=0.1535, simple_loss=0.1741, pruned_loss=0.06649, over 1893274.11 frames. ], batch size: 100, lr: 1.02e-02, grad_scale: 8.0 2022-12-07 17:46:40,851 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.375e+02 2.164e+02 2.734e+02 3.492e+02 7.190e+02, threshold=5.467e+02, percent-clipped=1.0 2022-12-07 17:47:32,870 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53897.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:48:00,632 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9411, 1.4949, 3.9598, 3.7549, 3.7037, 4.1016, 3.6137, 4.1156], device='cuda:3'), covar=tensor([0.1421, 0.1597, 0.0161, 0.0264, 0.0223, 0.0153, 0.0190, 0.0156], device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0156, 0.0115, 0.0160, 0.0131, 0.0128, 0.0106, 0.0112], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 17:48:05,625 INFO [train.py:873] (3/4) Epoch 8, batch 1000, loss[loss=0.1463, simple_loss=0.159, pruned_loss=0.06675, over 5958.00 frames. ], tot_loss[loss=0.1538, simple_loss=0.1746, pruned_loss=0.06647, over 1982278.48 frames. ], batch size: 100, lr: 1.02e-02, grad_scale: 8.0 2022-12-07 17:48:07,331 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.511e+01 2.177e+02 2.724e+02 3.766e+02 8.147e+02, threshold=5.449e+02, percent-clipped=5.0 2022-12-07 17:48:12,642 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.6611, 4.5110, 4.2794, 4.7442, 4.2518, 3.9664, 4.7532, 4.6259], device='cuda:3'), covar=tensor([0.0715, 0.0550, 0.0755, 0.0679, 0.0646, 0.0578, 0.0620, 0.0835], device='cuda:3'), in_proj_covar=tensor([0.0116, 0.0108, 0.0120, 0.0125, 0.0124, 0.0095, 0.0136, 0.0118], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 17:48:14,333 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53945.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:48:27,831 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53960.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:48:37,313 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53971.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:49:09,276 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=54008.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:49:12,247 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.46 vs. limit=5.0 2022-12-07 17:49:18,703 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=54019.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:49:21,349 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0481, 2.8511, 2.1289, 3.1928, 2.9250, 3.0075, 2.6933, 2.3073], device='cuda:3'), covar=tensor([0.0832, 0.1315, 0.3645, 0.0448, 0.0881, 0.0939, 0.1165, 0.3654], device='cuda:3'), in_proj_covar=tensor([0.0249, 0.0299, 0.0280, 0.0213, 0.0280, 0.0276, 0.0251, 0.0273], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 17:49:22,118 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=54023.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 17:49:22,412 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.36 vs. limit=5.0 2022-12-07 17:49:32,903 INFO [train.py:873] (3/4) Epoch 8, batch 1100, loss[loss=0.16, simple_loss=0.1828, pruned_loss=0.06862, over 14152.00 frames. ], tot_loss[loss=0.1531, simple_loss=0.1742, pruned_loss=0.06603, over 1982941.94 frames. ], batch size: 84, lr: 1.01e-02, grad_scale: 8.0 2022-12-07 17:49:34,932 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.169e+02 2.125e+02 2.772e+02 3.608e+02 1.248e+03, threshold=5.543e+02, percent-clipped=5.0 2022-12-07 17:50:04,032 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=54071.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 17:50:59,598 INFO [train.py:873] (3/4) Epoch 8, batch 1200, loss[loss=0.1735, simple_loss=0.1602, pruned_loss=0.09345, over 2627.00 frames. ], tot_loss[loss=0.1528, simple_loss=0.1743, pruned_loss=0.06571, over 1989549.98 frames. ], batch size: 100, lr: 1.01e-02, grad_scale: 8.0 2022-12-07 17:51:02,202 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.095e+02 2.508e+02 3.088e+02 3.577e+02 7.086e+02, threshold=6.177e+02, percent-clipped=4.0 2022-12-07 17:51:37,498 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6078, 1.2293, 2.0966, 1.9240, 1.9597, 2.0462, 1.6648, 2.0854], device='cuda:3'), covar=tensor([0.0444, 0.0860, 0.0091, 0.0256, 0.0281, 0.0135, 0.0333, 0.0151], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0156, 0.0115, 0.0159, 0.0132, 0.0128, 0.0106, 0.0113], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 17:52:09,019 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6096, 1.2688, 2.0810, 1.9150, 1.9517, 2.0537, 1.5857, 2.0616], device='cuda:3'), covar=tensor([0.0505, 0.0884, 0.0113, 0.0266, 0.0297, 0.0136, 0.0363, 0.0179], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0155, 0.0115, 0.0158, 0.0132, 0.0128, 0.0106, 0.0112], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 17:52:18,324 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.0111, 1.0347, 0.8901, 0.9758, 1.2532, 0.6677, 1.0239, 1.2007], device='cuda:3'), covar=tensor([0.0745, 0.1048, 0.0779, 0.0530, 0.0516, 0.0658, 0.0804, 0.0410], device='cuda:3'), in_proj_covar=tensor([0.0022, 0.0022, 0.0023, 0.0021, 0.0022, 0.0032, 0.0022, 0.0022], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2022-12-07 17:52:27,389 INFO [train.py:873] (3/4) Epoch 8, batch 1300, loss[loss=0.142, simple_loss=0.1745, pruned_loss=0.0548, over 14306.00 frames. ], tot_loss[loss=0.152, simple_loss=0.1737, pruned_loss=0.0652, over 2011339.35 frames. ], batch size: 39, lr: 1.01e-02, grad_scale: 8.0 2022-12-07 17:52:30,351 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.247e+02 2.196e+02 2.858e+02 3.785e+02 8.723e+02, threshold=5.716e+02, percent-clipped=2.0 2022-12-07 17:52:36,149 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2022-12-07 17:52:41,382 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.72 vs. limit=2.0 2022-12-07 17:52:47,144 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2022-12-07 17:53:39,609 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4657, 2.3808, 1.8709, 2.4693, 2.2069, 2.3612, 2.1380, 2.0200], device='cuda:3'), covar=tensor([0.0583, 0.0909, 0.2245, 0.0380, 0.0950, 0.0632, 0.1166, 0.1474], device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0298, 0.0277, 0.0215, 0.0276, 0.0273, 0.0249, 0.0269], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 17:53:54,867 INFO [train.py:873] (3/4) Epoch 8, batch 1400, loss[loss=0.1767, simple_loss=0.1555, pruned_loss=0.09895, over 1219.00 frames. ], tot_loss[loss=0.1529, simple_loss=0.1743, pruned_loss=0.06569, over 2012027.17 frames. ], batch size: 100, lr: 1.01e-02, grad_scale: 8.0 2022-12-07 17:53:57,393 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.247e+02 2.292e+02 2.799e+02 3.377e+02 6.597e+02, threshold=5.599e+02, percent-clipped=2.0 2022-12-07 17:54:29,952 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8526, 1.2166, 3.8960, 1.6983, 3.7605, 3.8937, 3.0646, 4.2092], device='cuda:3'), covar=tensor([0.0348, 0.4662, 0.0578, 0.3067, 0.0568, 0.0544, 0.0866, 0.0286], device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0159, 0.0151, 0.0168, 0.0163, 0.0165, 0.0133, 0.0133], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 17:54:43,779 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5820, 2.1171, 2.4166, 1.3020, 2.2178, 2.5209, 2.6222, 2.0804], device='cuda:3'), covar=tensor([0.0957, 0.2527, 0.1366, 0.3420, 0.1300, 0.0901, 0.0801, 0.2352], device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0187, 0.0125, 0.0125, 0.0117, 0.0121, 0.0100, 0.0133], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:3') 2022-12-07 17:55:22,065 INFO [train.py:873] (3/4) Epoch 8, batch 1500, loss[loss=0.1316, simple_loss=0.1309, pruned_loss=0.06613, over 2619.00 frames. ], tot_loss[loss=0.1533, simple_loss=0.1741, pruned_loss=0.06623, over 1974747.67 frames. ], batch size: 100, lr: 1.01e-02, grad_scale: 8.0 2022-12-07 17:55:25,014 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.978e+01 2.195e+02 2.660e+02 3.596e+02 7.442e+02, threshold=5.321e+02, percent-clipped=3.0 2022-12-07 17:55:33,438 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.94 vs. limit=5.0 2022-12-07 17:56:27,054 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.83 vs. limit=2.0 2022-12-07 17:56:31,812 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2022-12-07 17:56:49,480 INFO [train.py:873] (3/4) Epoch 8, batch 1600, loss[loss=0.1667, simple_loss=0.1757, pruned_loss=0.07882, over 8618.00 frames. ], tot_loss[loss=0.1532, simple_loss=0.1745, pruned_loss=0.06597, over 1999045.44 frames. ], batch size: 100, lr: 1.01e-02, grad_scale: 8.0 2022-12-07 17:56:51,887 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.201e+02 2.581e+02 3.079e+02 3.798e+02 6.375e+02, threshold=6.158e+02, percent-clipped=8.0 2022-12-07 17:57:12,794 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.3094, 1.9720, 4.9073, 2.7495, 4.4313, 5.0272, 4.8107, 5.6728], device='cuda:3'), covar=tensor([0.0151, 0.2574, 0.0315, 0.1727, 0.0353, 0.0391, 0.0265, 0.0094], device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0160, 0.0151, 0.0169, 0.0162, 0.0165, 0.0134, 0.0135], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 17:57:40,206 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2522, 1.8412, 4.9847, 4.6434, 4.3795, 5.1574, 4.8018, 5.1198], device='cuda:3'), covar=tensor([0.1179, 0.1294, 0.0066, 0.0125, 0.0140, 0.0059, 0.0079, 0.0087], device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0153, 0.0113, 0.0156, 0.0131, 0.0126, 0.0107, 0.0111], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 17:58:17,001 INFO [train.py:873] (3/4) Epoch 8, batch 1700, loss[loss=0.1387, simple_loss=0.1646, pruned_loss=0.05645, over 14522.00 frames. ], tot_loss[loss=0.151, simple_loss=0.173, pruned_loss=0.06452, over 1975506.81 frames. ], batch size: 43, lr: 1.01e-02, grad_scale: 8.0 2022-12-07 17:58:19,791 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.211e+02 2.375e+02 2.929e+02 3.612e+02 8.453e+02, threshold=5.858e+02, percent-clipped=6.0 2022-12-07 17:58:42,697 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=54664.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 17:59:36,605 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=54725.0, num_to_drop=1, layers_to_drop={3} 2022-12-07 17:59:44,436 INFO [train.py:873] (3/4) Epoch 8, batch 1800, loss[loss=0.1437, simple_loss=0.1312, pruned_loss=0.0781, over 2631.00 frames. ], tot_loss[loss=0.152, simple_loss=0.1735, pruned_loss=0.06522, over 1971332.54 frames. ], batch size: 100, lr: 1.01e-02, grad_scale: 8.0 2022-12-07 17:59:46,926 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.019e+02 2.304e+02 2.792e+02 3.452e+02 5.919e+02, threshold=5.584e+02, percent-clipped=1.0 2022-12-07 18:01:11,086 INFO [train.py:873] (3/4) Epoch 8, batch 1900, loss[loss=0.1367, simple_loss=0.1636, pruned_loss=0.05494, over 14277.00 frames. ], tot_loss[loss=0.1541, simple_loss=0.1748, pruned_loss=0.06667, over 1967026.20 frames. ], batch size: 63, lr: 1.01e-02, grad_scale: 8.0 2022-12-07 18:01:12,420 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2022-12-07 18:01:13,578 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.876e+01 2.431e+02 3.063e+02 3.945e+02 8.424e+02, threshold=6.126e+02, percent-clipped=6.0 2022-12-07 18:01:14,013 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0524, 1.9717, 2.0692, 2.0912, 1.9797, 1.7206, 1.3060, 1.8258], device='cuda:3'), covar=tensor([0.0365, 0.0352, 0.0466, 0.0292, 0.0351, 0.0849, 0.1673, 0.0305], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0150, 0.0131, 0.0125, 0.0179, 0.0122, 0.0151, 0.0169], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 18:01:28,051 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.1948, 4.4421, 4.8840, 5.0439, 4.7561, 4.3898, 4.9933, 3.9590], device='cuda:3'), covar=tensor([0.0773, 0.1659, 0.0571, 0.0893, 0.1212, 0.0547, 0.1123, 0.1036], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0243, 0.0167, 0.0159, 0.0167, 0.0133, 0.0249, 0.0152], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-07 18:01:32,516 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2448, 2.9080, 2.8155, 2.1628, 2.7536, 2.9396, 3.3381, 2.5294], device='cuda:3'), covar=tensor([0.0794, 0.2016, 0.1422, 0.2404, 0.1099, 0.0819, 0.0952, 0.1866], device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0188, 0.0125, 0.0126, 0.0117, 0.0123, 0.0101, 0.0133], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0007, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:3') 2022-12-07 18:01:35,165 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=54862.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:01:49,907 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.3874, 5.1329, 4.7612, 5.4070, 4.9116, 4.9300, 5.4137, 5.3339], device='cuda:3'), covar=tensor([0.0711, 0.0688, 0.0770, 0.0613, 0.0617, 0.0409, 0.0676, 0.0722], device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0114, 0.0127, 0.0130, 0.0131, 0.0101, 0.0145, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 18:01:57,136 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3716, 3.4052, 4.2369, 2.7134, 2.7720, 3.2899, 2.0405, 3.4349], device='cuda:3'), covar=tensor([0.1742, 0.0810, 0.0378, 0.1655, 0.2123, 0.1405, 0.4160, 0.0960], device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0088, 0.0082, 0.0090, 0.0113, 0.0075, 0.0132, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2022-12-07 18:01:58,790 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.7739, 4.4240, 4.3000, 4.7466, 4.5135, 4.1649, 4.7611, 3.9258], device='cuda:3'), covar=tensor([0.0327, 0.0856, 0.0286, 0.0447, 0.0666, 0.0631, 0.0504, 0.0474], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0242, 0.0167, 0.0158, 0.0166, 0.0133, 0.0248, 0.0151], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-07 18:02:29,211 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=54923.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 18:02:39,004 INFO [train.py:873] (3/4) Epoch 8, batch 2000, loss[loss=0.151, simple_loss=0.14, pruned_loss=0.08099, over 1229.00 frames. ], tot_loss[loss=0.1545, simple_loss=0.1751, pruned_loss=0.06695, over 1943124.77 frames. ], batch size: 100, lr: 1.01e-02, grad_scale: 8.0 2022-12-07 18:02:41,554 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.941e+01 2.269e+02 2.935e+02 3.791e+02 7.498e+02, threshold=5.870e+02, percent-clipped=3.0 2022-12-07 18:02:47,144 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.12 vs. limit=2.0 2022-12-07 18:02:52,006 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6329, 3.3885, 3.1454, 3.2901, 3.5287, 3.5142, 3.6072, 3.5919], device='cuda:3'), covar=tensor([0.0842, 0.0787, 0.1929, 0.2670, 0.0815, 0.0841, 0.1012, 0.0839], device='cuda:3'), in_proj_covar=tensor([0.0332, 0.0243, 0.0400, 0.0510, 0.0296, 0.0374, 0.0363, 0.0314], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 18:03:43,222 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5579, 3.6319, 3.8620, 3.5365, 3.6690, 3.5650, 1.3551, 3.4504], device='cuda:3'), covar=tensor([0.0265, 0.0290, 0.0357, 0.0375, 0.0315, 0.0485, 0.3107, 0.0281], device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0151, 0.0134, 0.0127, 0.0182, 0.0125, 0.0154, 0.0172], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 18:03:55,618 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3050, 2.4387, 2.3788, 2.5570, 1.9921, 2.7627, 2.3563, 1.1251], device='cuda:3'), covar=tensor([0.2349, 0.1008, 0.1252, 0.0841, 0.1232, 0.0476, 0.1351, 0.3600], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0064, 0.0055, 0.0056, 0.0082, 0.0061, 0.0088, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2022-12-07 18:03:57,246 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55020.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 18:03:58,829 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([6.0107, 5.4100, 5.4218, 5.9238, 5.4916, 4.8337, 5.8637, 4.9387], device='cuda:3'), covar=tensor([0.0303, 0.1033, 0.0237, 0.0394, 0.0734, 0.0345, 0.0469, 0.0503], device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0248, 0.0169, 0.0161, 0.0168, 0.0135, 0.0253, 0.0154], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-07 18:04:02,344 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0488, 1.9149, 1.9351, 2.0633, 2.0061, 1.9542, 2.1422, 1.7321], device='cuda:3'), covar=tensor([0.0866, 0.1213, 0.0698, 0.0764, 0.0781, 0.0797, 0.0858, 0.0682], device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0248, 0.0169, 0.0161, 0.0168, 0.0135, 0.0252, 0.0154], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-07 18:04:09,837 INFO [train.py:873] (3/4) Epoch 8, batch 2100, loss[loss=0.1631, simple_loss=0.1477, pruned_loss=0.08926, over 1255.00 frames. ], tot_loss[loss=0.1523, simple_loss=0.1736, pruned_loss=0.06551, over 1947611.58 frames. ], batch size: 100, lr: 1.01e-02, grad_scale: 8.0 2022-12-07 18:04:12,640 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.250e+02 2.381e+02 2.883e+02 3.565e+02 6.243e+02, threshold=5.765e+02, percent-clipped=1.0 2022-12-07 18:04:45,568 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4700, 4.1454, 3.9902, 4.4688, 4.2997, 3.9943, 4.4927, 3.7429], device='cuda:3'), covar=tensor([0.0460, 0.1114, 0.0416, 0.0477, 0.0782, 0.0780, 0.0595, 0.0577], device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0250, 0.0170, 0.0161, 0.0169, 0.0134, 0.0252, 0.0154], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-07 18:05:37,626 INFO [train.py:873] (3/4) Epoch 8, batch 2200, loss[loss=0.1759, simple_loss=0.1611, pruned_loss=0.09536, over 1245.00 frames. ], tot_loss[loss=0.153, simple_loss=0.1739, pruned_loss=0.06605, over 1907410.65 frames. ], batch size: 100, lr: 1.00e-02, grad_scale: 8.0 2022-12-07 18:05:39,867 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.445e+02 2.363e+02 3.011e+02 4.055e+02 7.336e+02, threshold=6.021e+02, percent-clipped=10.0 2022-12-07 18:05:44,858 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2022-12-07 18:05:48,459 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55148.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:05:52,369 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 2022-12-07 18:05:52,895 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.0530, 1.1987, 1.2718, 1.1470, 0.9864, 0.7217, 0.9740, 0.8455], device='cuda:3'), covar=tensor([0.0233, 0.0235, 0.0196, 0.0354, 0.0306, 0.0491, 0.0320, 0.0676], device='cuda:3'), in_proj_covar=tensor([0.0012, 0.0013, 0.0011, 0.0012, 0.0012, 0.0018, 0.0014, 0.0019], device='cuda:3'), out_proj_covar=tensor([8.2146e-05, 8.9846e-05, 8.0741e-05, 8.4616e-05, 8.3391e-05, 1.2263e-04, 1.0137e-04, 1.1871e-04], device='cuda:3') 2022-12-07 18:05:53,765 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1430, 1.6002, 1.6007, 1.7497, 1.5745, 1.8165, 1.3299, 1.1567], device='cuda:3'), covar=tensor([0.1741, 0.0743, 0.0387, 0.0305, 0.0953, 0.0432, 0.1588, 0.1617], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0064, 0.0055, 0.0056, 0.0082, 0.0061, 0.0087, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2022-12-07 18:06:43,453 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55209.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:06:44,986 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55211.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:06:51,022 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55218.0, num_to_drop=1, layers_to_drop={3} 2022-12-07 18:07:05,470 INFO [train.py:873] (3/4) Epoch 8, batch 2300, loss[loss=0.142, simple_loss=0.1659, pruned_loss=0.05908, over 14217.00 frames. ], tot_loss[loss=0.1524, simple_loss=0.1736, pruned_loss=0.06561, over 1874613.19 frames. ], batch size: 35, lr: 1.00e-02, grad_scale: 8.0 2022-12-07 18:07:08,046 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.599e+02 2.336e+02 2.949e+02 3.747e+02 7.724e+02, threshold=5.898e+02, percent-clipped=2.0 2022-12-07 18:07:39,138 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55272.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:08:21,841 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55320.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 18:08:34,796 INFO [train.py:873] (3/4) Epoch 8, batch 2400, loss[loss=0.1444, simple_loss=0.1745, pruned_loss=0.05717, over 14291.00 frames. ], tot_loss[loss=0.1532, simple_loss=0.1743, pruned_loss=0.06609, over 1900478.19 frames. ], batch size: 60, lr: 1.00e-02, grad_scale: 8.0 2022-12-07 18:08:36,533 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9272, 1.2670, 2.7915, 2.5557, 2.6894, 2.7794, 2.0640, 2.7863], device='cuda:3'), covar=tensor([0.0914, 0.1222, 0.0108, 0.0274, 0.0264, 0.0111, 0.0394, 0.0149], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0154, 0.0114, 0.0157, 0.0131, 0.0128, 0.0108, 0.0113], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 18:08:37,246 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.077e+02 2.285e+02 2.947e+02 4.114e+02 7.749e+02, threshold=5.894e+02, percent-clipped=7.0 2022-12-07 18:09:04,226 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=55368.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 18:09:25,736 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9222, 1.3216, 2.5039, 2.3600, 2.4624, 2.4971, 1.9205, 2.5350], device='cuda:3'), covar=tensor([0.0615, 0.0970, 0.0122, 0.0292, 0.0255, 0.0119, 0.0429, 0.0152], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0153, 0.0114, 0.0155, 0.0130, 0.0127, 0.0107, 0.0112], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 18:09:48,704 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0186, 3.0082, 2.9685, 3.0627, 3.0280, 2.4822, 1.2866, 2.8099], device='cuda:3'), covar=tensor([0.0449, 0.0533, 0.0736, 0.0539, 0.0509, 0.1262, 0.3811, 0.0432], device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0151, 0.0132, 0.0126, 0.0183, 0.0124, 0.0153, 0.0172], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 18:09:52,836 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 2022-12-07 18:10:03,415 INFO [train.py:873] (3/4) Epoch 8, batch 2500, loss[loss=0.137, simple_loss=0.1646, pruned_loss=0.05476, over 14266.00 frames. ], tot_loss[loss=0.152, simple_loss=0.1734, pruned_loss=0.06532, over 1906658.66 frames. ], batch size: 25, lr: 1.00e-02, grad_scale: 8.0 2022-12-07 18:10:05,936 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.292e+02 2.135e+02 2.789e+02 3.558e+02 8.555e+02, threshold=5.578e+02, percent-clipped=3.0 2022-12-07 18:10:22,211 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9203, 1.4205, 3.1277, 1.4513, 3.2226, 3.1207, 2.1076, 3.2621], device='cuda:3'), covar=tensor([0.0271, 0.2676, 0.0342, 0.2199, 0.0289, 0.0387, 0.0935, 0.0176], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0161, 0.0153, 0.0170, 0.0166, 0.0167, 0.0134, 0.0138], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 18:11:03,307 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55504.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:11:15,529 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55518.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 18:11:30,376 INFO [train.py:873] (3/4) Epoch 8, batch 2600, loss[loss=0.1606, simple_loss=0.1715, pruned_loss=0.07482, over 6030.00 frames. ], tot_loss[loss=0.1526, simple_loss=0.174, pruned_loss=0.06562, over 1965143.41 frames. ], batch size: 100, lr: 1.00e-02, grad_scale: 8.0 2022-12-07 18:11:32,918 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.429e+02 2.392e+02 2.780e+02 3.615e+02 5.645e+02, threshold=5.559e+02, percent-clipped=1.0 2022-12-07 18:11:57,683 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=55566.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:11:58,699 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55567.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:12:01,291 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1316, 1.7741, 4.6638, 4.2786, 4.1831, 4.7367, 4.4562, 4.7887], device='cuda:3'), covar=tensor([0.1253, 0.1344, 0.0079, 0.0157, 0.0133, 0.0091, 0.0112, 0.0093], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0155, 0.0116, 0.0158, 0.0132, 0.0130, 0.0109, 0.0114], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 18:12:16,134 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.5712, 5.4245, 5.0572, 5.7279, 5.2475, 4.9589, 5.7657, 5.5796], device='cuda:3'), covar=tensor([0.0666, 0.0559, 0.0601, 0.0451, 0.0598, 0.0376, 0.0524, 0.0668], device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0111, 0.0123, 0.0126, 0.0127, 0.0098, 0.0139, 0.0120], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 18:12:42,473 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.62 vs. limit=2.0 2022-12-07 18:12:57,535 INFO [train.py:873] (3/4) Epoch 8, batch 2700, loss[loss=0.1238, simple_loss=0.1538, pruned_loss=0.04688, over 14232.00 frames. ], tot_loss[loss=0.1525, simple_loss=0.1741, pruned_loss=0.0654, over 1950457.92 frames. ], batch size: 37, lr: 1.00e-02, grad_scale: 8.0 2022-12-07 18:13:00,129 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.995e+01 2.381e+02 3.010e+02 3.860e+02 1.123e+03, threshold=6.020e+02, percent-clipped=6.0 2022-12-07 18:13:08,252 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.0422, 1.1533, 1.0565, 0.9390, 1.2840, 0.6793, 1.2834, 1.1160], device='cuda:3'), covar=tensor([0.0704, 0.0592, 0.0538, 0.0761, 0.0622, 0.0867, 0.0369, 0.0656], device='cuda:3'), in_proj_covar=tensor([0.0021, 0.0021, 0.0022, 0.0021, 0.0022, 0.0031, 0.0022, 0.0022], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2022-12-07 18:13:32,995 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9941, 1.5955, 2.0121, 1.4030, 1.7349, 2.0511, 2.0010, 1.7573], device='cuda:3'), covar=tensor([0.0786, 0.1315, 0.1174, 0.2026, 0.1198, 0.0842, 0.0503, 0.1757], device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0190, 0.0127, 0.0126, 0.0118, 0.0123, 0.0104, 0.0134], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0007, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:3') 2022-12-07 18:14:27,409 INFO [train.py:873] (3/4) Epoch 8, batch 2800, loss[loss=0.1484, simple_loss=0.1717, pruned_loss=0.06253, over 14172.00 frames. ], tot_loss[loss=0.1511, simple_loss=0.1734, pruned_loss=0.06438, over 1925420.45 frames. ], batch size: 84, lr: 9.99e-03, grad_scale: 8.0 2022-12-07 18:14:30,868 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.032e+02 2.464e+02 2.891e+02 3.921e+02 7.148e+02, threshold=5.781e+02, percent-clipped=1.0 2022-12-07 18:14:38,337 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55748.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:14:46,194 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9962, 3.0449, 2.8304, 3.0338, 2.4483, 3.1806, 2.6673, 1.4967], device='cuda:3'), covar=tensor([0.2525, 0.0935, 0.1386, 0.0694, 0.1024, 0.0783, 0.1606, 0.3087], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0066, 0.0055, 0.0057, 0.0083, 0.0063, 0.0087, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2022-12-07 18:15:28,231 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55804.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:15:29,149 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8013, 2.1183, 3.8598, 2.6999, 3.6361, 1.9457, 2.8647, 3.6901], device='cuda:3'), covar=tensor([0.0441, 0.4426, 0.0309, 0.6481, 0.0558, 0.3780, 0.1198, 0.0353], device='cuda:3'), in_proj_covar=tensor([0.0228, 0.0238, 0.0184, 0.0314, 0.0207, 0.0241, 0.0226, 0.0188], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 18:15:30,862 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8899, 1.6037, 2.0669, 1.7145, 1.8886, 1.4640, 1.6456, 1.8137], device='cuda:3'), covar=tensor([0.1475, 0.2417, 0.0226, 0.1392, 0.0888, 0.1182, 0.1007, 0.0443], device='cuda:3'), in_proj_covar=tensor([0.0228, 0.0238, 0.0184, 0.0315, 0.0207, 0.0241, 0.0227, 0.0188], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 18:15:31,367 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.17 vs. limit=5.0 2022-12-07 18:15:32,948 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55809.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:15:42,946 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.2231, 1.6739, 1.6499, 1.7318, 1.5415, 1.8070, 1.4719, 1.0099], device='cuda:3'), covar=tensor([0.1746, 0.1047, 0.0646, 0.0461, 0.1234, 0.0687, 0.2100, 0.2950], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0065, 0.0055, 0.0056, 0.0083, 0.0063, 0.0087, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2022-12-07 18:15:53,058 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.2959, 2.5687, 4.1884, 4.3893, 4.4530, 2.5301, 4.3497, 3.3607], device='cuda:3'), covar=tensor([0.0232, 0.0563, 0.0823, 0.0243, 0.0150, 0.0869, 0.0204, 0.0555], device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0223, 0.0340, 0.0282, 0.0225, 0.0271, 0.0244, 0.0258], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 18:15:54,464 INFO [train.py:873] (3/4) Epoch 8, batch 2900, loss[loss=0.1204, simple_loss=0.1569, pruned_loss=0.04195, over 11281.00 frames. ], tot_loss[loss=0.1507, simple_loss=0.1729, pruned_loss=0.06427, over 1915541.66 frames. ], batch size: 14, lr: 9.99e-03, grad_scale: 8.0 2022-12-07 18:15:57,949 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.733e+01 2.419e+02 2.995e+02 3.748e+02 5.974e+02, threshold=5.989e+02, percent-clipped=2.0 2022-12-07 18:16:03,793 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.6355, 2.8331, 2.4552, 2.7913, 2.1661, 2.7758, 2.5743, 1.2039], device='cuda:3'), covar=tensor([0.2168, 0.0579, 0.1727, 0.0769, 0.1056, 0.0990, 0.1166, 0.3200], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0065, 0.0054, 0.0057, 0.0083, 0.0063, 0.0087, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2022-12-07 18:16:06,711 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55848.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:16:09,830 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=55852.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:16:22,839 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55867.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:16:59,473 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55909.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:17:04,334 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=55915.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:17:22,326 INFO [train.py:873] (3/4) Epoch 8, batch 3000, loss[loss=0.1398, simple_loss=0.1642, pruned_loss=0.05772, over 14406.00 frames. ], tot_loss[loss=0.1519, simple_loss=0.1735, pruned_loss=0.06516, over 1921104.95 frames. ], batch size: 53, lr: 9.98e-03, grad_scale: 8.0 2022-12-07 18:17:22,327 INFO [train.py:896] (3/4) Computing validation loss 2022-12-07 18:17:38,755 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.8326, 5.2304, 5.4984, 5.7264, 5.2334, 4.4694, 5.5642, 5.2332], device='cuda:3'), covar=tensor([0.0174, 0.0555, 0.0181, 0.0281, 0.0780, 0.0219, 0.0569, 0.0298], device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0247, 0.0168, 0.0159, 0.0166, 0.0133, 0.0249, 0.0151], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-07 18:17:42,270 INFO [train.py:905] (3/4) Epoch 8, validation: loss=0.1226, simple_loss=0.1659, pruned_loss=0.03968, over 857387.00 frames. 2022-12-07 18:17:42,270 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-07 18:17:45,688 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.182e+02 2.436e+02 3.032e+02 3.912e+02 1.128e+03, threshold=6.064e+02, percent-clipped=7.0 2022-12-07 18:18:49,470 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5960, 2.3168, 3.5582, 3.7003, 3.5256, 2.3478, 3.6335, 2.7783], device='cuda:3'), covar=tensor([0.0190, 0.0509, 0.0438, 0.0205, 0.0207, 0.0760, 0.0170, 0.0553], device='cuda:3'), in_proj_covar=tensor([0.0247, 0.0225, 0.0338, 0.0282, 0.0226, 0.0272, 0.0246, 0.0258], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 18:19:11,702 INFO [train.py:873] (3/4) Epoch 8, batch 3100, loss[loss=0.1072, simple_loss=0.1451, pruned_loss=0.03461, over 13997.00 frames. ], tot_loss[loss=0.1503, simple_loss=0.1724, pruned_loss=0.06412, over 1935693.63 frames. ], batch size: 19, lr: 9.97e-03, grad_scale: 8.0 2022-12-07 18:19:15,078 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.402e+01 2.241e+02 2.784e+02 3.482e+02 7.776e+02, threshold=5.568e+02, percent-clipped=3.0 2022-12-07 18:20:11,511 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56104.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:20:38,656 INFO [train.py:873] (3/4) Epoch 8, batch 3200, loss[loss=0.1491, simple_loss=0.1717, pruned_loss=0.06328, over 14428.00 frames. ], tot_loss[loss=0.1512, simple_loss=0.1734, pruned_loss=0.06447, over 2010089.69 frames. ], batch size: 73, lr: 9.96e-03, grad_scale: 8.0 2022-12-07 18:20:42,716 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.378e+02 2.292e+02 3.048e+02 3.641e+02 9.234e+02, threshold=6.095e+02, percent-clipped=4.0 2022-12-07 18:20:47,477 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56145.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:21:00,471 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5658, 3.6652, 3.9295, 3.5675, 3.7151, 3.6679, 1.3838, 3.5187], device='cuda:3'), covar=tensor([0.0295, 0.0286, 0.0341, 0.0455, 0.0309, 0.0405, 0.3305, 0.0253], device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0153, 0.0133, 0.0129, 0.0185, 0.0124, 0.0153, 0.0172], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 18:21:25,989 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8150, 1.8519, 3.1163, 2.2135, 2.8811, 1.7873, 2.3911, 2.8408], device='cuda:3'), covar=tensor([0.0787, 0.4412, 0.0545, 0.5707, 0.0645, 0.3514, 0.1191, 0.0628], device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0233, 0.0185, 0.0314, 0.0205, 0.0240, 0.0224, 0.0187], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 18:21:35,679 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.8469, 5.5532, 5.2963, 5.9958, 5.4191, 5.0484, 5.9511, 5.8817], device='cuda:3'), covar=tensor([0.0525, 0.0494, 0.0527, 0.0352, 0.0494, 0.0284, 0.0345, 0.0448], device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0115, 0.0125, 0.0131, 0.0130, 0.0102, 0.0143, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 18:21:40,191 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56204.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:21:42,153 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56206.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:22:08,270 INFO [train.py:873] (3/4) Epoch 8, batch 3300, loss[loss=0.1348, simple_loss=0.159, pruned_loss=0.05528, over 13977.00 frames. ], tot_loss[loss=0.1504, simple_loss=0.1726, pruned_loss=0.06407, over 2009268.41 frames. ], batch size: 19, lr: 9.95e-03, grad_scale: 8.0 2022-12-07 18:22:12,320 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.476e+02 2.398e+02 3.040e+02 3.643e+02 7.318e+02, threshold=6.080e+02, percent-clipped=1.0 2022-12-07 18:22:58,827 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8653, 4.6558, 4.2492, 4.3894, 4.4835, 4.7223, 4.8474, 4.7920], device='cuda:3'), covar=tensor([0.0731, 0.0577, 0.2535, 0.2960, 0.0760, 0.0878, 0.0816, 0.1026], device='cuda:3'), in_proj_covar=tensor([0.0335, 0.0244, 0.0405, 0.0511, 0.0289, 0.0373, 0.0371, 0.0326], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 18:23:06,703 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1480, 3.4603, 4.2826, 2.9579, 2.5247, 3.5000, 1.9685, 3.4575], device='cuda:3'), covar=tensor([0.1250, 0.0731, 0.0354, 0.1893, 0.2256, 0.0940, 0.3947, 0.1068], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0087, 0.0081, 0.0090, 0.0110, 0.0076, 0.0128, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2022-12-07 18:23:30,305 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1519, 1.8775, 2.2358, 2.3311, 1.9758, 1.9610, 2.3378, 2.1291], device='cuda:3'), covar=tensor([0.0126, 0.0244, 0.0113, 0.0081, 0.0188, 0.0293, 0.0137, 0.0115], device='cuda:3'), in_proj_covar=tensor([0.0247, 0.0224, 0.0343, 0.0283, 0.0226, 0.0274, 0.0249, 0.0260], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 18:23:34,243 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.73 vs. limit=2.0 2022-12-07 18:23:35,254 INFO [train.py:873] (3/4) Epoch 8, batch 3400, loss[loss=0.1641, simple_loss=0.1883, pruned_loss=0.06996, over 14673.00 frames. ], tot_loss[loss=0.1512, simple_loss=0.1726, pruned_loss=0.06492, over 1879171.86 frames. ], batch size: 23, lr: 9.94e-03, grad_scale: 8.0 2022-12-07 18:23:39,442 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.326e+02 2.408e+02 2.894e+02 3.622e+02 6.016e+02, threshold=5.789e+02, percent-clipped=0.0 2022-12-07 18:24:36,603 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=56404.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:24:43,701 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2022-12-07 18:24:46,401 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.07 vs. limit=2.0 2022-12-07 18:25:03,324 INFO [train.py:873] (3/4) Epoch 8, batch 3500, loss[loss=0.1357, simple_loss=0.1646, pruned_loss=0.0534, over 14649.00 frames. ], tot_loss[loss=0.1508, simple_loss=0.1726, pruned_loss=0.06448, over 1935947.52 frames. ], batch size: 23, lr: 9.93e-03, grad_scale: 8.0 2022-12-07 18:25:07,749 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.106e+02 2.219e+02 2.782e+02 3.549e+02 6.206e+02, threshold=5.564e+02, percent-clipped=1.0 2022-12-07 18:25:09,684 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0832, 3.2736, 2.9051, 2.9501, 2.5585, 3.5387, 2.9890, 1.4909], device='cuda:3'), covar=tensor([0.2678, 0.0748, 0.1807, 0.0998, 0.0923, 0.0430, 0.1327, 0.3197], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0066, 0.0055, 0.0057, 0.0083, 0.0064, 0.0088, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2022-12-07 18:25:18,099 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=56452.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:26:00,373 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0027, 3.7139, 3.5139, 3.6441, 3.8493, 3.9125, 3.9776, 3.9615], device='cuda:3'), covar=tensor([0.0713, 0.0620, 0.1975, 0.2596, 0.0644, 0.0796, 0.0907, 0.0769], device='cuda:3'), in_proj_covar=tensor([0.0330, 0.0239, 0.0396, 0.0500, 0.0282, 0.0369, 0.0360, 0.0319], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 18:26:01,192 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56501.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:26:01,614 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.93 vs. limit=2.0 2022-12-07 18:26:04,241 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=56504.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:26:07,424 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.91 vs. limit=2.0 2022-12-07 18:26:20,877 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56523.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 18:26:30,712 INFO [train.py:873] (3/4) Epoch 8, batch 3600, loss[loss=0.1673, simple_loss=0.1818, pruned_loss=0.0764, over 14129.00 frames. ], tot_loss[loss=0.1496, simple_loss=0.172, pruned_loss=0.06358, over 1957669.15 frames. ], batch size: 99, lr: 9.92e-03, grad_scale: 8.0 2022-12-07 18:26:35,120 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.906e+01 2.280e+02 2.872e+02 3.796e+02 9.491e+02, threshold=5.743e+02, percent-clipped=4.0 2022-12-07 18:26:46,148 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=56552.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:27:14,199 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56584.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 18:27:22,911 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2022-12-07 18:27:52,860 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8073, 0.6708, 0.6266, 0.6745, 0.6890, 0.3450, 0.4336, 0.5456], device='cuda:3'), covar=tensor([0.0114, 0.0103, 0.0078, 0.0091, 0.0124, 0.0316, 0.0216, 0.0293], device='cuda:3'), in_proj_covar=tensor([0.0012, 0.0014, 0.0012, 0.0012, 0.0012, 0.0019, 0.0015, 0.0019], device='cuda:3'), out_proj_covar=tensor([8.7724e-05, 9.4424e-05, 8.5044e-05, 8.8851e-05, 8.6407e-05, 1.2833e-04, 1.0818e-04, 1.2336e-04], device='cuda:3') 2022-12-07 18:27:59,552 INFO [train.py:873] (3/4) Epoch 8, batch 3700, loss[loss=0.1723, simple_loss=0.1859, pruned_loss=0.07936, over 14226.00 frames. ], tot_loss[loss=0.1503, simple_loss=0.1723, pruned_loss=0.06413, over 1952226.99 frames. ], batch size: 69, lr: 9.92e-03, grad_scale: 8.0 2022-12-07 18:28:03,912 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.177e+02 2.170e+02 2.783e+02 3.492e+02 8.002e+02, threshold=5.566e+02, percent-clipped=4.0 2022-12-07 18:28:35,142 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7497, 2.3893, 3.1557, 2.1622, 2.3286, 2.7024, 1.4651, 2.4992], device='cuda:3'), covar=tensor([0.1109, 0.1426, 0.0735, 0.2327, 0.2115, 0.0928, 0.4696, 0.1200], device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0086, 0.0080, 0.0089, 0.0109, 0.0075, 0.0127, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2022-12-07 18:29:25,510 INFO [train.py:873] (3/4) Epoch 8, batch 3800, loss[loss=0.1564, simple_loss=0.1777, pruned_loss=0.06756, over 14222.00 frames. ], tot_loss[loss=0.1505, simple_loss=0.1726, pruned_loss=0.06417, over 2025159.01 frames. ], batch size: 94, lr: 9.91e-03, grad_scale: 8.0 2022-12-07 18:29:29,584 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.270e+02 2.607e+02 3.160e+02 3.908e+02 8.359e+02, threshold=6.320e+02, percent-clipped=5.0 2022-12-07 18:30:07,987 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0289, 1.9956, 2.0861, 2.0844, 2.0296, 1.7455, 1.2991, 1.8500], device='cuda:3'), covar=tensor([0.0378, 0.0379, 0.0448, 0.0294, 0.0353, 0.0891, 0.1657, 0.0369], device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0152, 0.0131, 0.0127, 0.0183, 0.0123, 0.0151, 0.0172], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 18:30:14,777 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0734, 1.8799, 2.0221, 2.0776, 2.0437, 1.9334, 2.1935, 1.6948], device='cuda:3'), covar=tensor([0.0949, 0.1912, 0.0794, 0.0906, 0.1218, 0.0853, 0.1001, 0.0903], device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0246, 0.0167, 0.0161, 0.0164, 0.0130, 0.0244, 0.0149], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 18:30:24,865 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=56801.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:30:26,560 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56803.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:30:37,623 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5536, 4.4075, 4.9374, 4.1521, 4.6991, 5.0576, 1.7419, 4.3626], device='cuda:3'), covar=tensor([0.0223, 0.0320, 0.0329, 0.0340, 0.0283, 0.0132, 0.3139, 0.0237], device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0152, 0.0131, 0.0127, 0.0182, 0.0123, 0.0150, 0.0172], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 18:30:42,003 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2973, 1.9368, 2.3599, 2.4668, 2.2066, 1.8930, 2.5051, 2.1059], device='cuda:3'), covar=tensor([0.0195, 0.0408, 0.0217, 0.0166, 0.0227, 0.0518, 0.0170, 0.0306], device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0224, 0.0338, 0.0280, 0.0224, 0.0271, 0.0247, 0.0259], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 18:30:54,316 INFO [train.py:873] (3/4) Epoch 8, batch 3900, loss[loss=0.17, simple_loss=0.1889, pruned_loss=0.0756, over 14305.00 frames. ], tot_loss[loss=0.1496, simple_loss=0.1721, pruned_loss=0.06355, over 1942665.78 frames. ], batch size: 37, lr: 9.90e-03, grad_scale: 8.0 2022-12-07 18:30:58,940 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.085e+02 2.187e+02 2.677e+02 3.401e+02 6.261e+02, threshold=5.355e+02, percent-clipped=0.0 2022-12-07 18:31:00,372 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.66 vs. limit=2.0 2022-12-07 18:31:06,463 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=56849.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:31:19,881 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56864.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:31:33,509 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56879.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 18:31:45,675 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3195, 2.0003, 2.3573, 2.4411, 2.2784, 1.9424, 2.5172, 2.1458], device='cuda:3'), covar=tensor([0.0136, 0.0386, 0.0189, 0.0132, 0.0192, 0.0448, 0.0154, 0.0232], device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0224, 0.0338, 0.0279, 0.0224, 0.0269, 0.0246, 0.0258], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 18:31:58,575 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.39 vs. limit=5.0 2022-12-07 18:31:59,259 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56908.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:32:22,733 INFO [train.py:873] (3/4) Epoch 8, batch 4000, loss[loss=0.1713, simple_loss=0.1574, pruned_loss=0.09264, over 2556.00 frames. ], tot_loss[loss=0.1483, simple_loss=0.1711, pruned_loss=0.06274, over 1908924.37 frames. ], batch size: 100, lr: 9.89e-03, grad_scale: 8.0 2022-12-07 18:32:27,091 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.130e+02 2.242e+02 2.985e+02 3.676e+02 7.027e+02, threshold=5.970e+02, percent-clipped=5.0 2022-12-07 18:32:52,991 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56969.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:32:59,083 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9072, 3.9731, 4.0590, 3.7486, 3.9175, 4.2023, 1.6194, 3.6034], device='cuda:3'), covar=tensor([0.0348, 0.0428, 0.0525, 0.0582, 0.0534, 0.0293, 0.3670, 0.0434], device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0153, 0.0132, 0.0128, 0.0184, 0.0125, 0.0151, 0.0175], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 18:33:00,793 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0570, 2.8484, 2.5732, 2.7244, 2.9124, 2.9747, 3.0383, 2.9986], device='cuda:3'), covar=tensor([0.0991, 0.0810, 0.2362, 0.3131, 0.0927, 0.0982, 0.1240, 0.0995], device='cuda:3'), in_proj_covar=tensor([0.0336, 0.0243, 0.0402, 0.0514, 0.0294, 0.0373, 0.0369, 0.0323], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 18:33:14,716 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56993.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:33:16,381 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56995.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:33:44,394 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1198, 1.2764, 1.4370, 1.2122, 1.6569, 1.0179, 1.1007, 1.2231], device='cuda:3'), covar=tensor([0.0857, 0.0743, 0.1018, 0.0984, 0.0697, 0.0600, 0.0653, 0.0977], device='cuda:3'), in_proj_covar=tensor([0.0013, 0.0014, 0.0012, 0.0012, 0.0012, 0.0019, 0.0015, 0.0020], device='cuda:3'), out_proj_covar=tensor([8.8902e-05, 9.6562e-05, 8.5172e-05, 9.0818e-05, 8.7080e-05, 1.3139e-04, 1.0991e-04, 1.2537e-04], device='cuda:3') 2022-12-07 18:33:51,681 INFO [train.py:873] (3/4) Epoch 8, batch 4100, loss[loss=0.2023, simple_loss=0.201, pruned_loss=0.1018, over 7770.00 frames. ], tot_loss[loss=0.1489, simple_loss=0.1714, pruned_loss=0.06318, over 1900953.05 frames. ], batch size: 100, lr: 9.88e-03, grad_scale: 8.0 2022-12-07 18:33:56,137 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.189e+02 2.377e+02 3.103e+02 3.855e+02 1.239e+03, threshold=6.205e+02, percent-clipped=7.0 2022-12-07 18:34:02,139 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2022-12-07 18:34:08,942 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57054.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:34:10,589 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57056.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:34:13,535 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.73 vs. limit=5.0 2022-12-07 18:34:17,677 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2212, 3.0304, 3.0588, 3.2441, 3.1139, 3.1315, 3.3157, 2.7359], device='cuda:3'), covar=tensor([0.0476, 0.1164, 0.0484, 0.0524, 0.0879, 0.0530, 0.0675, 0.0629], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0245, 0.0167, 0.0160, 0.0164, 0.0133, 0.0250, 0.0150], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-07 18:34:48,116 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3704, 2.1535, 3.3604, 3.4803, 3.3697, 2.2548, 3.3425, 2.6039], device='cuda:3'), covar=tensor([0.0229, 0.0567, 0.0536, 0.0286, 0.0226, 0.0842, 0.0244, 0.0623], device='cuda:3'), in_proj_covar=tensor([0.0251, 0.0230, 0.0345, 0.0289, 0.0230, 0.0275, 0.0252, 0.0265], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 18:35:20,886 INFO [train.py:873] (3/4) Epoch 8, batch 4200, loss[loss=0.1213, simple_loss=0.1592, pruned_loss=0.04169, over 14281.00 frames. ], tot_loss[loss=0.1491, simple_loss=0.1716, pruned_loss=0.06331, over 1826410.67 frames. ], batch size: 25, lr: 9.87e-03, grad_scale: 8.0 2022-12-07 18:35:25,040 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.452e+02 2.260e+02 2.916e+02 3.546e+02 6.128e+02, threshold=5.833e+02, percent-clipped=0.0 2022-12-07 18:35:42,124 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57159.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:35:59,655 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57179.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 18:36:06,096 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57186.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:36:07,261 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2022-12-07 18:36:15,632 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57197.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:36:41,986 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57227.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 18:36:48,821 INFO [train.py:873] (3/4) Epoch 8, batch 4300, loss[loss=0.1341, simple_loss=0.1686, pruned_loss=0.04983, over 14442.00 frames. ], tot_loss[loss=0.1502, simple_loss=0.1727, pruned_loss=0.06381, over 1904013.91 frames. ], batch size: 51, lr: 9.86e-03, grad_scale: 8.0 2022-12-07 18:36:53,099 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.149e+02 2.274e+02 2.646e+02 3.498e+02 8.636e+02, threshold=5.293e+02, percent-clipped=3.0 2022-12-07 18:36:57,108 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57244.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:36:59,651 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57247.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:37:08,831 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9555, 4.7464, 4.4882, 4.9818, 4.4852, 4.1839, 5.0060, 4.8736], device='cuda:3'), covar=tensor([0.0617, 0.0616, 0.0683, 0.0566, 0.0766, 0.0557, 0.0550, 0.0686], device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0113, 0.0122, 0.0129, 0.0128, 0.0100, 0.0142, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 18:37:08,983 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57258.0, num_to_drop=1, layers_to_drop={3} 2022-12-07 18:37:14,092 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57264.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:37:28,911 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0203, 4.0368, 4.4486, 3.5515, 4.1338, 4.4690, 1.7128, 3.9125], device='cuda:3'), covar=tensor([0.0293, 0.0348, 0.0339, 0.0550, 0.0327, 0.0239, 0.3052, 0.0280], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0155, 0.0134, 0.0131, 0.0188, 0.0128, 0.0154, 0.0176], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 18:37:49,492 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57305.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:38:15,736 INFO [train.py:873] (3/4) Epoch 8, batch 4400, loss[loss=0.183, simple_loss=0.1721, pruned_loss=0.09694, over 1268.00 frames. ], tot_loss[loss=0.1493, simple_loss=0.1721, pruned_loss=0.06332, over 1877331.01 frames. ], batch size: 100, lr: 9.86e-03, grad_scale: 8.0 2022-12-07 18:38:19,429 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.42 vs. limit=5.0 2022-12-07 18:38:19,656 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 6.383e+01 2.226e+02 2.824e+02 3.594e+02 6.936e+02, threshold=5.649e+02, percent-clipped=3.0 2022-12-07 18:38:27,807 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57349.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:38:29,516 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57351.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:38:37,374 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9801, 2.5995, 5.1448, 3.3828, 4.6463, 2.1641, 3.6583, 4.5186], device='cuda:3'), covar=tensor([0.0466, 0.4180, 0.0248, 0.7669, 0.0554, 0.3560, 0.1136, 0.0388], device='cuda:3'), in_proj_covar=tensor([0.0234, 0.0232, 0.0185, 0.0315, 0.0204, 0.0236, 0.0224, 0.0192], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 18:39:10,899 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57398.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:39:12,940 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3411, 2.2499, 5.1455, 4.7791, 4.5720, 5.2482, 5.0984, 5.3006], device='cuda:3'), covar=tensor([0.1201, 0.1137, 0.0063, 0.0134, 0.0135, 0.0076, 0.0067, 0.0086], device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0155, 0.0116, 0.0160, 0.0135, 0.0131, 0.0109, 0.0114], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 18:39:43,843 INFO [train.py:873] (3/4) Epoch 8, batch 4500, loss[loss=0.1214, simple_loss=0.1553, pruned_loss=0.04375, over 13888.00 frames. ], tot_loss[loss=0.148, simple_loss=0.1716, pruned_loss=0.06219, over 1970328.43 frames. ], batch size: 20, lr: 9.85e-03, grad_scale: 8.0 2022-12-07 18:39:47,867 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.244e+02 2.294e+02 2.953e+02 3.643e+02 6.271e+02, threshold=5.907e+02, percent-clipped=3.0 2022-12-07 18:40:04,363 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57459.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:40:04,428 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57459.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:40:07,065 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57462.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:40:11,864 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.64 vs. limit=2.0 2022-12-07 18:40:35,436 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57495.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:40:45,764 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57507.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:40:47,186 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.66 vs. limit=5.0 2022-12-07 18:40:49,263 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4838, 1.3675, 3.5253, 1.5363, 3.3961, 3.4501, 2.4328, 3.7917], device='cuda:3'), covar=tensor([0.0203, 0.3089, 0.0315, 0.2337, 0.0718, 0.0374, 0.0862, 0.0167], device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0158, 0.0151, 0.0168, 0.0164, 0.0164, 0.0131, 0.0137], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 18:41:00,095 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57523.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:41:03,426 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=8.82 vs. limit=5.0 2022-12-07 18:41:04,745 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.2427, 1.4015, 1.3466, 1.2600, 1.5948, 0.7262, 1.0058, 0.8722], device='cuda:3'), covar=tensor([0.0341, 0.0625, 0.0657, 0.0280, 0.0334, 0.0361, 0.0443, 0.0573], device='cuda:3'), in_proj_covar=tensor([0.0012, 0.0013, 0.0011, 0.0012, 0.0012, 0.0019, 0.0015, 0.0019], device='cuda:3'), out_proj_covar=tensor([8.7105e-05, 9.2963e-05, 8.3622e-05, 8.8044e-05, 8.6964e-05, 1.2918e-04, 1.0900e-04, 1.2350e-04], device='cuda:3') 2022-12-07 18:41:10,290 INFO [train.py:873] (3/4) Epoch 8, batch 4600, loss[loss=0.1562, simple_loss=0.1776, pruned_loss=0.06737, over 14441.00 frames. ], tot_loss[loss=0.1491, simple_loss=0.1722, pruned_loss=0.06302, over 1958085.17 frames. ], batch size: 53, lr: 9.84e-03, grad_scale: 8.0 2022-12-07 18:41:14,686 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.272e+02 2.523e+02 3.027e+02 3.692e+02 7.181e+02, threshold=6.055e+02, percent-clipped=3.0 2022-12-07 18:41:16,409 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57542.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:41:26,157 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57553.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 18:41:29,206 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57556.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:41:36,033 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57564.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:42:07,596 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57600.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:42:18,281 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57612.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:42:24,878 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.90 vs. limit=5.0 2022-12-07 18:42:33,239 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2487, 2.3394, 2.8696, 2.0709, 1.9328, 2.4254, 1.3327, 2.5879], device='cuda:3'), covar=tensor([0.1982, 0.1772, 0.1104, 0.2077, 0.3420, 0.1760, 0.6829, 0.1427], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0087, 0.0083, 0.0089, 0.0110, 0.0076, 0.0128, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2022-12-07 18:42:38,956 INFO [train.py:873] (3/4) Epoch 8, batch 4700, loss[loss=0.1375, simple_loss=0.1699, pruned_loss=0.05254, over 13945.00 frames. ], tot_loss[loss=0.1488, simple_loss=0.1719, pruned_loss=0.06287, over 1940522.54 frames. ], batch size: 23, lr: 9.83e-03, grad_scale: 8.0 2022-12-07 18:42:39,592 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2022-12-07 18:42:43,279 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.474e+02 2.451e+02 2.895e+02 3.498e+02 7.656e+02, threshold=5.791e+02, percent-clipped=2.0 2022-12-07 18:42:51,342 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57649.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:42:53,015 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57651.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:43:33,484 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57697.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:43:35,285 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57699.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:43:40,754 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.68 vs. limit=5.0 2022-12-07 18:44:05,167 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57733.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:44:06,684 INFO [train.py:873] (3/4) Epoch 8, batch 4800, loss[loss=0.1628, simple_loss=0.1756, pruned_loss=0.07499, over 5998.00 frames. ], tot_loss[loss=0.1474, simple_loss=0.171, pruned_loss=0.06186, over 2038279.51 frames. ], batch size: 100, lr: 9.82e-03, grad_scale: 8.0 2022-12-07 18:44:10,753 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.453e+02 2.410e+02 2.943e+02 4.035e+02 9.695e+02, threshold=5.886e+02, percent-clipped=7.0 2022-12-07 18:44:23,454 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57754.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:44:23,539 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8236, 1.0556, 1.0055, 0.7348, 0.8246, 0.6826, 0.5647, 0.6884], device='cuda:3'), covar=tensor([0.0249, 0.0240, 0.0194, 0.0199, 0.0240, 0.0469, 0.0316, 0.0517], device='cuda:3'), in_proj_covar=tensor([0.0012, 0.0013, 0.0011, 0.0012, 0.0012, 0.0018, 0.0015, 0.0019], device='cuda:3'), out_proj_covar=tensor([8.6119e-05, 9.2109e-05, 8.3502e-05, 8.8321e-05, 8.6268e-05, 1.2642e-04, 1.0639e-04, 1.2242e-04], device='cuda:3') 2022-12-07 18:44:28,077 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9697, 2.4116, 4.0273, 4.0064, 3.8853, 2.3382, 3.9966, 3.0586], device='cuda:3'), covar=tensor([0.0250, 0.0630, 0.0512, 0.0295, 0.0251, 0.1037, 0.0224, 0.0612], device='cuda:3'), in_proj_covar=tensor([0.0250, 0.0231, 0.0347, 0.0290, 0.0230, 0.0276, 0.0252, 0.0264], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 18:44:43,714 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9262, 2.5049, 5.0126, 3.2723, 4.5609, 2.0208, 3.9255, 4.6345], device='cuda:3'), covar=tensor([0.0392, 0.4305, 0.0270, 0.7679, 0.0507, 0.3740, 0.0983, 0.0229], device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0228, 0.0182, 0.0307, 0.0201, 0.0235, 0.0219, 0.0187], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 18:44:44,085 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.96 vs. limit=5.0 2022-12-07 18:44:58,958 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57794.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:45:14,795 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2022-12-07 18:45:20,572 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57818.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:45:36,250 INFO [train.py:873] (3/4) Epoch 8, batch 4900, loss[loss=0.1482, simple_loss=0.1811, pruned_loss=0.05763, over 14289.00 frames. ], tot_loss[loss=0.1474, simple_loss=0.1711, pruned_loss=0.06188, over 2074781.90 frames. ], batch size: 25, lr: 9.81e-03, grad_scale: 8.0 2022-12-07 18:45:40,366 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.284e+02 2.301e+02 2.954e+02 3.459e+02 6.054e+02, threshold=5.907e+02, percent-clipped=1.0 2022-12-07 18:45:42,321 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57842.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:45:43,594 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 2022-12-07 18:45:50,042 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57851.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:45:51,584 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57853.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 18:46:03,484 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.23 vs. limit=5.0 2022-12-07 18:46:13,836 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.66 vs. limit=2.0 2022-12-07 18:46:24,222 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57890.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:46:32,435 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.1382, 4.7217, 4.6174, 5.0507, 4.7792, 4.3836, 5.1133, 4.2770], device='cuda:3'), covar=tensor([0.0275, 0.0922, 0.0319, 0.0401, 0.0723, 0.0529, 0.0493, 0.0462], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0248, 0.0169, 0.0162, 0.0166, 0.0135, 0.0255, 0.0154], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-07 18:46:33,313 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57900.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:46:34,129 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57901.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:46:44,819 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9543, 1.4063, 3.1430, 2.9505, 3.1110, 3.2021, 2.5837, 3.2119], device='cuda:3'), covar=tensor([0.1072, 0.1319, 0.0127, 0.0256, 0.0219, 0.0133, 0.0279, 0.0141], device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0155, 0.0117, 0.0159, 0.0135, 0.0130, 0.0109, 0.0115], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 18:47:03,937 INFO [train.py:873] (3/4) Epoch 8, batch 5000, loss[loss=0.1783, simple_loss=0.1832, pruned_loss=0.0867, over 7742.00 frames. ], tot_loss[loss=0.1497, simple_loss=0.1722, pruned_loss=0.06356, over 1953194.48 frames. ], batch size: 100, lr: 9.80e-03, grad_scale: 8.0 2022-12-07 18:47:08,079 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.420e+02 2.289e+02 3.008e+02 3.693e+02 7.100e+02, threshold=6.016e+02, percent-clipped=4.0 2022-12-07 18:47:15,352 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57948.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:47:47,170 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0861, 2.4789, 2.4777, 2.3691, 1.9005, 2.5509, 2.3053, 0.9656], device='cuda:3'), covar=tensor([0.3333, 0.1442, 0.1694, 0.1348, 0.1724, 0.0981, 0.2058, 0.4869], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0067, 0.0056, 0.0057, 0.0086, 0.0063, 0.0087, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2022-12-07 18:48:31,877 INFO [train.py:873] (3/4) Epoch 8, batch 5100, loss[loss=0.1463, simple_loss=0.1706, pruned_loss=0.06094, over 14220.00 frames. ], tot_loss[loss=0.1495, simple_loss=0.1722, pruned_loss=0.06337, over 1984657.63 frames. ], batch size: 60, lr: 9.80e-03, grad_scale: 8.0 2022-12-07 18:48:36,395 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.248e+02 2.346e+02 3.025e+02 3.888e+02 7.693e+02, threshold=6.049e+02, percent-clipped=2.0 2022-12-07 18:48:48,808 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58054.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:49:19,605 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58089.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:49:30,703 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58102.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:49:42,751 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.2613, 1.7207, 1.0883, 1.2298, 1.3897, 1.0292, 1.9509, 1.5518], device='cuda:3'), covar=tensor([0.1332, 0.2961, 0.0860, 0.1353, 0.1023, 0.0570, 0.0404, 0.0840], device='cuda:3'), in_proj_covar=tensor([0.0012, 0.0013, 0.0011, 0.0012, 0.0012, 0.0019, 0.0015, 0.0019], device='cuda:3'), out_proj_covar=tensor([8.6436e-05, 9.3310e-05, 8.2587e-05, 8.8557e-05, 8.6171e-05, 1.2792e-04, 1.0720e-04, 1.2248e-04], device='cuda:3') 2022-12-07 18:49:45,286 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58118.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:49:59,579 INFO [train.py:873] (3/4) Epoch 8, batch 5200, loss[loss=0.1355, simple_loss=0.1635, pruned_loss=0.05371, over 14411.00 frames. ], tot_loss[loss=0.1489, simple_loss=0.1721, pruned_loss=0.06282, over 2028474.17 frames. ], batch size: 53, lr: 9.79e-03, grad_scale: 16.0 2022-12-07 18:50:04,075 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.185e+02 2.199e+02 2.811e+02 3.642e+02 7.366e+02, threshold=5.622e+02, percent-clipped=3.0 2022-12-07 18:50:05,172 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58141.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:50:07,825 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58144.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:50:13,780 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58151.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:50:26,691 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58166.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:50:36,355 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8390, 1.2800, 2.4807, 2.2921, 2.4492, 2.4716, 1.6659, 2.4905], device='cuda:3'), covar=tensor([0.0869, 0.1089, 0.0124, 0.0303, 0.0277, 0.0138, 0.0560, 0.0161], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0156, 0.0118, 0.0160, 0.0137, 0.0131, 0.0111, 0.0116], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 18:50:56,069 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58199.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:50:59,087 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58202.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:51:01,538 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58205.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:51:27,323 INFO [train.py:873] (3/4) Epoch 8, batch 5300, loss[loss=0.1606, simple_loss=0.1836, pruned_loss=0.06879, over 14261.00 frames. ], tot_loss[loss=0.1487, simple_loss=0.1717, pruned_loss=0.06279, over 2013208.53 frames. ], batch size: 35, lr: 9.78e-03, grad_scale: 8.0 2022-12-07 18:51:32,495 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.244e+02 2.430e+02 3.290e+02 3.970e+02 9.558e+02, threshold=6.580e+02, percent-clipped=6.0 2022-12-07 18:52:55,483 INFO [train.py:873] (3/4) Epoch 8, batch 5400, loss[loss=0.1525, simple_loss=0.1397, pruned_loss=0.0827, over 1237.00 frames. ], tot_loss[loss=0.1491, simple_loss=0.1715, pruned_loss=0.06331, over 1939173.10 frames. ], batch size: 100, lr: 9.77e-03, grad_scale: 8.0 2022-12-07 18:53:00,690 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.457e+02 2.324e+02 2.944e+02 3.627e+02 6.890e+02, threshold=5.887e+02, percent-clipped=1.0 2022-12-07 18:53:10,520 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.7060, 5.4013, 5.2598, 5.7581, 5.2375, 4.9892, 5.7269, 5.6873], device='cuda:3'), covar=tensor([0.0605, 0.0536, 0.0572, 0.0459, 0.0679, 0.0362, 0.0620, 0.0513], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0119, 0.0128, 0.0135, 0.0131, 0.0104, 0.0146, 0.0127], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 18:53:27,447 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1634, 1.8601, 4.5897, 4.2814, 4.2582, 4.8022, 4.4861, 4.8008], device='cuda:3'), covar=tensor([0.1243, 0.1328, 0.0088, 0.0142, 0.0142, 0.0073, 0.0084, 0.0089], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0155, 0.0117, 0.0159, 0.0136, 0.0130, 0.0109, 0.0115], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 18:53:31,732 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0287, 1.9519, 2.0229, 2.1012, 1.9448, 1.6249, 1.3101, 1.7907], device='cuda:3'), covar=tensor([0.0360, 0.0481, 0.0530, 0.0307, 0.0418, 0.1447, 0.2053, 0.0403], device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0156, 0.0132, 0.0129, 0.0186, 0.0129, 0.0153, 0.0173], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 18:53:43,282 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58389.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:54:01,034 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.86 vs. limit=2.0 2022-12-07 18:54:15,290 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8203, 1.3782, 1.7128, 1.1898, 1.4857, 1.8686, 1.6836, 1.5777], device='cuda:3'), covar=tensor([0.0521, 0.0906, 0.0582, 0.0932, 0.1071, 0.0603, 0.0434, 0.1487], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0188, 0.0130, 0.0128, 0.0121, 0.0129, 0.0106, 0.0135], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0007, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005], device='cuda:3') 2022-12-07 18:54:23,918 INFO [train.py:873] (3/4) Epoch 8, batch 5500, loss[loss=0.1425, simple_loss=0.1356, pruned_loss=0.07477, over 2568.00 frames. ], tot_loss[loss=0.1489, simple_loss=0.1712, pruned_loss=0.06335, over 1896901.01 frames. ], batch size: 100, lr: 9.76e-03, grad_scale: 8.0 2022-12-07 18:54:26,070 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58437.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:54:29,553 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.638e+01 2.350e+02 2.867e+02 3.886e+02 1.534e+03, threshold=5.734e+02, percent-clipped=10.0 2022-12-07 18:55:18,754 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58497.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:55:21,715 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58500.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:55:28,810 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1056, 1.3321, 1.1258, 1.0639, 0.9764, 0.9595, 0.7282, 0.7499], device='cuda:3'), covar=tensor([0.0291, 0.0457, 0.0302, 0.0388, 0.0428, 0.0377, 0.0282, 0.0605], device='cuda:3'), in_proj_covar=tensor([0.0012, 0.0013, 0.0011, 0.0012, 0.0012, 0.0018, 0.0015, 0.0019], device='cuda:3'), out_proj_covar=tensor([8.5051e-05, 9.1380e-05, 8.1864e-05, 8.6403e-05, 8.5086e-05, 1.2488e-04, 1.0557e-04, 1.1989e-04], device='cuda:3') 2022-12-07 18:55:48,158 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0404, 2.2141, 2.8619, 2.3851, 2.7991, 2.8329, 2.7241, 2.4465], device='cuda:3'), covar=tensor([0.0832, 0.2702, 0.0900, 0.1730, 0.0554, 0.0907, 0.0931, 0.1862], device='cuda:3'), in_proj_covar=tensor([0.0320, 0.0322, 0.0391, 0.0309, 0.0376, 0.0305, 0.0356, 0.0328], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 18:55:52,342 INFO [train.py:873] (3/4) Epoch 8, batch 5600, loss[loss=0.1461, simple_loss=0.1655, pruned_loss=0.06337, over 11164.00 frames. ], tot_loss[loss=0.1504, simple_loss=0.1722, pruned_loss=0.06431, over 1879482.72 frames. ], batch size: 100, lr: 9.75e-03, grad_scale: 8.0 2022-12-07 18:55:57,836 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.352e+02 2.292e+02 2.814e+02 3.387e+02 5.498e+02, threshold=5.627e+02, percent-clipped=0.0 2022-12-07 18:56:31,566 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.3051, 1.4312, 1.2116, 1.0495, 1.5914, 0.7377, 1.3454, 1.5475], device='cuda:3'), covar=tensor([0.1305, 0.0680, 0.1033, 0.1118, 0.0868, 0.0758, 0.0888, 0.0560], device='cuda:3'), in_proj_covar=tensor([0.0022, 0.0022, 0.0022, 0.0021, 0.0022, 0.0032, 0.0023, 0.0023], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2022-12-07 18:57:21,322 INFO [train.py:873] (3/4) Epoch 8, batch 5700, loss[loss=0.1715, simple_loss=0.1918, pruned_loss=0.07557, over 14114.00 frames. ], tot_loss[loss=0.1493, simple_loss=0.1719, pruned_loss=0.0633, over 1929925.93 frames. ], batch size: 29, lr: 9.75e-03, grad_scale: 8.0 2022-12-07 18:57:26,806 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.215e+02 2.316e+02 2.828e+02 3.502e+02 7.696e+02, threshold=5.656e+02, percent-clipped=3.0 2022-12-07 18:57:38,419 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.18 vs. limit=5.0 2022-12-07 18:57:40,044 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.83 vs. limit=2.0 2022-12-07 18:57:59,151 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.83 vs. limit=5.0 2022-12-07 18:58:02,504 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58681.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:58:49,962 INFO [train.py:873] (3/4) Epoch 8, batch 5800, loss[loss=0.173, simple_loss=0.1495, pruned_loss=0.09824, over 1164.00 frames. ], tot_loss[loss=0.1491, simple_loss=0.1722, pruned_loss=0.06304, over 1997664.54 frames. ], batch size: 100, lr: 9.74e-03, grad_scale: 8.0 2022-12-07 18:58:52,704 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.8287, 5.5802, 5.4043, 5.8169, 5.3608, 5.2802, 5.8501, 5.6752], device='cuda:3'), covar=tensor([0.0628, 0.0703, 0.0598, 0.0478, 0.0634, 0.0322, 0.0581, 0.0608], device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0121, 0.0130, 0.0138, 0.0134, 0.0105, 0.0148, 0.0129], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 18:58:55,173 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.238e+02 2.215e+02 2.766e+02 3.426e+02 6.991e+02, threshold=5.532e+02, percent-clipped=1.0 2022-12-07 18:58:56,254 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58742.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:59:25,781 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58775.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:59:45,997 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58797.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 18:59:48,898 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58800.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:00:19,698 INFO [train.py:873] (3/4) Epoch 8, batch 5900, loss[loss=0.1674, simple_loss=0.1883, pruned_loss=0.07323, over 14097.00 frames. ], tot_loss[loss=0.1487, simple_loss=0.172, pruned_loss=0.06266, over 2003437.52 frames. ], batch size: 29, lr: 9.73e-03, grad_scale: 8.0 2022-12-07 19:00:20,718 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58836.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:00:25,227 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.436e+02 2.290e+02 2.920e+02 3.537e+02 8.146e+02, threshold=5.840e+02, percent-clipped=6.0 2022-12-07 19:00:28,995 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58845.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:00:31,575 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58848.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:01:22,570 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58906.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:01:47,423 INFO [train.py:873] (3/4) Epoch 8, batch 6000, loss[loss=0.1327, simple_loss=0.1337, pruned_loss=0.06587, over 2650.00 frames. ], tot_loss[loss=0.15, simple_loss=0.172, pruned_loss=0.06401, over 1888347.51 frames. ], batch size: 100, lr: 9.72e-03, grad_scale: 8.0 2022-12-07 19:01:47,424 INFO [train.py:896] (3/4) Computing validation loss 2022-12-07 19:02:10,414 INFO [train.py:905] (3/4) Epoch 8, validation: loss=0.1228, simple_loss=0.1649, pruned_loss=0.04039, over 857387.00 frames. 2022-12-07 19:02:10,415 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-07 19:02:16,061 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.501e+02 2.300e+02 3.027e+02 3.751e+02 1.020e+03, threshold=6.055e+02, percent-clipped=3.0 2022-12-07 19:02:30,257 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0348, 1.9277, 2.0038, 2.1104, 2.0144, 1.7167, 1.2936, 1.7863], device='cuda:3'), covar=tensor([0.0420, 0.0474, 0.0504, 0.0266, 0.0395, 0.1122, 0.1902, 0.0402], device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0153, 0.0131, 0.0128, 0.0185, 0.0127, 0.0152, 0.0173], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 19:02:30,684 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.52 vs. limit=2.0 2022-12-07 19:02:38,941 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58967.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:02:53,627 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2022-12-07 19:03:14,044 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([6.0286, 5.4793, 5.5784, 5.9544, 5.6343, 4.9523, 5.9540, 4.8713], device='cuda:3'), covar=tensor([0.0275, 0.0981, 0.0214, 0.0399, 0.0683, 0.0349, 0.0444, 0.0467], device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0247, 0.0169, 0.0160, 0.0166, 0.0134, 0.0253, 0.0152], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-07 19:03:38,337 INFO [train.py:873] (3/4) Epoch 8, batch 6100, loss[loss=0.1728, simple_loss=0.182, pruned_loss=0.08181, over 4999.00 frames. ], tot_loss[loss=0.1483, simple_loss=0.1711, pruned_loss=0.06273, over 1928142.71 frames. ], batch size: 100, lr: 9.71e-03, grad_scale: 8.0 2022-12-07 19:03:40,160 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59037.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:03:43,429 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.600e+02 2.326e+02 2.830e+02 3.436e+02 6.113e+02, threshold=5.661e+02, percent-clipped=1.0 2022-12-07 19:03:43,982 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2022-12-07 19:04:39,517 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8787, 1.6062, 4.4877, 2.3009, 4.4891, 4.8147, 4.5421, 5.3094], device='cuda:3'), covar=tensor([0.0164, 0.2962, 0.0366, 0.1984, 0.0258, 0.0252, 0.0210, 0.0103], device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0158, 0.0152, 0.0170, 0.0164, 0.0167, 0.0132, 0.0138], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 19:05:02,933 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59131.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:05:06,214 INFO [train.py:873] (3/4) Epoch 8, batch 6200, loss[loss=0.158, simple_loss=0.1808, pruned_loss=0.06758, over 14413.00 frames. ], tot_loss[loss=0.1471, simple_loss=0.1701, pruned_loss=0.06206, over 1896659.18 frames. ], batch size: 41, lr: 9.71e-03, grad_scale: 8.0 2022-12-07 19:05:11,776 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.400e+01 2.402e+02 3.111e+02 3.796e+02 6.098e+02, threshold=6.221e+02, percent-clipped=5.0 2022-12-07 19:05:32,087 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59164.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:06:10,684 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2022-12-07 19:06:26,865 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59225.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:06:35,505 INFO [train.py:873] (3/4) Epoch 8, batch 6300, loss[loss=0.1513, simple_loss=0.1519, pruned_loss=0.07534, over 2644.00 frames. ], tot_loss[loss=0.147, simple_loss=0.1703, pruned_loss=0.06182, over 1938360.29 frames. ], batch size: 100, lr: 9.70e-03, grad_scale: 8.0 2022-12-07 19:06:40,625 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.486e+02 2.206e+02 2.756e+02 3.412e+02 6.293e+02, threshold=5.512e+02, percent-clipped=2.0 2022-12-07 19:06:41,766 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6972, 1.5554, 4.3222, 3.9729, 3.9826, 4.4958, 4.0982, 4.4208], device='cuda:3'), covar=tensor([0.1911, 0.1755, 0.0162, 0.0244, 0.0243, 0.0169, 0.0163, 0.0191], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0158, 0.0118, 0.0162, 0.0137, 0.0132, 0.0111, 0.0118], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 19:06:44,285 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3573, 3.4042, 3.6062, 3.2994, 3.4741, 3.4049, 1.4134, 3.2977], device='cuda:3'), covar=tensor([0.0283, 0.0332, 0.0381, 0.0498, 0.0318, 0.0425, 0.3173, 0.0272], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0156, 0.0133, 0.0130, 0.0187, 0.0128, 0.0154, 0.0174], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 19:07:00,051 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59262.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:07:07,635 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2022-12-07 19:07:48,899 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4575, 0.9626, 1.3763, 0.9058, 1.2093, 1.4163, 1.2201, 1.2088], device='cuda:3'), covar=tensor([0.0278, 0.0923, 0.0414, 0.0472, 0.0913, 0.0479, 0.0327, 0.0878], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0188, 0.0129, 0.0125, 0.0122, 0.0127, 0.0106, 0.0134], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0007, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005], device='cuda:3') 2022-12-07 19:07:51,749 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2022-12-07 19:08:02,570 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5702, 3.3318, 3.0012, 3.1935, 3.4213, 3.4263, 3.5064, 3.5060], device='cuda:3'), covar=tensor([0.0884, 0.0670, 0.2078, 0.2828, 0.0774, 0.0850, 0.1112, 0.0852], device='cuda:3'), in_proj_covar=tensor([0.0338, 0.0240, 0.0399, 0.0513, 0.0287, 0.0378, 0.0367, 0.0324], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 19:08:04,133 INFO [train.py:873] (3/4) Epoch 8, batch 6400, loss[loss=0.1165, simple_loss=0.1489, pruned_loss=0.04205, over 14324.00 frames. ], tot_loss[loss=0.1464, simple_loss=0.1704, pruned_loss=0.06121, over 1931756.63 frames. ], batch size: 28, lr: 9.69e-03, grad_scale: 8.0 2022-12-07 19:08:05,993 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59337.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:08:09,582 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.267e+01 2.304e+02 2.776e+02 3.758e+02 5.907e+02, threshold=5.553e+02, percent-clipped=4.0 2022-12-07 19:08:48,447 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=59385.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:08:58,108 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1599, 2.0467, 2.5361, 1.6044, 1.6940, 2.4125, 1.2809, 2.1308], device='cuda:3'), covar=tensor([0.1083, 0.2072, 0.0622, 0.2292, 0.3216, 0.0764, 0.4675, 0.0936], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0094, 0.0084, 0.0092, 0.0114, 0.0078, 0.0133, 0.0082], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2022-12-07 19:09:09,244 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9729, 1.6721, 2.0436, 1.7088, 2.1340, 1.8140, 1.6628, 1.9204], device='cuda:3'), covar=tensor([0.0418, 0.1199, 0.0211, 0.0349, 0.0222, 0.0510, 0.0197, 0.0301], device='cuda:3'), in_proj_covar=tensor([0.0321, 0.0320, 0.0391, 0.0306, 0.0376, 0.0305, 0.0353, 0.0323], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 19:09:29,257 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59431.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:09:32,601 INFO [train.py:873] (3/4) Epoch 8, batch 6500, loss[loss=0.1658, simple_loss=0.1834, pruned_loss=0.07406, over 14041.00 frames. ], tot_loss[loss=0.1469, simple_loss=0.1705, pruned_loss=0.06166, over 1931595.68 frames. ], batch size: 22, lr: 9.68e-03, grad_scale: 8.0 2022-12-07 19:09:37,915 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.592e+02 2.343e+02 2.966e+02 3.672e+02 1.051e+03, threshold=5.932e+02, percent-clipped=2.0 2022-12-07 19:10:09,347 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2915, 1.9835, 2.1696, 1.3463, 1.9752, 2.1689, 2.3128, 1.9715], device='cuda:3'), covar=tensor([0.0830, 0.1139, 0.1224, 0.2510, 0.1126, 0.0889, 0.0797, 0.1781], device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0189, 0.0130, 0.0126, 0.0123, 0.0129, 0.0107, 0.0135], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0007, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005], device='cuda:3') 2022-12-07 19:10:10,832 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=59479.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:10:31,102 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59502.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:10:46,630 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59520.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:10:59,607 INFO [train.py:873] (3/4) Epoch 8, batch 6600, loss[loss=0.146, simple_loss=0.1704, pruned_loss=0.06081, over 14219.00 frames. ], tot_loss[loss=0.1473, simple_loss=0.1707, pruned_loss=0.06191, over 1941279.86 frames. ], batch size: 60, lr: 9.67e-03, grad_scale: 8.0 2022-12-07 19:11:04,956 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.142e+02 2.283e+02 2.818e+02 3.412e+02 6.206e+02, threshold=5.635e+02, percent-clipped=2.0 2022-12-07 19:11:23,729 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59562.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:11:24,620 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59563.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:11:35,111 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.66 vs. limit=2.0 2022-12-07 19:12:00,932 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9293, 3.7082, 3.4085, 2.5713, 3.3535, 3.6604, 3.9581, 3.1451], device='cuda:3'), covar=tensor([0.0605, 0.1753, 0.1124, 0.2108, 0.0885, 0.0547, 0.0690, 0.1414], device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0186, 0.0128, 0.0125, 0.0122, 0.0128, 0.0106, 0.0134], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0007, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005], device='cuda:3') 2022-12-07 19:12:05,781 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=59610.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:12:23,177 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2022-12-07 19:12:27,868 INFO [train.py:873] (3/4) Epoch 8, batch 6700, loss[loss=0.1658, simple_loss=0.1841, pruned_loss=0.07379, over 14261.00 frames. ], tot_loss[loss=0.1478, simple_loss=0.171, pruned_loss=0.06229, over 1934039.05 frames. ], batch size: 80, lr: 9.66e-03, grad_scale: 8.0 2022-12-07 19:12:32,694 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.081e+02 2.361e+02 2.930e+02 3.597e+02 6.848e+02, threshold=5.861e+02, percent-clipped=3.0 2022-12-07 19:12:34,674 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0932, 2.7973, 2.1298, 3.1992, 2.9336, 3.0728, 2.5357, 2.1575], device='cuda:3'), covar=tensor([0.0822, 0.1726, 0.4249, 0.0484, 0.0997, 0.1158, 0.1702, 0.4399], device='cuda:3'), in_proj_covar=tensor([0.0248, 0.0294, 0.0277, 0.0223, 0.0287, 0.0276, 0.0253, 0.0266], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2022-12-07 19:12:45,197 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.81 vs. limit=2.0 2022-12-07 19:13:08,691 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8925, 1.8012, 4.6629, 2.5624, 4.3869, 4.8265, 4.4943, 5.2003], device='cuda:3'), covar=tensor([0.0156, 0.2692, 0.0243, 0.1691, 0.0266, 0.0262, 0.0225, 0.0117], device='cuda:3'), in_proj_covar=tensor([0.0162, 0.0156, 0.0151, 0.0167, 0.0163, 0.0164, 0.0132, 0.0137], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 19:13:18,213 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.3660, 1.3262, 1.1812, 1.0312, 1.3623, 0.8065, 1.2965, 1.1835], device='cuda:3'), covar=tensor([0.0809, 0.0984, 0.0773, 0.0925, 0.0888, 0.0840, 0.0615, 0.0950], device='cuda:3'), in_proj_covar=tensor([0.0022, 0.0024, 0.0024, 0.0022, 0.0024, 0.0033, 0.0023, 0.0023], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2022-12-07 19:13:18,979 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7526, 1.5901, 1.7670, 2.0623, 1.3783, 1.7105, 1.6782, 1.8715], device='cuda:3'), covar=tensor([0.0088, 0.0159, 0.0093, 0.0060, 0.0172, 0.0182, 0.0107, 0.0069], device='cuda:3'), in_proj_covar=tensor([0.0251, 0.0229, 0.0344, 0.0289, 0.0230, 0.0275, 0.0251, 0.0263], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 19:13:55,716 INFO [train.py:873] (3/4) Epoch 8, batch 6800, loss[loss=0.1429, simple_loss=0.1736, pruned_loss=0.05613, over 14484.00 frames. ], tot_loss[loss=0.1475, simple_loss=0.1709, pruned_loss=0.06207, over 2001227.76 frames. ], batch size: 49, lr: 9.66e-03, grad_scale: 8.0 2022-12-07 19:14:01,555 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.385e+02 2.278e+02 2.761e+02 3.768e+02 7.478e+02, threshold=5.523e+02, percent-clipped=3.0 2022-12-07 19:14:07,487 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.9527, 0.9456, 0.9333, 0.7770, 0.9651, 0.5458, 0.9493, 0.8649], device='cuda:3'), covar=tensor([0.0656, 0.0879, 0.0780, 0.0629, 0.0422, 0.0695, 0.0552, 0.0508], device='cuda:3'), in_proj_covar=tensor([0.0022, 0.0023, 0.0023, 0.0021, 0.0024, 0.0032, 0.0022, 0.0023], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2022-12-07 19:14:30,252 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.9511, 5.3305, 5.3144, 5.8524, 5.3228, 4.7535, 5.8546, 4.8252], device='cuda:3'), covar=tensor([0.0237, 0.0992, 0.0251, 0.0398, 0.0719, 0.0301, 0.0455, 0.0480], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0244, 0.0169, 0.0160, 0.0164, 0.0130, 0.0250, 0.0153], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 19:15:12,039 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59820.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:15:17,896 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4389, 1.7848, 2.3673, 2.0050, 2.5435, 2.2524, 2.0790, 2.1598], device='cuda:3'), covar=tensor([0.0452, 0.1790, 0.0511, 0.1063, 0.0399, 0.0825, 0.0552, 0.1158], device='cuda:3'), in_proj_covar=tensor([0.0329, 0.0324, 0.0399, 0.0309, 0.0380, 0.0311, 0.0361, 0.0329], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 19:15:24,930 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4184, 1.0207, 1.3672, 0.8885, 1.2628, 1.4389, 1.0948, 1.1977], device='cuda:3'), covar=tensor([0.0335, 0.0781, 0.0460, 0.0503, 0.0646, 0.0538, 0.0317, 0.0716], device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0183, 0.0126, 0.0123, 0.0121, 0.0126, 0.0103, 0.0133], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0007, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005], device='cuda:3') 2022-12-07 19:15:25,705 INFO [train.py:873] (3/4) Epoch 8, batch 6900, loss[loss=0.2338, simple_loss=0.1988, pruned_loss=0.1344, over 1248.00 frames. ], tot_loss[loss=0.1499, simple_loss=0.1723, pruned_loss=0.06369, over 1965710.60 frames. ], batch size: 100, lr: 9.65e-03, grad_scale: 8.0 2022-12-07 19:15:30,915 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.022e+02 2.523e+02 3.039e+02 3.738e+02 1.408e+03, threshold=6.079e+02, percent-clipped=6.0 2022-12-07 19:15:32,731 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9976, 2.1374, 2.1072, 2.2107, 1.8882, 2.2791, 2.0307, 1.2086], device='cuda:3'), covar=tensor([0.1654, 0.0701, 0.0902, 0.0697, 0.1098, 0.0491, 0.1305, 0.2740], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0068, 0.0056, 0.0058, 0.0086, 0.0065, 0.0091, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2022-12-07 19:15:36,992 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59848.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:15:45,655 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59858.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:15:54,679 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=59868.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:16:13,237 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59889.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:16:31,685 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59909.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:16:46,530 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2022-12-07 19:16:53,933 INFO [train.py:873] (3/4) Epoch 8, batch 7000, loss[loss=0.2032, simple_loss=0.1778, pruned_loss=0.1143, over 1220.00 frames. ], tot_loss[loss=0.1492, simple_loss=0.1717, pruned_loss=0.06335, over 1893707.88 frames. ], batch size: 100, lr: 9.64e-03, grad_scale: 8.0 2022-12-07 19:16:59,264 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.368e+02 2.338e+02 3.066e+02 3.722e+02 8.266e+02, threshold=6.131e+02, percent-clipped=2.0 2022-12-07 19:17:08,027 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59950.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:17:24,289 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.2214, 1.3642, 1.2215, 1.0862, 0.8862, 0.9268, 0.9615, 0.8733], device='cuda:3'), covar=tensor([0.0209, 0.0225, 0.0220, 0.0221, 0.0401, 0.0391, 0.0219, 0.0492], device='cuda:3'), in_proj_covar=tensor([0.0012, 0.0014, 0.0011, 0.0012, 0.0012, 0.0019, 0.0016, 0.0020], device='cuda:3'), out_proj_covar=tensor([9.0770e-05, 9.6840e-05, 8.5627e-05, 9.1193e-05, 8.7172e-05, 1.3442e-04, 1.1266e-04, 1.2738e-04], device='cuda:3') 2022-12-07 19:17:36,376 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.9619, 1.0720, 0.8798, 1.0204, 1.0626, 0.5656, 0.8452, 0.9565], device='cuda:3'), covar=tensor([0.0402, 0.0559, 0.0425, 0.0344, 0.0239, 0.0495, 0.0840, 0.0416], device='cuda:3'), in_proj_covar=tensor([0.0022, 0.0024, 0.0023, 0.0022, 0.0024, 0.0033, 0.0022, 0.0023], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2022-12-07 19:18:24,216 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3562, 3.9973, 3.7410, 3.9196, 4.1331, 4.2485, 4.3777, 4.3472], device='cuda:3'), covar=tensor([0.0889, 0.0527, 0.2186, 0.2697, 0.0763, 0.0793, 0.0868, 0.0751], device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0244, 0.0414, 0.0524, 0.0301, 0.0388, 0.0376, 0.0332], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 19:18:27,447 INFO [train.py:873] (3/4) Epoch 8, batch 7100, loss[loss=0.1703, simple_loss=0.1604, pruned_loss=0.09008, over 2671.00 frames. ], tot_loss[loss=0.1492, simple_loss=0.1715, pruned_loss=0.06342, over 1893246.30 frames. ], batch size: 100, lr: 9.63e-03, grad_scale: 8.0 2022-12-07 19:18:31,933 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60040.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:18:32,560 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.335e+02 2.178e+02 2.684e+02 3.271e+02 7.336e+02, threshold=5.369e+02, percent-clipped=1.0 2022-12-07 19:18:42,294 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60052.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:18:48,463 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8320, 0.7356, 0.6998, 0.7389, 0.7733, 0.2185, 0.6982, 0.7934], device='cuda:3'), covar=tensor([0.0283, 0.0625, 0.0413, 0.0211, 0.0283, 0.0260, 0.0759, 0.0402], device='cuda:3'), in_proj_covar=tensor([0.0022, 0.0024, 0.0024, 0.0022, 0.0024, 0.0033, 0.0023, 0.0023], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2022-12-07 19:18:54,044 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60065.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:19:00,707 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.69 vs. limit=2.0 2022-12-07 19:19:05,154 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60078.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:19:12,088 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2022-12-07 19:19:24,568 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60101.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 19:19:35,519 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60113.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 19:19:46,377 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60126.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:19:53,837 INFO [train.py:873] (3/4) Epoch 8, batch 7200, loss[loss=0.1753, simple_loss=0.1741, pruned_loss=0.08826, over 2600.00 frames. ], tot_loss[loss=0.1495, simple_loss=0.1716, pruned_loss=0.06369, over 1867556.17 frames. ], batch size: 100, lr: 9.63e-03, grad_scale: 8.0 2022-12-07 19:19:57,781 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60139.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:19:59,258 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.579e+01 2.475e+02 2.936e+02 3.540e+02 1.198e+03, threshold=5.873e+02, percent-clipped=8.0 2022-12-07 19:19:59,450 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8852, 1.6280, 2.0437, 1.7003, 1.8775, 1.4260, 1.7210, 1.8429], device='cuda:3'), covar=tensor([0.1451, 0.2459, 0.0246, 0.1676, 0.0747, 0.1273, 0.0900, 0.0536], device='cuda:3'), in_proj_covar=tensor([0.0236, 0.0234, 0.0185, 0.0316, 0.0208, 0.0236, 0.0225, 0.0196], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 19:20:14,436 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60158.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:20:49,231 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60198.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:20:54,600 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60204.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:20:56,397 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60206.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:21:22,475 INFO [train.py:873] (3/4) Epoch 8, batch 7300, loss[loss=0.1578, simple_loss=0.14, pruned_loss=0.08783, over 1237.00 frames. ], tot_loss[loss=0.1474, simple_loss=0.1705, pruned_loss=0.06219, over 1907174.79 frames. ], batch size: 100, lr: 9.62e-03, grad_scale: 16.0 2022-12-07 19:21:28,209 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 6.538e+01 2.297e+02 3.128e+02 4.022e+02 9.031e+02, threshold=6.256e+02, percent-clipped=2.0 2022-12-07 19:21:30,687 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60245.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:21:35,555 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.51 vs. limit=2.0 2022-12-07 19:21:43,251 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60259.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:21:57,009 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5413, 2.5680, 4.6937, 3.1392, 4.2794, 2.1629, 3.4157, 4.0899], device='cuda:3'), covar=tensor([0.0506, 0.4031, 0.0243, 0.7517, 0.0754, 0.3403, 0.1232, 0.0548], device='cuda:3'), in_proj_covar=tensor([0.0235, 0.0231, 0.0185, 0.0316, 0.0208, 0.0232, 0.0223, 0.0194], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 19:22:49,540 INFO [train.py:873] (3/4) Epoch 8, batch 7400, loss[loss=0.1328, simple_loss=0.1669, pruned_loss=0.04933, over 14439.00 frames. ], tot_loss[loss=0.1476, simple_loss=0.1707, pruned_loss=0.06226, over 1952252.50 frames. ], batch size: 53, lr: 9.61e-03, grad_scale: 8.0 2022-12-07 19:22:56,039 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.153e+02 2.246e+02 2.782e+02 3.435e+02 7.808e+02, threshold=5.565e+02, percent-clipped=3.0 2022-12-07 19:23:00,176 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2278, 2.0269, 3.1895, 3.2352, 3.2229, 2.1230, 3.1329, 2.3950], device='cuda:3'), covar=tensor([0.0257, 0.0616, 0.0494, 0.0360, 0.0226, 0.0868, 0.0232, 0.0687], device='cuda:3'), in_proj_covar=tensor([0.0254, 0.0232, 0.0347, 0.0294, 0.0233, 0.0279, 0.0257, 0.0266], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 19:23:43,976 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60396.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 19:23:46,590 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1485, 2.9127, 2.2252, 3.1973, 2.9381, 3.0446, 2.6464, 2.2693], device='cuda:3'), covar=tensor([0.0736, 0.1425, 0.3412, 0.0659, 0.0905, 0.1120, 0.1284, 0.3443], device='cuda:3'), in_proj_covar=tensor([0.0252, 0.0295, 0.0273, 0.0228, 0.0291, 0.0278, 0.0252, 0.0266], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2022-12-07 19:23:51,414 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=8.46 vs. limit=5.0 2022-12-07 19:23:54,278 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60408.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 19:24:06,086 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60421.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:24:16,537 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2022-12-07 19:24:17,747 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60434.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:24:18,544 INFO [train.py:873] (3/4) Epoch 8, batch 7500, loss[loss=0.1631, simple_loss=0.1439, pruned_loss=0.09108, over 1267.00 frames. ], tot_loss[loss=0.1472, simple_loss=0.1707, pruned_loss=0.06189, over 2012284.72 frames. ], batch size: 100, lr: 9.60e-03, grad_scale: 8.0 2022-12-07 19:24:24,272 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.353e+02 2.510e+02 2.850e+02 3.552e+02 6.747e+02, threshold=5.701e+02, percent-clipped=5.0 2022-12-07 19:24:28,299 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 2022-12-07 19:24:46,955 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=8.08 vs. limit=5.0 2022-12-07 19:24:57,386 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.74 vs. limit=2.0 2022-12-07 19:25:46,336 INFO [train.py:873] (3/4) Epoch 9, batch 0, loss[loss=0.1868, simple_loss=0.1975, pruned_loss=0.08804, over 9511.00 frames. ], tot_loss[loss=0.1868, simple_loss=0.1975, pruned_loss=0.08804, over 9511.00 frames. ], batch size: 100, lr: 9.08e-03, grad_scale: 8.0 2022-12-07 19:25:46,337 INFO [train.py:896] (3/4) Computing validation loss 2022-12-07 19:25:53,549 INFO [train.py:905] (3/4) Epoch 9, validation: loss=0.1275, simple_loss=0.1706, pruned_loss=0.04216, over 857387.00 frames. 2022-12-07 19:25:53,550 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-07 19:26:00,139 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60504.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:26:33,659 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 6.503e+01 2.217e+02 2.837e+02 3.988e+02 6.329e+02, threshold=5.675e+02, percent-clipped=3.0 2022-12-07 19:26:36,527 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60545.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:26:43,664 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60552.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:26:45,475 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60554.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:27:19,870 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60593.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:27:23,502 INFO [train.py:873] (3/4) Epoch 9, batch 100, loss[loss=0.141, simple_loss=0.1405, pruned_loss=0.07076, over 2553.00 frames. ], tot_loss[loss=0.1464, simple_loss=0.1704, pruned_loss=0.06118, over 848257.68 frames. ], batch size: 100, lr: 9.08e-03, grad_scale: 8.0 2022-12-07 19:27:58,044 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8312, 4.3928, 4.3372, 4.7568, 4.5479, 4.2911, 4.8251, 4.0486], device='cuda:3'), covar=tensor([0.0360, 0.1067, 0.0328, 0.0477, 0.0681, 0.0621, 0.0536, 0.0499], device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0243, 0.0165, 0.0161, 0.0162, 0.0132, 0.0249, 0.0151], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 19:28:03,331 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.184e+02 2.443e+02 2.952e+02 3.884e+02 8.156e+02, threshold=5.905e+02, percent-clipped=9.0 2022-12-07 19:28:04,992 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2022-12-07 19:28:47,428 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2022-12-07 19:28:51,026 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60696.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:28:51,888 INFO [train.py:873] (3/4) Epoch 9, batch 200, loss[loss=0.1616, simple_loss=0.164, pruned_loss=0.07964, over 6027.00 frames. ], tot_loss[loss=0.1457, simple_loss=0.1694, pruned_loss=0.06095, over 1267571.55 frames. ], batch size: 100, lr: 9.07e-03, grad_scale: 8.0 2022-12-07 19:28:53,652 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1318, 1.8287, 4.3636, 4.0385, 4.0192, 4.4400, 3.9435, 4.4075], device='cuda:3'), covar=tensor([0.1225, 0.1275, 0.0082, 0.0169, 0.0168, 0.0095, 0.0149, 0.0105], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0158, 0.0120, 0.0164, 0.0140, 0.0134, 0.0112, 0.0118], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 19:29:01,485 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60708.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:29:11,319 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60718.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:29:13,793 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60721.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:29:24,917 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60734.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:29:32,172 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.211e+01 2.496e+02 3.042e+02 4.274e+02 1.188e+03, threshold=6.084e+02, percent-clipped=4.0 2022-12-07 19:29:33,112 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60744.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:29:44,437 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60756.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:29:55,860 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60769.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:30:04,781 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60779.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:30:07,105 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60782.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:30:20,883 INFO [train.py:873] (3/4) Epoch 9, batch 300, loss[loss=0.1491, simple_loss=0.1714, pruned_loss=0.06344, over 14031.00 frames. ], tot_loss[loss=0.1466, simple_loss=0.1694, pruned_loss=0.06192, over 1457404.20 frames. ], batch size: 22, lr: 9.06e-03, grad_scale: 4.0 2022-12-07 19:30:49,349 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 2022-12-07 19:30:50,175 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.47 vs. limit=5.0 2022-12-07 19:31:01,441 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.410e+02 2.299e+02 2.848e+02 3.507e+02 1.292e+03, threshold=5.696e+02, percent-clipped=2.0 2022-12-07 19:31:11,049 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60854.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:31:37,614 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.7848, 0.8041, 0.7147, 0.7781, 0.8276, 0.1121, 0.7684, 0.8639], device='cuda:3'), covar=tensor([0.0203, 0.0400, 0.0328, 0.0344, 0.0247, 0.0151, 0.0743, 0.0270], device='cuda:3'), in_proj_covar=tensor([0.0022, 0.0023, 0.0024, 0.0021, 0.0023, 0.0033, 0.0023, 0.0024], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2022-12-07 19:31:49,308 INFO [train.py:873] (3/4) Epoch 9, batch 400, loss[loss=0.1238, simple_loss=0.1431, pruned_loss=0.05229, over 10042.00 frames. ], tot_loss[loss=0.1451, simple_loss=0.1689, pruned_loss=0.06071, over 1648537.87 frames. ], batch size: 12, lr: 9.06e-03, grad_scale: 8.0 2022-12-07 19:31:53,801 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60902.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:32:29,748 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.639e+01 2.151e+02 2.975e+02 3.759e+02 9.050e+02, threshold=5.950e+02, percent-clipped=6.0 2022-12-07 19:32:37,894 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8093, 4.3207, 4.2442, 4.7066, 4.3628, 4.1960, 4.7659, 3.9174], device='cuda:3'), covar=tensor([0.0332, 0.0981, 0.0372, 0.0537, 0.0832, 0.0607, 0.0541, 0.0597], device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0245, 0.0168, 0.0163, 0.0164, 0.0134, 0.0251, 0.0153], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-07 19:33:18,102 INFO [train.py:873] (3/4) Epoch 9, batch 500, loss[loss=0.1486, simple_loss=0.1703, pruned_loss=0.0635, over 14461.00 frames. ], tot_loss[loss=0.1457, simple_loss=0.1696, pruned_loss=0.06087, over 1806491.38 frames. ], batch size: 27, lr: 9.05e-03, grad_scale: 8.0 2022-12-07 19:33:20,656 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.73 vs. limit=2.0 2022-12-07 19:33:30,079 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1964, 1.4594, 4.0744, 1.7355, 3.9980, 4.1607, 3.3859, 4.5850], device='cuda:3'), covar=tensor([0.0208, 0.3175, 0.0421, 0.2215, 0.0394, 0.0429, 0.0578, 0.0130], device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0159, 0.0156, 0.0169, 0.0169, 0.0166, 0.0135, 0.0137], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 19:33:46,466 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7403, 2.6885, 2.6875, 2.7718, 2.1940, 2.9577, 2.5746, 1.3109], device='cuda:3'), covar=tensor([0.1871, 0.1216, 0.1307, 0.0773, 0.1278, 0.0496, 0.1751, 0.3104], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0068, 0.0055, 0.0057, 0.0085, 0.0064, 0.0090, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2022-12-07 19:33:59,480 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.027e+02 2.291e+02 2.820e+02 3.548e+02 5.650e+02, threshold=5.639e+02, percent-clipped=0.0 2022-12-07 19:34:01,298 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.5766, 1.7494, 1.9492, 1.2766, 1.3490, 1.7888, 1.2247, 1.6484], device='cuda:3'), covar=tensor([0.1631, 0.1826, 0.0709, 0.2510, 0.2578, 0.0890, 0.3488, 0.0983], device='cuda:3'), in_proj_covar=tensor([0.0073, 0.0090, 0.0082, 0.0089, 0.0109, 0.0076, 0.0127, 0.0078], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2022-12-07 19:34:03,399 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 2022-12-07 19:34:14,694 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=61061.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:34:22,712 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.02 vs. limit=2.0 2022-12-07 19:34:25,954 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=61074.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:34:42,562 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1838, 3.8955, 3.7937, 4.2274, 3.8210, 3.4491, 4.2975, 4.1092], device='cuda:3'), covar=tensor([0.0621, 0.0757, 0.0771, 0.0639, 0.0773, 0.0793, 0.0536, 0.0790], device='cuda:3'), in_proj_covar=tensor([0.0120, 0.0116, 0.0126, 0.0132, 0.0128, 0.0104, 0.0142, 0.0125], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-07 19:34:45,880 INFO [train.py:873] (3/4) Epoch 9, batch 600, loss[loss=0.1381, simple_loss=0.1704, pruned_loss=0.05294, over 14385.00 frames. ], tot_loss[loss=0.1446, simple_loss=0.1691, pruned_loss=0.06004, over 1895225.23 frames. ], batch size: 55, lr: 9.04e-03, grad_scale: 8.0 2022-12-07 19:34:46,605 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.43 vs. limit=5.0 2022-12-07 19:35:08,459 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=61122.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:35:25,944 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.178e+01 2.284e+02 2.726e+02 3.406e+02 8.336e+02, threshold=5.452e+02, percent-clipped=5.0 2022-12-07 19:35:34,060 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0268, 1.8898, 3.2383, 2.2776, 3.0324, 1.7517, 2.4312, 2.9913], device='cuda:3'), covar=tensor([0.0780, 0.4489, 0.0437, 0.5949, 0.0799, 0.3821, 0.1436, 0.0518], device='cuda:3'), in_proj_covar=tensor([0.0239, 0.0229, 0.0186, 0.0314, 0.0212, 0.0236, 0.0224, 0.0192], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 19:36:13,114 INFO [train.py:873] (3/4) Epoch 9, batch 700, loss[loss=0.1236, simple_loss=0.1461, pruned_loss=0.05057, over 4943.00 frames. ], tot_loss[loss=0.1441, simple_loss=0.1682, pruned_loss=0.06003, over 1857541.29 frames. ], batch size: 100, lr: 9.03e-03, grad_scale: 8.0 2022-12-07 19:36:47,164 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8381, 1.1398, 1.2468, 1.2877, 1.0852, 1.3665, 1.0883, 0.8678], device='cuda:3'), covar=tensor([0.2554, 0.0956, 0.0359, 0.0390, 0.1299, 0.0546, 0.1621, 0.1296], device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0068, 0.0055, 0.0056, 0.0086, 0.0064, 0.0089, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2022-12-07 19:36:54,420 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.093e+02 2.220e+02 2.923e+02 3.750e+02 6.888e+02, threshold=5.845e+02, percent-clipped=6.0 2022-12-07 19:37:35,754 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1878, 1.2022, 1.2819, 1.1366, 1.2554, 0.6303, 1.0710, 1.1302], device='cuda:3'), covar=tensor([0.1145, 0.0975, 0.0591, 0.0529, 0.0995, 0.0866, 0.0833, 0.1206], device='cuda:3'), in_proj_covar=tensor([0.0022, 0.0023, 0.0023, 0.0021, 0.0023, 0.0033, 0.0023, 0.0024], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2022-12-07 19:37:41,718 INFO [train.py:873] (3/4) Epoch 9, batch 800, loss[loss=0.132, simple_loss=0.1698, pruned_loss=0.0471, over 14098.00 frames. ], tot_loss[loss=0.1448, simple_loss=0.1686, pruned_loss=0.06052, over 1904900.44 frames. ], batch size: 29, lr: 9.03e-03, grad_scale: 8.0 2022-12-07 19:37:47,713 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 2022-12-07 19:38:00,405 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1757, 3.2176, 3.1345, 3.4436, 2.4591, 3.3846, 3.0325, 1.5552], device='cuda:3'), covar=tensor([0.2337, 0.0795, 0.1590, 0.0431, 0.1198, 0.0679, 0.1474, 0.2828], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0067, 0.0055, 0.0056, 0.0086, 0.0064, 0.0090, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2022-12-07 19:38:22,020 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.383e+02 2.254e+02 2.833e+02 3.522e+02 6.123e+02, threshold=5.665e+02, percent-clipped=1.0 2022-12-07 19:38:49,251 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=61374.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:38:59,869 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.77 vs. limit=2.0 2022-12-07 19:39:09,294 INFO [train.py:873] (3/4) Epoch 9, batch 900, loss[loss=0.1283, simple_loss=0.1585, pruned_loss=0.049, over 14302.00 frames. ], tot_loss[loss=0.143, simple_loss=0.1674, pruned_loss=0.0593, over 1931275.34 frames. ], batch size: 60, lr: 9.02e-03, grad_scale: 8.0 2022-12-07 19:39:20,085 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.94 vs. limit=5.0 2022-12-07 19:39:27,331 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=61417.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:39:31,538 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=61422.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:39:50,649 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.310e+02 2.290e+02 3.289e+02 4.035e+02 9.123e+02, threshold=6.578e+02, percent-clipped=5.0 2022-12-07 19:40:14,656 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9483, 2.5655, 3.9882, 3.9991, 3.9810, 2.4593, 3.9365, 3.0763], device='cuda:3'), covar=tensor([0.0266, 0.0611, 0.0527, 0.0273, 0.0281, 0.0955, 0.0267, 0.0645], device='cuda:3'), in_proj_covar=tensor([0.0255, 0.0232, 0.0347, 0.0293, 0.0236, 0.0281, 0.0257, 0.0265], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 19:40:37,279 INFO [train.py:873] (3/4) Epoch 9, batch 1000, loss[loss=0.1374, simple_loss=0.1666, pruned_loss=0.05405, over 11171.00 frames. ], tot_loss[loss=0.1443, simple_loss=0.1686, pruned_loss=0.05996, over 1978542.47 frames. ], batch size: 100, lr: 9.01e-03, grad_scale: 8.0 2022-12-07 19:41:11,618 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.79 vs. limit=5.0 2022-12-07 19:41:18,075 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.146e+02 2.101e+02 2.828e+02 3.770e+02 7.982e+02, threshold=5.656e+02, percent-clipped=1.0 2022-12-07 19:41:39,089 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8560, 3.6410, 3.6310, 3.9256, 3.4518, 3.1059, 3.9576, 3.8287], device='cuda:3'), covar=tensor([0.0602, 0.0751, 0.0660, 0.0555, 0.0908, 0.0730, 0.0629, 0.0699], device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0116, 0.0125, 0.0132, 0.0130, 0.0104, 0.0144, 0.0125], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 19:42:05,671 INFO [train.py:873] (3/4) Epoch 9, batch 1100, loss[loss=0.1553, simple_loss=0.1733, pruned_loss=0.06868, over 6945.00 frames. ], tot_loss[loss=0.1439, simple_loss=0.1683, pruned_loss=0.0597, over 2017412.15 frames. ], batch size: 100, lr: 9.00e-03, grad_scale: 8.0 2022-12-07 19:42:46,207 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.208e+02 2.543e+02 3.100e+02 3.894e+02 7.246e+02, threshold=6.200e+02, percent-clipped=7.0 2022-12-07 19:43:32,679 INFO [train.py:873] (3/4) Epoch 9, batch 1200, loss[loss=0.1929, simple_loss=0.1767, pruned_loss=0.1045, over 1279.00 frames. ], tot_loss[loss=0.1443, simple_loss=0.1685, pruned_loss=0.06002, over 1957465.37 frames. ], batch size: 100, lr: 9.00e-03, grad_scale: 8.0 2022-12-07 19:43:49,941 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=61717.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:44:03,907 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=61732.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:44:13,227 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.333e+02 2.189e+02 2.580e+02 3.174e+02 5.497e+02, threshold=5.160e+02, percent-clipped=0.0 2022-12-07 19:44:13,412 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=61743.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:44:32,967 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=61765.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:44:56,522 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=61793.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:44:59,742 INFO [train.py:873] (3/4) Epoch 9, batch 1300, loss[loss=0.1635, simple_loss=0.1765, pruned_loss=0.07522, over 11185.00 frames. ], tot_loss[loss=0.1446, simple_loss=0.1684, pruned_loss=0.06039, over 1939616.01 frames. ], batch size: 100, lr: 8.99e-03, grad_scale: 8.0 2022-12-07 19:45:06,725 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 2022-12-07 19:45:07,360 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=61804.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:45:41,029 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.161e+02 2.302e+02 2.881e+02 3.480e+02 7.451e+02, threshold=5.762e+02, percent-clipped=5.0 2022-12-07 19:45:41,501 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7311, 1.9625, 2.0257, 2.1169, 1.9203, 2.0743, 1.7827, 1.3044], device='cuda:3'), covar=tensor([0.1614, 0.0998, 0.0748, 0.0467, 0.0923, 0.0586, 0.1504, 0.2558], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0068, 0.0055, 0.0056, 0.0087, 0.0066, 0.0091, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2022-12-07 19:46:02,059 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0062, 1.6831, 1.9078, 1.9821, 1.9179, 1.9732, 2.0224, 1.6775], device='cuda:3'), covar=tensor([0.1451, 0.2777, 0.1397, 0.1323, 0.1852, 0.0934, 0.1717, 0.1618], device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0246, 0.0169, 0.0167, 0.0164, 0.0132, 0.0250, 0.0152], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 19:46:28,405 INFO [train.py:873] (3/4) Epoch 9, batch 1400, loss[loss=0.1423, simple_loss=0.1633, pruned_loss=0.06065, over 6933.00 frames. ], tot_loss[loss=0.144, simple_loss=0.168, pruned_loss=0.05999, over 1909318.83 frames. ], batch size: 100, lr: 8.98e-03, grad_scale: 8.0 2022-12-07 19:46:40,259 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9281, 4.5726, 4.2329, 4.4241, 4.5215, 4.7961, 4.8760, 4.8047], device='cuda:3'), covar=tensor([0.0651, 0.0522, 0.1682, 0.2516, 0.0619, 0.0609, 0.0842, 0.0827], device='cuda:3'), in_proj_covar=tensor([0.0340, 0.0240, 0.0402, 0.0514, 0.0296, 0.0378, 0.0373, 0.0332], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 19:46:43,234 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=61914.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:46:43,244 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.9127, 0.9182, 0.8762, 0.9346, 1.0143, 0.5172, 0.7905, 0.8429], device='cuda:3'), covar=tensor([0.0430, 0.0731, 0.0350, 0.0327, 0.0233, 0.0445, 0.0755, 0.0493], device='cuda:3'), in_proj_covar=tensor([0.0022, 0.0023, 0.0023, 0.0021, 0.0023, 0.0033, 0.0023, 0.0023], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2022-12-07 19:47:08,293 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.377e+02 2.184e+02 2.770e+02 3.491e+02 7.127e+02, threshold=5.539e+02, percent-clipped=2.0 2022-12-07 19:47:37,243 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=61975.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:47:56,374 INFO [train.py:873] (3/4) Epoch 9, batch 1500, loss[loss=0.1365, simple_loss=0.1698, pruned_loss=0.05156, over 14511.00 frames. ], tot_loss[loss=0.1434, simple_loss=0.1681, pruned_loss=0.05933, over 2010671.07 frames. ], batch size: 49, lr: 8.98e-03, grad_scale: 8.0 2022-12-07 19:47:57,315 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8786, 1.4142, 3.1987, 2.9099, 3.0830, 3.1874, 2.3645, 3.1932], device='cuda:3'), covar=tensor([0.1174, 0.1418, 0.0137, 0.0346, 0.0263, 0.0153, 0.0493, 0.0162], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0157, 0.0120, 0.0161, 0.0140, 0.0133, 0.0116, 0.0117], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 19:48:36,605 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.50 vs. limit=5.0 2022-12-07 19:48:37,573 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.350e+02 2.345e+02 2.854e+02 3.466e+02 6.884e+02, threshold=5.708e+02, percent-clipped=1.0 2022-12-07 19:48:45,583 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8637, 3.1024, 2.8332, 3.2316, 2.3841, 3.1979, 2.8730, 1.1924], device='cuda:3'), covar=tensor([0.2403, 0.0671, 0.2649, 0.0550, 0.1111, 0.0703, 0.1285, 0.3504], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0068, 0.0056, 0.0056, 0.0088, 0.0067, 0.0092, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2022-12-07 19:49:18,242 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62088.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:49:25,700 INFO [train.py:873] (3/4) Epoch 9, batch 1600, loss[loss=0.1378, simple_loss=0.1455, pruned_loss=0.06512, over 3865.00 frames. ], tot_loss[loss=0.1438, simple_loss=0.1682, pruned_loss=0.05968, over 2018116.39 frames. ], batch size: 100, lr: 8.97e-03, grad_scale: 8.0 2022-12-07 19:49:27,551 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62099.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:49:48,628 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62122.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:50:07,093 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.173e+02 2.069e+02 2.487e+02 3.323e+02 7.707e+02, threshold=4.973e+02, percent-clipped=4.0 2022-12-07 19:50:09,463 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.45 vs. limit=5.0 2022-12-07 19:50:42,589 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62183.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:50:54,729 INFO [train.py:873] (3/4) Epoch 9, batch 1700, loss[loss=0.1611, simple_loss=0.178, pruned_loss=0.07206, over 10327.00 frames. ], tot_loss[loss=0.1437, simple_loss=0.1682, pruned_loss=0.05959, over 2011545.74 frames. ], batch size: 100, lr: 8.96e-03, grad_scale: 8.0 2022-12-07 19:51:32,002 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5207, 1.4251, 3.4387, 1.5157, 3.3953, 3.6172, 2.3652, 3.8656], device='cuda:3'), covar=tensor([0.0236, 0.3178, 0.0475, 0.2428, 0.0672, 0.0354, 0.0935, 0.0163], device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0156, 0.0154, 0.0170, 0.0167, 0.0166, 0.0133, 0.0139], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 19:51:35,608 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.148e+02 2.316e+02 3.145e+02 3.790e+02 7.968e+02, threshold=6.289e+02, percent-clipped=7.0 2022-12-07 19:51:59,356 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62270.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:52:01,962 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62273.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:52:23,524 INFO [train.py:873] (3/4) Epoch 9, batch 1800, loss[loss=0.1392, simple_loss=0.1689, pruned_loss=0.05471, over 14242.00 frames. ], tot_loss[loss=0.1437, simple_loss=0.1682, pruned_loss=0.0596, over 2040924.23 frames. ], batch size: 80, lr: 8.95e-03, grad_scale: 8.0 2022-12-07 19:52:24,843 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.02 vs. limit=5.0 2022-12-07 19:52:56,840 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62334.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:53:04,487 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.227e+02 2.235e+02 2.778e+02 3.404e+02 6.725e+02, threshold=5.556e+02, percent-clipped=1.0 2022-12-07 19:53:44,425 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62388.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:53:47,076 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62391.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:53:49,306 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2022-12-07 19:53:52,235 INFO [train.py:873] (3/4) Epoch 9, batch 1900, loss[loss=0.1246, simple_loss=0.1598, pruned_loss=0.04466, over 14036.00 frames. ], tot_loss[loss=0.1441, simple_loss=0.1681, pruned_loss=0.06002, over 2025901.60 frames. ], batch size: 26, lr: 8.95e-03, grad_scale: 8.0 2022-12-07 19:53:54,355 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62399.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:53:58,972 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.6238, 4.4444, 4.2945, 4.6480, 4.2877, 3.9995, 4.6782, 4.4385], device='cuda:3'), covar=tensor([0.0513, 0.0708, 0.0601, 0.0527, 0.0713, 0.0558, 0.0553, 0.0672], device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0121, 0.0131, 0.0138, 0.0136, 0.0108, 0.0152, 0.0130], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 19:54:26,834 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62436.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:54:33,571 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.440e+02 2.190e+02 2.770e+02 3.608e+02 5.661e+02, threshold=5.540e+02, percent-clipped=1.0 2022-12-07 19:54:37,190 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62447.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:54:41,696 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62452.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:55:04,122 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62478.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:55:20,940 INFO [train.py:873] (3/4) Epoch 9, batch 2000, loss[loss=0.1371, simple_loss=0.1651, pruned_loss=0.05456, over 13935.00 frames. ], tot_loss[loss=0.1448, simple_loss=0.1682, pruned_loss=0.06067, over 1943387.62 frames. ], batch size: 20, lr: 8.94e-03, grad_scale: 8.0 2022-12-07 19:56:01,038 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.163e+02 2.315e+02 2.841e+02 3.698e+02 9.209e+02, threshold=5.682e+02, percent-clipped=7.0 2022-12-07 19:56:03,011 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62545.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:56:03,697 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3992, 2.5862, 2.6833, 2.5464, 2.0048, 2.8687, 2.4482, 1.1766], device='cuda:3'), covar=tensor([0.2011, 0.0978, 0.0712, 0.0820, 0.1412, 0.0644, 0.1527, 0.3557], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0069, 0.0056, 0.0058, 0.0088, 0.0067, 0.0092, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2022-12-07 19:56:24,871 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62570.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:56:37,563 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62585.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:56:47,848 INFO [train.py:873] (3/4) Epoch 9, batch 2100, loss[loss=0.1296, simple_loss=0.1628, pruned_loss=0.0482, over 14476.00 frames. ], tot_loss[loss=0.1438, simple_loss=0.1679, pruned_loss=0.05988, over 1964691.30 frames. ], batch size: 51, lr: 8.93e-03, grad_scale: 8.0 2022-12-07 19:56:56,387 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62606.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:57:07,124 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62618.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:57:16,366 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62629.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:57:28,901 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.193e+02 2.143e+02 2.774e+02 3.519e+02 5.598e+02, threshold=5.548e+02, percent-clipped=0.0 2022-12-07 19:57:31,541 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62646.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:57:48,781 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62666.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:58:15,320 INFO [train.py:873] (3/4) Epoch 9, batch 2200, loss[loss=0.24, simple_loss=0.1966, pruned_loss=0.1417, over 1256.00 frames. ], tot_loss[loss=0.1438, simple_loss=0.1678, pruned_loss=0.05991, over 1933641.35 frames. ], batch size: 100, lr: 8.93e-03, grad_scale: 8.0 2022-12-07 19:58:41,984 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62727.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:58:53,890 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 2022-12-07 19:58:56,803 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.196e+02 2.557e+02 3.266e+02 4.324e+02 1.294e+03, threshold=6.533e+02, percent-clipped=13.0 2022-12-07 19:58:59,498 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62747.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:59:21,851 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62773.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:59:25,966 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62778.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 19:59:29,073 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.32 vs. limit=5.0 2022-12-07 19:59:38,958 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.72 vs. limit=5.0 2022-12-07 19:59:41,695 INFO [train.py:873] (3/4) Epoch 9, batch 2300, loss[loss=0.1478, simple_loss=0.1506, pruned_loss=0.07252, over 3887.00 frames. ], tot_loss[loss=0.1432, simple_loss=0.1673, pruned_loss=0.05952, over 1912383.86 frames. ], batch size: 100, lr: 8.92e-03, grad_scale: 8.0 2022-12-07 20:00:07,218 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62826.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:00:14,366 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62834.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:00:23,176 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.000e+02 2.029e+02 2.536e+02 3.250e+02 8.518e+02, threshold=5.072e+02, percent-clipped=1.0 2022-12-07 20:01:05,988 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2841, 1.4289, 2.5408, 1.4862, 2.5149, 2.4815, 1.8592, 2.5837], device='cuda:3'), covar=tensor([0.0304, 0.2070, 0.0311, 0.1812, 0.0408, 0.0482, 0.1151, 0.0262], device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0156, 0.0155, 0.0170, 0.0167, 0.0166, 0.0134, 0.0138], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 20:01:09,228 INFO [train.py:873] (3/4) Epoch 9, batch 2400, loss[loss=0.1412, simple_loss=0.1698, pruned_loss=0.0563, over 14291.00 frames. ], tot_loss[loss=0.1444, simple_loss=0.168, pruned_loss=0.06033, over 1899867.10 frames. ], batch size: 60, lr: 8.91e-03, grad_scale: 8.0 2022-12-07 20:01:12,687 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62901.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:01:19,005 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2022-12-07 20:01:36,911 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62929.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:01:47,461 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62941.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:01:48,565 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=10.41 vs. limit=5.0 2022-12-07 20:01:49,790 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.047e+02 2.251e+02 2.910e+02 3.795e+02 1.080e+03, threshold=5.819e+02, percent-clipped=6.0 2022-12-07 20:02:18,749 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62977.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:02:28,307 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2022-12-07 20:02:35,718 INFO [train.py:873] (3/4) Epoch 9, batch 2500, loss[loss=0.1444, simple_loss=0.1669, pruned_loss=0.06095, over 13542.00 frames. ], tot_loss[loss=0.1446, simple_loss=0.168, pruned_loss=0.06058, over 1861816.65 frames. ], batch size: 100, lr: 8.90e-03, grad_scale: 8.0 2022-12-07 20:02:58,127 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63022.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:03:06,400 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.59 vs. limit=5.0 2022-12-07 20:03:17,451 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.170e+02 2.289e+02 2.848e+02 3.593e+02 5.714e+02, threshold=5.696e+02, percent-clipped=0.0 2022-12-07 20:03:20,547 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63047.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:04:02,994 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63095.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:04:04,717 INFO [train.py:873] (3/4) Epoch 9, batch 2600, loss[loss=0.1416, simple_loss=0.1686, pruned_loss=0.0573, over 12752.00 frames. ], tot_loss[loss=0.1443, simple_loss=0.1683, pruned_loss=0.06014, over 1988661.06 frames. ], batch size: 100, lr: 8.90e-03, grad_scale: 8.0 2022-12-07 20:04:12,611 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.89 vs. limit=2.0 2022-12-07 20:04:32,676 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63129.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:04:45,881 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.145e+02 2.604e+02 3.223e+02 3.963e+02 1.117e+03, threshold=6.445e+02, percent-clipped=7.0 2022-12-07 20:05:32,672 INFO [train.py:873] (3/4) Epoch 9, batch 2700, loss[loss=0.1595, simple_loss=0.1783, pruned_loss=0.07035, over 10318.00 frames. ], tot_loss[loss=0.1446, simple_loss=0.1683, pruned_loss=0.06043, over 1964784.69 frames. ], batch size: 100, lr: 8.89e-03, grad_scale: 8.0 2022-12-07 20:05:36,740 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63201.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:05:54,609 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63221.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:06:11,831 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63241.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:06:14,513 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.252e+02 2.383e+02 3.002e+02 3.565e+02 6.845e+02, threshold=6.004e+02, percent-clipped=1.0 2022-12-07 20:06:19,157 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63249.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:06:47,897 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63282.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:06:48,829 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.5502, 1.5644, 1.5973, 1.5057, 1.6658, 1.0465, 1.1942, 1.0094], device='cuda:3'), covar=tensor([0.0442, 0.0561, 0.0379, 0.0565, 0.0305, 0.0316, 0.0304, 0.0625], device='cuda:3'), in_proj_covar=tensor([0.0013, 0.0014, 0.0012, 0.0013, 0.0012, 0.0020, 0.0016, 0.0021], device='cuda:3'), out_proj_covar=tensor([9.4880e-05, 1.0193e-04, 8.9125e-05, 9.9296e-05, 9.3032e-05, 1.4049e-04, 1.1759e-04, 1.3436e-04], device='cuda:3') 2022-12-07 20:06:54,041 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63289.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:07:00,710 INFO [train.py:873] (3/4) Epoch 9, batch 2800, loss[loss=0.159, simple_loss=0.1569, pruned_loss=0.0805, over 2640.00 frames. ], tot_loss[loss=0.1441, simple_loss=0.1684, pruned_loss=0.05989, over 1983418.15 frames. ], batch size: 100, lr: 8.88e-03, grad_scale: 8.0 2022-12-07 20:07:21,390 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.58 vs. limit=5.0 2022-12-07 20:07:22,867 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63322.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:07:42,340 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.098e+02 2.099e+02 2.541e+02 3.230e+02 5.695e+02, threshold=5.082e+02, percent-clipped=0.0 2022-12-07 20:08:04,947 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63370.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:08:05,363 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.99 vs. limit=2.0 2022-12-07 20:08:08,176 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1985, 1.2794, 1.1756, 0.9824, 1.1231, 0.8768, 0.6339, 0.8761], device='cuda:3'), covar=tensor([0.0195, 0.0281, 0.0253, 0.0336, 0.0209, 0.0394, 0.0310, 0.0471], device='cuda:3'), in_proj_covar=tensor([0.0013, 0.0014, 0.0012, 0.0013, 0.0013, 0.0020, 0.0016, 0.0021], device='cuda:3'), out_proj_covar=tensor([9.5763e-05, 1.0242e-04, 8.9553e-05, 9.9239e-05, 9.3891e-05, 1.4040e-04, 1.1748e-04, 1.3489e-04], device='cuda:3') 2022-12-07 20:08:13,239 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.66 vs. limit=2.0 2022-12-07 20:08:27,863 INFO [train.py:873] (3/4) Epoch 9, batch 2900, loss[loss=0.1054, simple_loss=0.1486, pruned_loss=0.03105, over 11281.00 frames. ], tot_loss[loss=0.1441, simple_loss=0.1683, pruned_loss=0.05996, over 1983505.71 frames. ], batch size: 14, lr: 8.88e-03, grad_scale: 8.0 2022-12-07 20:08:41,509 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.6481, 4.5985, 4.9340, 4.0474, 4.6754, 5.0495, 1.7496, 4.4546], device='cuda:3'), covar=tensor([0.0220, 0.0246, 0.0338, 0.0425, 0.0281, 0.0120, 0.3101, 0.0270], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0156, 0.0131, 0.0128, 0.0187, 0.0127, 0.0154, 0.0175], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 20:08:48,367 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63420.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:08:50,040 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63422.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:08:55,926 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63429.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:09:09,073 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.322e+02 2.350e+02 2.806e+02 3.593e+02 6.969e+02, threshold=5.612e+02, percent-clipped=3.0 2022-12-07 20:09:37,903 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63477.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:09:37,990 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.7780, 0.7692, 0.6544, 0.7790, 0.8251, 0.1287, 0.6514, 0.8082], device='cuda:3'), covar=tensor([0.0141, 0.0377, 0.0256, 0.0170, 0.0193, 0.0174, 0.0918, 0.0363], device='cuda:3'), in_proj_covar=tensor([0.0022, 0.0023, 0.0022, 0.0022, 0.0023, 0.0032, 0.0023, 0.0024], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:3') 2022-12-07 20:09:41,717 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63481.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:09:43,620 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63483.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:09:55,509 INFO [train.py:873] (3/4) Epoch 9, batch 3000, loss[loss=0.1316, simple_loss=0.1595, pruned_loss=0.05182, over 14406.00 frames. ], tot_loss[loss=0.1447, simple_loss=0.1686, pruned_loss=0.06043, over 1906548.22 frames. ], batch size: 53, lr: 8.87e-03, grad_scale: 8.0 2022-12-07 20:09:55,509 INFO [train.py:896] (3/4) Computing validation loss 2022-12-07 20:10:05,839 INFO [train.py:905] (3/4) Epoch 9, validation: loss=0.124, simple_loss=0.1667, pruned_loss=0.04063, over 857387.00 frames. 2022-12-07 20:10:05,839 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-07 20:10:18,729 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.16 vs. limit=5.0 2022-12-07 20:10:20,154 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0612, 1.3404, 3.3149, 1.4592, 2.9248, 3.2134, 2.2599, 3.4321], device='cuda:3'), covar=tensor([0.0298, 0.3025, 0.0339, 0.2429, 0.1189, 0.0392, 0.0917, 0.0228], device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0156, 0.0155, 0.0171, 0.0168, 0.0165, 0.0134, 0.0138], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 20:10:46,173 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.041e+02 2.483e+02 2.932e+02 3.564e+02 6.010e+02, threshold=5.864e+02, percent-clipped=1.0 2022-12-07 20:10:48,355 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1983, 1.9263, 3.3329, 2.4305, 3.2359, 1.8187, 2.5530, 3.1711], device='cuda:3'), covar=tensor([0.0926, 0.4911, 0.0757, 0.6836, 0.0745, 0.4414, 0.1606, 0.0615], device='cuda:3'), in_proj_covar=tensor([0.0237, 0.0228, 0.0192, 0.0308, 0.0210, 0.0232, 0.0219, 0.0192], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 20:11:08,818 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63569.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:11:12,283 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0715, 3.5380, 2.8289, 4.1346, 3.9760, 3.9943, 3.4699, 2.9186], device='cuda:3'), covar=tensor([0.0574, 0.1252, 0.3614, 0.0475, 0.0711, 0.0977, 0.1163, 0.3509], device='cuda:3'), in_proj_covar=tensor([0.0258, 0.0298, 0.0276, 0.0233, 0.0293, 0.0285, 0.0256, 0.0263], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2022-12-07 20:11:15,576 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63577.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:11:19,403 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.65 vs. limit=2.0 2022-12-07 20:11:23,857 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63586.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:11:33,376 INFO [train.py:873] (3/4) Epoch 9, batch 3100, loss[loss=0.1486, simple_loss=0.1686, pruned_loss=0.06428, over 14244.00 frames. ], tot_loss[loss=0.1434, simple_loss=0.1674, pruned_loss=0.05963, over 1915152.13 frames. ], batch size: 69, lr: 8.86e-03, grad_scale: 8.0 2022-12-07 20:12:03,499 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63630.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:12:15,373 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.211e+02 2.304e+02 2.853e+02 3.296e+02 8.516e+02, threshold=5.706e+02, percent-clipped=4.0 2022-12-07 20:12:18,585 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63647.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:13:02,480 INFO [train.py:873] (3/4) Epoch 9, batch 3200, loss[loss=0.147, simple_loss=0.1526, pruned_loss=0.07072, over 5020.00 frames. ], tot_loss[loss=0.1429, simple_loss=0.1676, pruned_loss=0.05905, over 1977236.68 frames. ], batch size: 100, lr: 8.86e-03, grad_scale: 8.0 2022-12-07 20:13:21,830 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.2344, 1.2286, 1.4183, 0.9289, 0.8873, 1.2628, 0.8363, 1.1424], device='cuda:3'), covar=tensor([0.1427, 0.2363, 0.0627, 0.2106, 0.3005, 0.0751, 0.1485, 0.0906], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0091, 0.0085, 0.0091, 0.0109, 0.0078, 0.0127, 0.0081], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2022-12-07 20:13:35,440 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1697, 1.2210, 1.4299, 0.8618, 0.9044, 1.1698, 0.7771, 1.1269], device='cuda:3'), covar=tensor([0.1254, 0.2549, 0.0615, 0.2158, 0.2603, 0.0726, 0.1897, 0.0857], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0091, 0.0084, 0.0091, 0.0108, 0.0077, 0.0126, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2022-12-07 20:13:43,328 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.200e+02 2.396e+02 2.898e+02 3.837e+02 7.541e+02, threshold=5.796e+02, percent-clipped=6.0 2022-12-07 20:14:11,286 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63776.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:14:13,021 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63778.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:14:17,777 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9888, 2.5974, 5.0711, 3.3543, 4.6537, 2.3291, 3.8939, 4.4076], device='cuda:3'), covar=tensor([0.0425, 0.4523, 0.0322, 0.8196, 0.0524, 0.3708, 0.1066, 0.0353], device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0230, 0.0192, 0.0306, 0.0211, 0.0233, 0.0220, 0.0193], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 20:14:29,971 INFO [train.py:873] (3/4) Epoch 9, batch 3300, loss[loss=0.1509, simple_loss=0.1723, pruned_loss=0.06479, over 14214.00 frames. ], tot_loss[loss=0.1428, simple_loss=0.1677, pruned_loss=0.059, over 2008057.17 frames. ], batch size: 94, lr: 8.85e-03, grad_scale: 8.0 2022-12-07 20:14:41,660 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63810.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:14:52,999 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.3707, 0.9735, 1.3029, 0.8246, 1.0548, 1.4291, 1.0253, 1.0368], device='cuda:3'), covar=tensor([0.0396, 0.0739, 0.0543, 0.0489, 0.1080, 0.0491, 0.0457, 0.1287], device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0186, 0.0132, 0.0124, 0.0126, 0.0132, 0.0108, 0.0137], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0007, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005], device='cuda:3') 2022-12-07 20:15:10,959 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.258e+02 2.144e+02 2.539e+02 3.073e+02 7.001e+02, threshold=5.078e+02, percent-clipped=2.0 2022-12-07 20:15:12,039 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63845.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:15:12,999 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4801, 1.7227, 1.4674, 1.7569, 1.6639, 1.3491, 1.1612, 1.0874], device='cuda:3'), covar=tensor([0.0688, 0.0674, 0.1045, 0.0437, 0.0365, 0.0559, 0.0541, 0.1188], device='cuda:3'), in_proj_covar=tensor([0.0013, 0.0014, 0.0012, 0.0013, 0.0013, 0.0021, 0.0016, 0.0021], device='cuda:3'), out_proj_covar=tensor([9.6263e-05, 1.0307e-04, 9.1159e-05, 9.9126e-05, 9.4823e-05, 1.4425e-04, 1.2069e-04, 1.3739e-04], device='cuda:3') 2022-12-07 20:15:34,965 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63871.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:15:40,221 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63877.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:15:56,935 INFO [train.py:873] (3/4) Epoch 9, batch 3400, loss[loss=0.1645, simple_loss=0.1698, pruned_loss=0.0796, over 7790.00 frames. ], tot_loss[loss=0.1429, simple_loss=0.1674, pruned_loss=0.05918, over 1992929.50 frames. ], batch size: 100, lr: 8.84e-03, grad_scale: 8.0 2022-12-07 20:16:01,304 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.06 vs. limit=2.0 2022-12-07 20:16:05,643 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63906.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:16:14,137 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63916.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:16:16,607 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2599, 1.7501, 2.4454, 1.9780, 2.2738, 1.6275, 1.9552, 2.1228], device='cuda:3'), covar=tensor([0.1536, 0.3091, 0.0325, 0.3480, 0.0901, 0.1991, 0.1026, 0.0939], device='cuda:3'), in_proj_covar=tensor([0.0235, 0.0229, 0.0190, 0.0306, 0.0210, 0.0230, 0.0219, 0.0193], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 20:16:21,653 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63925.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:16:21,692 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63925.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:16:32,911 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8061, 4.7875, 5.2906, 4.3279, 4.9529, 5.4742, 2.1919, 4.7153], device='cuda:3'), covar=tensor([0.0195, 0.0314, 0.0323, 0.0437, 0.0261, 0.0089, 0.2924, 0.0270], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0159, 0.0132, 0.0128, 0.0187, 0.0128, 0.0155, 0.0176], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 20:16:36,691 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63942.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:16:38,551 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.322e+02 2.128e+02 2.612e+02 3.647e+02 5.829e+02, threshold=5.224e+02, percent-clipped=2.0 2022-12-07 20:16:49,962 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2513, 3.5332, 4.1690, 2.9923, 2.4238, 3.5622, 1.9505, 3.6211], device='cuda:3'), covar=tensor([0.1222, 0.0927, 0.0468, 0.2554, 0.2513, 0.1183, 0.4216, 0.1094], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0090, 0.0086, 0.0091, 0.0108, 0.0077, 0.0126, 0.0081], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2022-12-07 20:17:00,678 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0667, 1.1826, 3.2699, 1.4493, 2.9299, 3.1798, 2.2072, 3.3474], device='cuda:3'), covar=tensor([0.0348, 0.3485, 0.0445, 0.2598, 0.1275, 0.0530, 0.1170, 0.0301], device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0153, 0.0153, 0.0169, 0.0166, 0.0163, 0.0132, 0.0136], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 20:17:07,602 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63977.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:17:25,061 INFO [train.py:873] (3/4) Epoch 9, batch 3500, loss[loss=0.1677, simple_loss=0.1531, pruned_loss=0.0911, over 2674.00 frames. ], tot_loss[loss=0.1424, simple_loss=0.167, pruned_loss=0.05892, over 1953707.68 frames. ], batch size: 100, lr: 8.83e-03, grad_scale: 8.0 2022-12-07 20:17:40,404 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.62 vs. limit=2.0 2022-12-07 20:17:48,256 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64023.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:17:49,185 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9193, 2.1208, 2.8644, 2.9399, 2.8982, 2.0868, 2.8761, 2.3271], device='cuda:3'), covar=tensor([0.0284, 0.0571, 0.0429, 0.0278, 0.0285, 0.0870, 0.0253, 0.0685], device='cuda:3'), in_proj_covar=tensor([0.0262, 0.0238, 0.0351, 0.0301, 0.0240, 0.0287, 0.0268, 0.0269], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 20:18:06,279 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.249e+02 2.173e+02 2.574e+02 3.096e+02 7.573e+02, threshold=5.148e+02, percent-clipped=3.0 2022-12-07 20:18:33,635 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64076.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:18:35,486 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64078.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:18:40,722 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64084.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:18:51,458 INFO [train.py:873] (3/4) Epoch 9, batch 3600, loss[loss=0.1666, simple_loss=0.1432, pruned_loss=0.09499, over 2641.00 frames. ], tot_loss[loss=0.1423, simple_loss=0.167, pruned_loss=0.05879, over 1894281.26 frames. ], batch size: 100, lr: 8.83e-03, grad_scale: 8.0 2022-12-07 20:18:56,140 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3062, 3.1104, 2.8059, 3.0073, 3.2051, 3.1799, 3.2741, 3.2587], device='cuda:3'), covar=tensor([0.1020, 0.0658, 0.2450, 0.2850, 0.0833, 0.0979, 0.1225, 0.0979], device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0242, 0.0409, 0.0518, 0.0296, 0.0386, 0.0379, 0.0342], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 20:19:15,029 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64124.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:19:16,705 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64126.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:19:29,058 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7403, 1.8694, 1.5729, 1.4159, 1.4364, 0.7333, 1.5920, 1.4865], device='cuda:3'), covar=tensor([0.1221, 0.1430, 0.0587, 0.1122, 0.1587, 0.0768, 0.0834, 0.1139], device='cuda:3'), in_proj_covar=tensor([0.0023, 0.0023, 0.0023, 0.0023, 0.0024, 0.0034, 0.0023, 0.0025], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001], device='cuda:3') 2022-12-07 20:19:32,556 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.135e+01 2.573e+02 3.171e+02 4.040e+02 8.310e+02, threshold=6.341e+02, percent-clipped=10.0 2022-12-07 20:19:34,510 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1153, 3.0510, 2.9776, 3.2364, 2.7629, 2.8924, 3.2295, 3.1572], device='cuda:3'), covar=tensor([0.0639, 0.0850, 0.0770, 0.0554, 0.1113, 0.0647, 0.0682, 0.0737], device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0117, 0.0128, 0.0131, 0.0129, 0.0104, 0.0146, 0.0127], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 20:19:43,632 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.70 vs. limit=2.0 2022-12-07 20:19:51,879 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64166.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:20:18,992 INFO [train.py:873] (3/4) Epoch 9, batch 3700, loss[loss=0.143, simple_loss=0.1714, pruned_loss=0.05725, over 14250.00 frames. ], tot_loss[loss=0.1433, simple_loss=0.1676, pruned_loss=0.05949, over 1896042.45 frames. ], batch size: 44, lr: 8.82e-03, grad_scale: 8.0 2022-12-07 20:20:22,606 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64201.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:20:43,619 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8778, 1.9537, 2.1327, 1.3540, 1.4924, 2.0380, 1.2052, 1.9194], device='cuda:3'), covar=tensor([0.1041, 0.1955, 0.0723, 0.2342, 0.2684, 0.0793, 0.3594, 0.0892], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0092, 0.0086, 0.0092, 0.0108, 0.0078, 0.0126, 0.0082], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2022-12-07 20:20:44,421 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64225.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:20:46,276 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.3660, 1.3164, 1.3492, 1.2606, 1.2847, 1.0141, 0.8583, 0.9193], device='cuda:3'), covar=tensor([0.0178, 0.0361, 0.0203, 0.0194, 0.0219, 0.0312, 0.0207, 0.0387], device='cuda:3'), in_proj_covar=tensor([0.0012, 0.0014, 0.0012, 0.0013, 0.0012, 0.0020, 0.0016, 0.0020], device='cuda:3'), out_proj_covar=tensor([9.3760e-05, 1.0082e-04, 8.8531e-05, 9.6704e-05, 9.2388e-05, 1.4035e-04, 1.1718e-04, 1.3293e-04], device='cuda:3') 2022-12-07 20:20:50,538 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64232.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:20:51,991 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2022-12-07 20:20:59,023 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64242.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:21:00,575 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.090e+02 2.323e+02 2.858e+02 3.638e+02 5.583e+02, threshold=5.716e+02, percent-clipped=0.0 2022-12-07 20:21:25,160 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64272.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:21:25,985 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64273.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:21:41,384 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64290.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:21:44,024 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64293.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:21:44,787 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6578, 1.4321, 3.6833, 1.6150, 3.6204, 3.8779, 2.7272, 4.0361], device='cuda:3'), covar=tensor([0.0240, 0.3171, 0.0421, 0.2269, 0.0484, 0.0352, 0.0785, 0.0162], device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0156, 0.0156, 0.0170, 0.0167, 0.0165, 0.0134, 0.0139], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 20:21:47,278 INFO [train.py:873] (3/4) Epoch 9, batch 3800, loss[loss=0.1213, simple_loss=0.1585, pruned_loss=0.04201, over 13893.00 frames. ], tot_loss[loss=0.1425, simple_loss=0.1673, pruned_loss=0.05887, over 1909455.29 frames. ], batch size: 20, lr: 8.81e-03, grad_scale: 8.0 2022-12-07 20:22:00,757 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.89 vs. limit=2.0 2022-12-07 20:22:28,598 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.324e+02 2.269e+02 2.644e+02 3.472e+02 7.922e+02, threshold=5.288e+02, percent-clipped=2.0 2022-12-07 20:22:29,749 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64345.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:22:59,875 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64379.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:23:14,389 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1093, 3.8633, 3.8413, 4.1962, 3.7490, 3.3275, 4.2158, 4.1072], device='cuda:3'), covar=tensor([0.0585, 0.0711, 0.0696, 0.0479, 0.0772, 0.0669, 0.0558, 0.0641], device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0118, 0.0130, 0.0133, 0.0130, 0.0106, 0.0148, 0.0129], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 20:23:16,037 INFO [train.py:873] (3/4) Epoch 9, batch 3900, loss[loss=0.1379, simple_loss=0.17, pruned_loss=0.05293, over 14576.00 frames. ], tot_loss[loss=0.1416, simple_loss=0.1665, pruned_loss=0.0584, over 1908018.86 frames. ], batch size: 34, lr: 8.81e-03, grad_scale: 8.0 2022-12-07 20:23:24,310 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64406.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:23:57,837 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.354e+02 2.154e+02 2.432e+02 3.241e+02 6.644e+02, threshold=4.864e+02, percent-clipped=3.0 2022-12-07 20:24:13,079 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.7057, 1.0929, 1.2940, 1.2832, 1.1098, 1.2779, 1.1564, 0.8621], device='cuda:3'), covar=tensor([0.2580, 0.1241, 0.0278, 0.0309, 0.0910, 0.0690, 0.1578, 0.1187], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0071, 0.0056, 0.0060, 0.0088, 0.0067, 0.0092, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2022-12-07 20:24:17,332 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64465.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:24:18,150 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64466.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:24:45,108 INFO [train.py:873] (3/4) Epoch 9, batch 4000, loss[loss=0.1334, simple_loss=0.1564, pruned_loss=0.05522, over 6006.00 frames. ], tot_loss[loss=0.1417, simple_loss=0.1664, pruned_loss=0.05852, over 1917625.37 frames. ], batch size: 100, lr: 8.80e-03, grad_scale: 8.0 2022-12-07 20:24:48,717 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64501.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:25:00,314 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64514.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:25:11,125 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64526.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:25:21,541 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.79 vs. limit=2.0 2022-12-07 20:25:26,921 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.253e+02 2.384e+02 2.847e+02 3.821e+02 7.495e+02, threshold=5.695e+02, percent-clipped=8.0 2022-12-07 20:25:31,228 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 2022-12-07 20:25:31,864 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64549.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:25:36,527 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.23 vs. limit=5.0 2022-12-07 20:25:51,456 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64572.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:26:06,382 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64588.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:26:07,635 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64589.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:26:14,197 INFO [train.py:873] (3/4) Epoch 9, batch 4100, loss[loss=0.1841, simple_loss=0.1776, pruned_loss=0.09527, over 5023.00 frames. ], tot_loss[loss=0.1427, simple_loss=0.1674, pruned_loss=0.05904, over 1890773.44 frames. ], batch size: 100, lr: 8.79e-03, grad_scale: 8.0 2022-12-07 20:26:17,762 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9454, 1.5893, 3.2764, 2.9822, 3.1826, 3.2734, 2.5955, 3.3178], device='cuda:3'), covar=tensor([0.1243, 0.1352, 0.0115, 0.0268, 0.0219, 0.0125, 0.0336, 0.0131], device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0156, 0.0119, 0.0161, 0.0139, 0.0134, 0.0113, 0.0117], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 20:26:34,416 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64620.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:26:55,696 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.345e+01 2.214e+02 2.881e+02 3.617e+02 6.891e+02, threshold=5.761e+02, percent-clipped=1.0 2022-12-07 20:26:59,467 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64648.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:27:01,168 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4135, 2.3798, 2.5241, 2.5289, 2.4527, 2.1401, 1.3937, 2.2055], device='cuda:3'), covar=tensor([0.0460, 0.0470, 0.0438, 0.0326, 0.0406, 0.0997, 0.2440, 0.0399], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0158, 0.0134, 0.0129, 0.0188, 0.0127, 0.0154, 0.0178], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 20:27:01,235 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64650.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:27:16,486 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.11 vs. limit=5.0 2022-12-07 20:27:26,431 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64679.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:27:41,955 INFO [train.py:873] (3/4) Epoch 9, batch 4200, loss[loss=0.1834, simple_loss=0.1837, pruned_loss=0.09149, over 4961.00 frames. ], tot_loss[loss=0.1425, simple_loss=0.1672, pruned_loss=0.05894, over 1906033.42 frames. ], batch size: 100, lr: 8.79e-03, grad_scale: 8.0 2022-12-07 20:27:45,397 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64701.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:27:46,955 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64703.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:27:52,814 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64709.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:27:56,825 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([6.0888, 5.8282, 5.4971, 6.0406, 5.5283, 5.4530, 6.0624, 5.9290], device='cuda:3'), covar=tensor([0.0488, 0.0475, 0.0604, 0.0484, 0.0548, 0.0288, 0.0606, 0.0570], device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0121, 0.0133, 0.0137, 0.0133, 0.0109, 0.0151, 0.0132], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 20:28:07,782 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64727.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:28:15,666 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7952, 1.4958, 3.8470, 1.5825, 3.6252, 3.9046, 2.8425, 4.1288], device='cuda:3'), covar=tensor([0.0205, 0.2933, 0.0330, 0.2238, 0.0532, 0.0316, 0.0671, 0.0148], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0158, 0.0157, 0.0169, 0.0169, 0.0169, 0.0135, 0.0140], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 20:28:23,777 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.263e+02 2.280e+02 2.745e+02 3.613e+02 6.444e+02, threshold=5.491e+02, percent-clipped=3.0 2022-12-07 20:28:23,995 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64745.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:28:30,372 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64752.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:28:40,164 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64764.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:29:09,366 INFO [train.py:873] (3/4) Epoch 9, batch 4300, loss[loss=0.1317, simple_loss=0.1658, pruned_loss=0.0488, over 14550.00 frames. ], tot_loss[loss=0.1429, simple_loss=0.1672, pruned_loss=0.05931, over 1897976.36 frames. ], batch size: 43, lr: 8.78e-03, grad_scale: 8.0 2022-12-07 20:29:17,498 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64806.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:29:23,544 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64813.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:29:30,679 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64821.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:29:43,856 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.66 vs. limit=5.0 2022-12-07 20:29:49,473 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.7593, 0.7096, 0.5446, 0.8078, 0.7306, 0.1454, 0.7175, 0.8048], device='cuda:3'), covar=tensor([0.0482, 0.0597, 0.0358, 0.0218, 0.0193, 0.0236, 0.0941, 0.0541], device='cuda:3'), in_proj_covar=tensor([0.0023, 0.0023, 0.0024, 0.0023, 0.0024, 0.0034, 0.0024, 0.0025], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001], device='cuda:3') 2022-12-07 20:29:51,886 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.265e+02 2.326e+02 2.910e+02 3.508e+02 7.763e+02, threshold=5.820e+02, percent-clipped=3.0 2022-12-07 20:30:09,856 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5975, 1.9519, 3.7047, 2.5344, 3.5370, 1.7971, 2.7000, 3.5316], device='cuda:3'), covar=tensor([0.0716, 0.4605, 0.0398, 0.6254, 0.0569, 0.3886, 0.1382, 0.0418], device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0229, 0.0194, 0.0308, 0.0215, 0.0233, 0.0222, 0.0196], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 20:30:29,541 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64888.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:30:30,482 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4374, 3.1766, 3.0316, 2.1762, 2.9227, 3.1712, 3.3927, 2.8055], device='cuda:3'), covar=tensor([0.0602, 0.1339, 0.1031, 0.1740, 0.0916, 0.0608, 0.0765, 0.1250], device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0182, 0.0130, 0.0123, 0.0124, 0.0131, 0.0107, 0.0132], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0007, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005], device='cuda:3') 2022-12-07 20:30:37,026 INFO [train.py:873] (3/4) Epoch 9, batch 4400, loss[loss=0.1318, simple_loss=0.1643, pruned_loss=0.04966, over 14265.00 frames. ], tot_loss[loss=0.1433, simple_loss=0.1675, pruned_loss=0.05959, over 1939586.09 frames. ], batch size: 31, lr: 8.77e-03, grad_scale: 8.0 2022-12-07 20:31:10,766 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64936.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:31:18,743 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.162e+02 2.178e+02 2.781e+02 3.445e+02 8.866e+02, threshold=5.563e+02, percent-clipped=5.0 2022-12-07 20:31:18,901 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64945.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:31:45,899 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64976.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:32:04,726 INFO [train.py:873] (3/4) Epoch 9, batch 4500, loss[loss=0.1386, simple_loss=0.1639, pruned_loss=0.05664, over 14233.00 frames. ], tot_loss[loss=0.1423, simple_loss=0.1671, pruned_loss=0.0587, over 1954499.71 frames. ], batch size: 35, lr: 8.77e-03, grad_scale: 8.0 2022-12-07 20:32:04,906 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.3635, 1.4188, 1.4140, 1.2066, 1.1252, 1.0047, 0.9222, 0.9053], device='cuda:3'), covar=tensor([0.0271, 0.0228, 0.0179, 0.0183, 0.0252, 0.0387, 0.0250, 0.0735], device='cuda:3'), in_proj_covar=tensor([0.0013, 0.0014, 0.0012, 0.0013, 0.0013, 0.0021, 0.0017, 0.0021], device='cuda:3'), out_proj_covar=tensor([9.8026e-05, 1.0424e-04, 9.2242e-05, 9.9522e-05, 9.6716e-05, 1.4399e-04, 1.2212e-04, 1.3903e-04], device='cuda:3') 2022-12-07 20:32:06,673 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4772, 2.1612, 3.4787, 3.6172, 3.5978, 2.2521, 3.4136, 2.5881], device='cuda:3'), covar=tensor([0.0314, 0.0724, 0.0609, 0.0383, 0.0254, 0.1052, 0.0304, 0.0756], device='cuda:3'), in_proj_covar=tensor([0.0264, 0.0239, 0.0353, 0.0300, 0.0243, 0.0286, 0.0270, 0.0270], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 20:32:11,695 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65001.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:32:14,093 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65004.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:32:43,524 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65037.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 20:32:50,125 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.126e+02 2.276e+02 2.761e+02 3.494e+02 6.954e+02, threshold=5.522e+02, percent-clipped=1.0 2022-12-07 20:32:53,592 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65049.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:32:58,595 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65055.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:33:02,190 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65059.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:33:35,181 INFO [train.py:873] (3/4) Epoch 9, batch 4600, loss[loss=0.1329, simple_loss=0.1726, pruned_loss=0.04663, over 14636.00 frames. ], tot_loss[loss=0.1429, simple_loss=0.1677, pruned_loss=0.05905, over 1968718.84 frames. ], batch size: 22, lr: 8.76e-03, grad_scale: 8.0 2022-12-07 20:33:39,030 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65101.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:33:40,713 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.5732, 1.3078, 2.0860, 1.8573, 1.9407, 2.1124, 1.5882, 2.0261], device='cuda:3'), covar=tensor([0.0533, 0.0830, 0.0141, 0.0305, 0.0356, 0.0143, 0.0401, 0.0214], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0157, 0.0120, 0.0165, 0.0140, 0.0136, 0.0114, 0.0118], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 20:33:45,512 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65108.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:33:52,313 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65116.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:33:56,596 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65121.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:34:17,217 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.207e+01 2.339e+02 3.104e+02 3.874e+02 1.034e+03, threshold=6.208e+02, percent-clipped=6.0 2022-12-07 20:34:20,005 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.47 vs. limit=2.0 2022-12-07 20:34:28,040 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.16 vs. limit=5.0 2022-12-07 20:34:38,809 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65169.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:35:03,850 INFO [train.py:873] (3/4) Epoch 9, batch 4700, loss[loss=0.1523, simple_loss=0.1784, pruned_loss=0.06308, over 14207.00 frames. ], tot_loss[loss=0.1428, simple_loss=0.1675, pruned_loss=0.05905, over 1951847.29 frames. ], batch size: 89, lr: 8.75e-03, grad_scale: 8.0 2022-12-07 20:35:32,324 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.6050, 2.9154, 4.3464, 3.2871, 4.4021, 4.1926, 3.9964, 3.5672], device='cuda:3'), covar=tensor([0.0485, 0.2871, 0.0741, 0.1585, 0.0772, 0.0636, 0.1464, 0.1873], device='cuda:3'), in_proj_covar=tensor([0.0331, 0.0316, 0.0394, 0.0306, 0.0373, 0.0310, 0.0358, 0.0317], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 20:35:42,039 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1004, 2.1917, 2.0058, 2.1895, 1.8303, 2.0757, 2.1554, 2.1263], device='cuda:3'), covar=tensor([0.0933, 0.0954, 0.1129, 0.0772, 0.1265, 0.0821, 0.1055, 0.0875], device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0118, 0.0130, 0.0134, 0.0130, 0.0106, 0.0147, 0.0128], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 20:35:46,990 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.148e+02 2.284e+02 2.894e+02 3.637e+02 5.922e+02, threshold=5.788e+02, percent-clipped=0.0 2022-12-07 20:35:47,151 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65245.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:36:21,928 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65285.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:36:28,432 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65293.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:36:31,082 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.0704, 1.9233, 4.7734, 2.5686, 4.4971, 4.9945, 4.6818, 5.4686], device='cuda:3'), covar=tensor([0.0159, 0.2605, 0.0392, 0.1775, 0.0264, 0.0366, 0.0244, 0.0114], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0157, 0.0157, 0.0168, 0.0167, 0.0169, 0.0134, 0.0139], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 20:36:31,782 INFO [train.py:873] (3/4) Epoch 9, batch 4800, loss[loss=0.1004, simple_loss=0.1347, pruned_loss=0.03308, over 14299.00 frames. ], tot_loss[loss=0.1427, simple_loss=0.1673, pruned_loss=0.059, over 1967594.14 frames. ], batch size: 18, lr: 8.75e-03, grad_scale: 8.0 2022-12-07 20:36:38,348 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65304.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:37:02,582 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65332.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 20:37:13,782 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.537e+02 2.162e+02 2.794e+02 3.355e+02 5.841e+02, threshold=5.587e+02, percent-clipped=1.0 2022-12-07 20:37:15,284 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65346.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:37:20,534 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65352.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:37:20,547 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3349, 3.2186, 3.1429, 3.4520, 3.0060, 2.8907, 3.4497, 3.3429], device='cuda:3'), covar=tensor([0.0716, 0.0824, 0.0907, 0.0586, 0.1092, 0.0730, 0.0676, 0.0704], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0120, 0.0133, 0.0136, 0.0132, 0.0108, 0.0149, 0.0130], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 20:37:26,015 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65359.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:37:58,500 INFO [train.py:873] (3/4) Epoch 9, batch 4900, loss[loss=0.1668, simple_loss=0.1547, pruned_loss=0.0895, over 1284.00 frames. ], tot_loss[loss=0.1432, simple_loss=0.1676, pruned_loss=0.05937, over 1924335.00 frames. ], batch size: 100, lr: 8.74e-03, grad_scale: 8.0 2022-12-07 20:38:02,288 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65401.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:38:07,261 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65407.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:38:08,245 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65408.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:38:10,632 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65411.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:38:40,336 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.955e+01 2.399e+02 2.803e+02 3.609e+02 7.825e+02, threshold=5.606e+02, percent-clipped=2.0 2022-12-07 20:38:43,760 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65449.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:38:49,505 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65456.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:39:01,894 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.11 vs. limit=5.0 2022-12-07 20:39:06,485 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.58 vs. limit=2.0 2022-12-07 20:39:25,267 INFO [train.py:873] (3/4) Epoch 9, batch 5000, loss[loss=0.142, simple_loss=0.1445, pruned_loss=0.06973, over 2674.00 frames. ], tot_loss[loss=0.1439, simple_loss=0.168, pruned_loss=0.05995, over 1912452.04 frames. ], batch size: 100, lr: 8.73e-03, grad_scale: 8.0 2022-12-07 20:39:25,504 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7542, 2.4521, 3.5287, 2.6436, 3.5608, 3.5320, 3.3097, 2.7694], device='cuda:3'), covar=tensor([0.0658, 0.2898, 0.0980, 0.1991, 0.0917, 0.0826, 0.1593, 0.2091], device='cuda:3'), in_proj_covar=tensor([0.0329, 0.0313, 0.0389, 0.0302, 0.0370, 0.0307, 0.0353, 0.0313], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 20:39:56,499 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65532.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 20:40:07,679 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.681e+01 2.367e+02 2.989e+02 3.809e+02 1.039e+03, threshold=5.978e+02, percent-clipped=5.0 2022-12-07 20:40:40,463 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1391, 1.5507, 1.6826, 1.7264, 1.4891, 1.7036, 1.3549, 1.1716], device='cuda:3'), covar=tensor([0.1858, 0.1132, 0.0420, 0.0344, 0.0922, 0.0637, 0.2024, 0.1682], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0071, 0.0057, 0.0059, 0.0089, 0.0067, 0.0093, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2022-12-07 20:40:42,202 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.3145, 5.0238, 4.6308, 4.9320, 4.8724, 5.2009, 5.3303, 5.3163], device='cuda:3'), covar=tensor([0.0942, 0.0395, 0.2299, 0.2394, 0.0767, 0.0778, 0.0867, 0.0782], device='cuda:3'), in_proj_covar=tensor([0.0348, 0.0242, 0.0408, 0.0521, 0.0298, 0.0395, 0.0385, 0.0342], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 20:40:50,544 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65593.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 20:40:53,899 INFO [train.py:873] (3/4) Epoch 9, batch 5100, loss[loss=0.1383, simple_loss=0.165, pruned_loss=0.05582, over 14220.00 frames. ], tot_loss[loss=0.1438, simple_loss=0.1676, pruned_loss=0.05998, over 1859017.46 frames. ], batch size: 60, lr: 8.73e-03, grad_scale: 8.0 2022-12-07 20:41:08,280 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.40 vs. limit=5.0 2022-12-07 20:41:25,525 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65632.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:41:32,999 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65641.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:41:36,452 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.407e+02 2.306e+02 2.958e+02 3.852e+02 6.809e+02, threshold=5.917e+02, percent-clipped=2.0 2022-12-07 20:41:52,315 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4123, 1.7104, 1.4161, 1.3643, 1.2150, 0.9837, 0.8799, 0.8338], device='cuda:3'), covar=tensor([0.0241, 0.0130, 0.0164, 0.0170, 0.0280, 0.0298, 0.0234, 0.0483], device='cuda:3'), in_proj_covar=tensor([0.0013, 0.0014, 0.0012, 0.0013, 0.0013, 0.0021, 0.0017, 0.0021], device='cuda:3'), out_proj_covar=tensor([9.7072e-05, 1.0584e-04, 9.3419e-05, 1.0043e-04, 9.8340e-05, 1.4607e-04, 1.2216e-04, 1.4130e-04], device='cuda:3') 2022-12-07 20:42:07,906 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65680.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:42:11,343 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1815, 1.9999, 2.1091, 2.1862, 2.1505, 2.0430, 2.2450, 1.9093], device='cuda:3'), covar=tensor([0.0807, 0.1282, 0.0662, 0.0713, 0.0874, 0.0776, 0.0839, 0.0755], device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0255, 0.0174, 0.0171, 0.0169, 0.0140, 0.0260, 0.0159], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-07 20:42:13,201 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.3824, 5.1360, 4.6675, 4.9953, 4.9022, 5.2965, 5.3890, 5.2974], device='cuda:3'), covar=tensor([0.0752, 0.0448, 0.1968, 0.2499, 0.0677, 0.0685, 0.0883, 0.1034], device='cuda:3'), in_proj_covar=tensor([0.0346, 0.0242, 0.0406, 0.0523, 0.0297, 0.0395, 0.0384, 0.0341], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 20:42:22,491 INFO [train.py:873] (3/4) Epoch 9, batch 5200, loss[loss=0.2076, simple_loss=0.1703, pruned_loss=0.1224, over 1261.00 frames. ], tot_loss[loss=0.1435, simple_loss=0.1671, pruned_loss=0.05989, over 1795900.30 frames. ], batch size: 100, lr: 8.72e-03, grad_scale: 8.0 2022-12-07 20:42:35,258 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9062, 3.6647, 3.5402, 3.9290, 3.7060, 3.6449, 3.9588, 3.2993], device='cuda:3'), covar=tensor([0.0466, 0.0897, 0.0371, 0.0415, 0.0730, 0.0823, 0.0509, 0.0535], device='cuda:3'), in_proj_covar=tensor([0.0162, 0.0254, 0.0173, 0.0170, 0.0169, 0.0140, 0.0259, 0.0158], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-07 20:42:35,310 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65711.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:43:04,806 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.244e+02 2.172e+02 3.153e+02 3.803e+02 8.139e+02, threshold=6.306e+02, percent-clipped=3.0 2022-12-07 20:43:17,070 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6475, 1.8864, 3.7988, 2.6331, 3.6468, 1.9107, 2.9586, 3.6390], device='cuda:3'), covar=tensor([0.0817, 0.5172, 0.0559, 0.5992, 0.0614, 0.4140, 0.1323, 0.0533], device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0225, 0.0195, 0.0306, 0.0213, 0.0229, 0.0219, 0.0196], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 20:43:17,698 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65759.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:43:32,866 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9872, 1.9675, 2.0665, 2.0736, 2.0420, 1.6731, 1.2909, 1.8061], device='cuda:3'), covar=tensor([0.0506, 0.0439, 0.0462, 0.0327, 0.0450, 0.1392, 0.2082, 0.0416], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0157, 0.0133, 0.0128, 0.0188, 0.0130, 0.0154, 0.0177], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 20:43:50,698 INFO [train.py:873] (3/4) Epoch 9, batch 5300, loss[loss=0.1373, simple_loss=0.1656, pruned_loss=0.05452, over 14139.00 frames. ], tot_loss[loss=0.144, simple_loss=0.1679, pruned_loss=0.06006, over 1920537.63 frames. ], batch size: 84, lr: 8.71e-03, grad_scale: 8.0 2022-12-07 20:44:26,728 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65838.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 20:44:32,252 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.588e+01 2.128e+02 2.955e+02 3.598e+02 1.062e+03, threshold=5.909e+02, percent-clipped=4.0 2022-12-07 20:45:03,930 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1946, 2.4256, 4.2291, 4.3292, 4.2574, 2.5746, 4.2884, 3.3024], device='cuda:3'), covar=tensor([0.0274, 0.0671, 0.0698, 0.0261, 0.0232, 0.1029, 0.0281, 0.0614], device='cuda:3'), in_proj_covar=tensor([0.0264, 0.0237, 0.0350, 0.0298, 0.0241, 0.0283, 0.0270, 0.0265], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 20:45:04,258 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2022-12-07 20:45:09,732 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65888.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 20:45:17,382 INFO [train.py:873] (3/4) Epoch 9, batch 5400, loss[loss=0.2351, simple_loss=0.1919, pruned_loss=0.1392, over 1260.00 frames. ], tot_loss[loss=0.1437, simple_loss=0.1675, pruned_loss=0.05993, over 1860732.74 frames. ], batch size: 100, lr: 8.71e-03, grad_scale: 8.0 2022-12-07 20:45:17,586 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7136, 1.8909, 2.6292, 2.1248, 2.6701, 2.5464, 2.4234, 2.1887], device='cuda:3'), covar=tensor([0.0694, 0.2771, 0.0832, 0.1734, 0.0544, 0.1006, 0.0944, 0.1648], device='cuda:3'), in_proj_covar=tensor([0.0336, 0.0318, 0.0397, 0.0307, 0.0379, 0.0313, 0.0362, 0.0321], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 20:45:19,381 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65899.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 20:45:33,251 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1846, 3.7690, 2.8748, 4.6017, 4.0573, 4.2420, 3.7287, 3.1159], device='cuda:3'), covar=tensor([0.0846, 0.1386, 0.4296, 0.0463, 0.1075, 0.1821, 0.1169, 0.4021], device='cuda:3'), in_proj_covar=tensor([0.0257, 0.0297, 0.0278, 0.0231, 0.0294, 0.0285, 0.0259, 0.0262], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2022-12-07 20:45:56,482 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65941.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:46:00,157 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.009e+02 2.028e+02 2.560e+02 3.349e+02 6.726e+02, threshold=5.119e+02, percent-clipped=1.0 2022-12-07 20:46:38,787 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65989.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:46:45,302 INFO [train.py:873] (3/4) Epoch 9, batch 5500, loss[loss=0.1638, simple_loss=0.1884, pruned_loss=0.06958, over 14269.00 frames. ], tot_loss[loss=0.1427, simple_loss=0.1669, pruned_loss=0.05921, over 1912739.96 frames. ], batch size: 94, lr: 8.70e-03, grad_scale: 8.0 2022-12-07 20:47:00,170 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.73 vs. limit=2.0 2022-12-07 20:47:09,076 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9007, 3.6174, 3.3940, 3.5515, 3.7555, 3.7616, 3.8787, 3.8518], device='cuda:3'), covar=tensor([0.0692, 0.0557, 0.1864, 0.2237, 0.0706, 0.0718, 0.0963, 0.0758], device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0241, 0.0410, 0.0525, 0.0301, 0.0395, 0.0382, 0.0347], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 20:47:27,195 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.123e+02 2.224e+02 2.797e+02 3.647e+02 1.046e+03, threshold=5.594e+02, percent-clipped=7.0 2022-12-07 20:47:58,720 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3920, 1.9618, 3.5519, 2.4797, 3.3576, 1.8379, 2.7924, 3.3378], device='cuda:3'), covar=tensor([0.0670, 0.4448, 0.0450, 0.6055, 0.0610, 0.3758, 0.1278, 0.0495], device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0225, 0.0193, 0.0305, 0.0213, 0.0234, 0.0222, 0.0198], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 20:48:11,088 INFO [train.py:873] (3/4) Epoch 9, batch 5600, loss[loss=0.212, simple_loss=0.1807, pruned_loss=0.1216, over 1215.00 frames. ], tot_loss[loss=0.1427, simple_loss=0.1668, pruned_loss=0.0593, over 1891409.91 frames. ], batch size: 100, lr: 8.69e-03, grad_scale: 8.0 2022-12-07 20:48:52,436 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66144.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:48:53,077 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.414e+02 2.262e+02 2.866e+02 3.556e+02 6.245e+02, threshold=5.731e+02, percent-clipped=4.0 2022-12-07 20:49:07,533 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66160.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 20:49:32,061 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66188.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 20:49:37,654 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66194.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 20:49:38,537 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4099, 2.4111, 1.9776, 2.4788, 2.2476, 2.3832, 2.1687, 2.0308], device='cuda:3'), covar=tensor([0.0817, 0.0780, 0.2363, 0.0625, 0.0923, 0.0923, 0.1351, 0.1595], device='cuda:3'), in_proj_covar=tensor([0.0261, 0.0299, 0.0278, 0.0235, 0.0298, 0.0288, 0.0260, 0.0264], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2022-12-07 20:49:40,171 INFO [train.py:873] (3/4) Epoch 9, batch 5700, loss[loss=0.1442, simple_loss=0.1708, pruned_loss=0.0588, over 11977.00 frames. ], tot_loss[loss=0.1441, simple_loss=0.1681, pruned_loss=0.06012, over 1903390.80 frames. ], batch size: 100, lr: 8.69e-03, grad_scale: 8.0 2022-12-07 20:49:43,196 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1429, 2.0865, 2.3994, 1.6424, 1.7614, 2.3151, 1.3363, 2.1346], device='cuda:3'), covar=tensor([0.1054, 0.1518, 0.0694, 0.2627, 0.2834, 0.0896, 0.4004, 0.1076], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0091, 0.0085, 0.0092, 0.0110, 0.0077, 0.0128, 0.0084], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2022-12-07 20:49:47,585 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66205.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 20:50:01,229 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66221.0, num_to_drop=1, layers_to_drop={3} 2022-12-07 20:50:15,207 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66236.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 20:50:23,973 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.364e+02 2.292e+02 2.851e+02 3.352e+02 5.991e+02, threshold=5.702e+02, percent-clipped=1.0 2022-12-07 20:50:34,209 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66258.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:50:53,371 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1457, 1.8550, 2.1909, 1.2943, 1.8808, 2.1642, 2.2937, 1.9231], device='cuda:3'), covar=tensor([0.1015, 0.0945, 0.1154, 0.2066, 0.1777, 0.0853, 0.0596, 0.2106], device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0184, 0.0132, 0.0126, 0.0127, 0.0134, 0.0110, 0.0137], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0007, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005], device='cuda:3') 2022-12-07 20:50:56,908 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.69 vs. limit=2.0 2022-12-07 20:51:08,647 INFO [train.py:873] (3/4) Epoch 9, batch 5800, loss[loss=0.1575, simple_loss=0.1814, pruned_loss=0.06676, over 14243.00 frames. ], tot_loss[loss=0.1448, simple_loss=0.1683, pruned_loss=0.06062, over 1908176.71 frames. ], batch size: 46, lr: 8.68e-03, grad_scale: 4.0 2022-12-07 20:51:28,616 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66319.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:51:51,563 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.104e+02 2.481e+02 2.990e+02 4.125e+02 8.241e+02, threshold=5.981e+02, percent-clipped=12.0 2022-12-07 20:52:36,917 INFO [train.py:873] (3/4) Epoch 9, batch 5900, loss[loss=0.1562, simple_loss=0.1598, pruned_loss=0.07628, over 5008.00 frames. ], tot_loss[loss=0.1433, simple_loss=0.1677, pruned_loss=0.05946, over 1914979.37 frames. ], batch size: 100, lr: 8.67e-03, grad_scale: 4.0 2022-12-07 20:53:19,921 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.110e+02 2.452e+02 3.007e+02 3.915e+02 6.171e+02, threshold=6.013e+02, percent-clipped=2.0 2022-12-07 20:53:26,910 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4186, 2.3368, 4.4825, 2.9485, 4.2157, 2.2284, 3.3480, 4.0828], device='cuda:3'), covar=tensor([0.0483, 0.4090, 0.0347, 0.8137, 0.0415, 0.3408, 0.1243, 0.0408], device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0223, 0.0193, 0.0303, 0.0212, 0.0232, 0.0222, 0.0199], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 20:53:29,748 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.66 vs. limit=2.0 2022-12-07 20:54:02,043 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66494.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 20:54:04,618 INFO [train.py:873] (3/4) Epoch 9, batch 6000, loss[loss=0.1556, simple_loss=0.1768, pruned_loss=0.06718, over 13545.00 frames. ], tot_loss[loss=0.1417, simple_loss=0.1664, pruned_loss=0.05846, over 1896647.88 frames. ], batch size: 100, lr: 8.67e-03, grad_scale: 8.0 2022-12-07 20:54:04,618 INFO [train.py:896] (3/4) Computing validation loss 2022-12-07 20:54:16,409 INFO [train.py:905] (3/4) Epoch 9, validation: loss=0.1244, simple_loss=0.166, pruned_loss=0.04137, over 857387.00 frames. 2022-12-07 20:54:16,410 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-07 20:54:19,134 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66500.0, num_to_drop=1, layers_to_drop={3} 2022-12-07 20:54:29,156 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0 2022-12-07 20:54:32,912 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66516.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 20:54:37,342 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66521.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 20:54:56,091 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66542.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 20:54:59,453 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.842e+01 2.449e+02 2.904e+02 3.531e+02 7.481e+02, threshold=5.809e+02, percent-clipped=4.0 2022-12-07 20:55:32,154 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66582.0, num_to_drop=1, layers_to_drop={3} 2022-12-07 20:55:45,075 INFO [train.py:873] (3/4) Epoch 9, batch 6100, loss[loss=0.115, simple_loss=0.1527, pruned_loss=0.03865, over 14659.00 frames. ], tot_loss[loss=0.1422, simple_loss=0.1669, pruned_loss=0.05877, over 1930137.96 frames. ], batch size: 33, lr: 8.66e-03, grad_scale: 8.0 2022-12-07 20:55:57,768 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0564, 2.8169, 2.5897, 2.7114, 2.9837, 2.9313, 3.0534, 3.0073], device='cuda:3'), covar=tensor([0.1109, 0.0936, 0.2631, 0.3287, 0.0936, 0.1134, 0.1435, 0.1057], device='cuda:3'), in_proj_covar=tensor([0.0346, 0.0242, 0.0409, 0.0525, 0.0303, 0.0391, 0.0382, 0.0343], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 20:56:00,276 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66614.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:56:26,395 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8460, 0.8000, 0.7083, 0.8274, 0.7318, 0.5821, 0.5036, 0.7089], device='cuda:3'), covar=tensor([0.0103, 0.0101, 0.0099, 0.0089, 0.0129, 0.0268, 0.0171, 0.0254], device='cuda:3'), in_proj_covar=tensor([0.0013, 0.0014, 0.0012, 0.0013, 0.0013, 0.0021, 0.0017, 0.0021], device='cuda:3'), out_proj_covar=tensor([9.7420e-05, 1.0621e-04, 9.4726e-05, 1.0073e-04, 9.8892e-05, 1.4654e-04, 1.2292e-04, 1.4231e-04], device='cuda:3') 2022-12-07 20:56:28,746 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.254e+02 2.301e+02 2.963e+02 3.616e+02 7.389e+02, threshold=5.926e+02, percent-clipped=2.0 2022-12-07 20:57:13,996 INFO [train.py:873] (3/4) Epoch 9, batch 6200, loss[loss=0.1271, simple_loss=0.1576, pruned_loss=0.04833, over 14328.00 frames. ], tot_loss[loss=0.1413, simple_loss=0.1665, pruned_loss=0.05807, over 1980803.91 frames. ], batch size: 60, lr: 8.66e-03, grad_scale: 8.0 2022-12-07 20:57:44,165 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66731.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 20:57:57,360 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.480e+01 2.278e+02 2.813e+02 3.570e+02 7.547e+02, threshold=5.626e+02, percent-clipped=4.0 2022-12-07 20:58:02,700 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0276, 2.8668, 2.5797, 2.7196, 2.9383, 2.9673, 3.0317, 3.0132], device='cuda:3'), covar=tensor([0.1061, 0.0630, 0.2230, 0.2750, 0.0794, 0.0887, 0.1147, 0.0876], device='cuda:3'), in_proj_covar=tensor([0.0347, 0.0241, 0.0406, 0.0522, 0.0304, 0.0387, 0.0377, 0.0342], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 20:58:37,909 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66792.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 20:58:42,096 INFO [train.py:873] (3/4) Epoch 9, batch 6300, loss[loss=0.1019, simple_loss=0.1306, pruned_loss=0.03656, over 14405.00 frames. ], tot_loss[loss=0.1416, simple_loss=0.1664, pruned_loss=0.05835, over 1956221.90 frames. ], batch size: 18, lr: 8.65e-03, grad_scale: 8.0 2022-12-07 20:58:45,148 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66800.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 20:58:59,235 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66816.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 20:59:25,549 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.076e+02 2.178e+02 2.874e+02 3.614e+02 8.081e+02, threshold=5.749e+02, percent-clipped=6.0 2022-12-07 20:59:27,377 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66848.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 20:59:42,265 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66864.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 20:59:53,366 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66877.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 21:00:10,828 INFO [train.py:873] (3/4) Epoch 9, batch 6400, loss[loss=0.1537, simple_loss=0.1747, pruned_loss=0.06637, over 14178.00 frames. ], tot_loss[loss=0.1424, simple_loss=0.1672, pruned_loss=0.05885, over 1977268.98 frames. ], batch size: 84, lr: 8.64e-03, grad_scale: 8.0 2022-12-07 21:00:26,424 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66914.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:00:54,936 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.137e+02 2.352e+02 2.835e+02 3.387e+02 8.466e+02, threshold=5.670e+02, percent-clipped=3.0 2022-12-07 21:00:55,145 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4032, 2.4275, 1.9389, 2.4839, 2.2999, 2.4217, 2.1759, 2.1516], device='cuda:3'), covar=tensor([0.0788, 0.0831, 0.2105, 0.0559, 0.0865, 0.0593, 0.1168, 0.1481], device='cuda:3'), in_proj_covar=tensor([0.0261, 0.0302, 0.0280, 0.0236, 0.0301, 0.0291, 0.0259, 0.0266], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2022-12-07 21:01:04,621 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4871, 4.3093, 4.0951, 4.1462, 4.3239, 4.4789, 4.5692, 4.5489], device='cuda:3'), covar=tensor([0.0996, 0.0518, 0.2163, 0.2783, 0.0684, 0.0743, 0.0990, 0.0774], device='cuda:3'), in_proj_covar=tensor([0.0347, 0.0242, 0.0405, 0.0521, 0.0303, 0.0387, 0.0381, 0.0342], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 21:01:04,717 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66957.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:01:09,014 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66962.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:01:17,051 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8000, 0.7675, 0.7565, 0.7949, 0.6703, 0.4301, 0.4451, 0.6751], device='cuda:3'), covar=tensor([0.0142, 0.0156, 0.0129, 0.0143, 0.0162, 0.0354, 0.0233, 0.0320], device='cuda:3'), in_proj_covar=tensor([0.0013, 0.0014, 0.0012, 0.0013, 0.0013, 0.0021, 0.0017, 0.0021], device='cuda:3'), out_proj_covar=tensor([9.5845e-05, 1.0541e-04, 9.2846e-05, 9.9853e-05, 9.8018e-05, 1.4593e-04, 1.2239e-04, 1.4239e-04], device='cuda:3') 2022-12-07 21:01:40,465 INFO [train.py:873] (3/4) Epoch 9, batch 6500, loss[loss=0.1788, simple_loss=0.1538, pruned_loss=0.1019, over 1249.00 frames. ], tot_loss[loss=0.1416, simple_loss=0.1669, pruned_loss=0.05809, over 1966426.52 frames. ], batch size: 100, lr: 8.64e-03, grad_scale: 8.0 2022-12-07 21:01:53,106 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9248, 2.7891, 3.5269, 2.6123, 2.2577, 3.0771, 1.6055, 3.0253], device='cuda:3'), covar=tensor([0.1556, 0.1284, 0.0867, 0.1603, 0.2455, 0.1117, 0.4756, 0.1599], device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0092, 0.0086, 0.0093, 0.0111, 0.0078, 0.0128, 0.0084], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0005, 0.0003], device='cuda:3') 2022-12-07 21:01:59,344 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67018.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 21:02:24,261 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.174e+02 2.311e+02 2.795e+02 3.453e+02 6.517e+02, threshold=5.590e+02, percent-clipped=2.0 2022-12-07 21:02:54,335 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0833, 3.2719, 3.2705, 2.9822, 2.4395, 3.3983, 3.0309, 1.4790], device='cuda:3'), covar=tensor([0.2292, 0.0729, 0.1294, 0.0944, 0.1148, 0.0403, 0.1361, 0.2946], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0071, 0.0057, 0.0060, 0.0088, 0.0067, 0.0093, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2022-12-07 21:02:57,877 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67083.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:03:01,139 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67087.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 21:03:10,247 INFO [train.py:873] (3/4) Epoch 9, batch 6600, loss[loss=0.1576, simple_loss=0.1774, pruned_loss=0.06894, over 6932.00 frames. ], tot_loss[loss=0.141, simple_loss=0.1663, pruned_loss=0.05784, over 1929225.29 frames. ], batch size: 100, lr: 8.63e-03, grad_scale: 8.0 2022-12-07 21:03:12,062 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9122, 1.1900, 2.0258, 1.2854, 2.0650, 2.0752, 1.6267, 2.1298], device='cuda:3'), covar=tensor([0.0343, 0.1788, 0.0396, 0.1501, 0.0353, 0.0514, 0.1050, 0.0309], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0157, 0.0158, 0.0169, 0.0169, 0.0168, 0.0136, 0.0138], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 21:03:53,366 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67144.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:03:54,921 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.422e+02 2.185e+02 2.883e+02 3.643e+02 7.006e+02, threshold=5.766e+02, percent-clipped=2.0 2022-12-07 21:04:22,359 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67177.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 21:04:38,182 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2588, 2.5436, 2.5497, 2.5638, 2.1581, 2.6204, 2.3444, 1.1803], device='cuda:3'), covar=tensor([0.1895, 0.0730, 0.0755, 0.0552, 0.1168, 0.0638, 0.1223, 0.3079], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0070, 0.0057, 0.0060, 0.0087, 0.0067, 0.0092, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2022-12-07 21:04:40,790 INFO [train.py:873] (3/4) Epoch 9, batch 6700, loss[loss=0.1488, simple_loss=0.1705, pruned_loss=0.0636, over 14275.00 frames. ], tot_loss[loss=0.1413, simple_loss=0.1663, pruned_loss=0.05811, over 1897638.27 frames. ], batch size: 44, lr: 8.62e-03, grad_scale: 8.0 2022-12-07 21:04:45,988 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2022-12-07 21:05:01,843 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8052, 3.6547, 3.2328, 2.5771, 3.1803, 3.5352, 3.9275, 3.1056], device='cuda:3'), covar=tensor([0.0602, 0.1534, 0.1076, 0.1790, 0.0972, 0.0649, 0.0636, 0.1354], device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0182, 0.0131, 0.0124, 0.0125, 0.0135, 0.0111, 0.0134], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0007, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005], device='cuda:3') 2022-12-07 21:05:06,536 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=67225.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 21:05:11,486 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.24 vs. limit=5.0 2022-12-07 21:05:24,549 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.307e+02 2.308e+02 2.851e+02 3.420e+02 6.457e+02, threshold=5.703e+02, percent-clipped=2.0 2022-12-07 21:05:56,542 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5263, 4.2797, 4.1960, 4.5974, 4.1958, 3.7930, 4.6180, 4.5024], device='cuda:3'), covar=tensor([0.0749, 0.0648, 0.0775, 0.0493, 0.0658, 0.0609, 0.0634, 0.0670], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0119, 0.0130, 0.0137, 0.0132, 0.0107, 0.0149, 0.0130], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 21:06:10,531 INFO [train.py:873] (3/4) Epoch 9, batch 6800, loss[loss=0.1102, simple_loss=0.1445, pruned_loss=0.03791, over 14001.00 frames. ], tot_loss[loss=0.1405, simple_loss=0.166, pruned_loss=0.05746, over 1959917.65 frames. ], batch size: 19, lr: 8.62e-03, grad_scale: 8.0 2022-12-07 21:06:25,252 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67313.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 21:06:35,067 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4088, 1.6627, 4.3098, 2.1343, 4.1466, 4.4075, 3.6626, 4.7484], device='cuda:3'), covar=tensor([0.0170, 0.2826, 0.0272, 0.1939, 0.0320, 0.0281, 0.0412, 0.0121], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0159, 0.0159, 0.0172, 0.0172, 0.0171, 0.0137, 0.0140], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 21:06:54,534 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.150e+02 2.355e+02 2.871e+02 3.825e+02 6.310e+02, threshold=5.742e+02, percent-clipped=3.0 2022-12-07 21:07:05,524 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4917, 1.3188, 1.3935, 1.4864, 1.4059, 0.8189, 1.3188, 1.6201], device='cuda:3'), covar=tensor([0.1026, 0.1141, 0.1167, 0.0541, 0.2371, 0.0734, 0.0708, 0.0763], device='cuda:3'), in_proj_covar=tensor([0.0023, 0.0023, 0.0026, 0.0023, 0.0025, 0.0034, 0.0024, 0.0025], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001], device='cuda:3') 2022-12-07 21:07:19,198 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2022-12-07 21:07:31,776 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67387.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 21:07:40,408 INFO [train.py:873] (3/4) Epoch 9, batch 6900, loss[loss=0.1622, simple_loss=0.1803, pruned_loss=0.07207, over 10358.00 frames. ], tot_loss[loss=0.1421, simple_loss=0.1669, pruned_loss=0.05864, over 1960779.84 frames. ], batch size: 100, lr: 8.61e-03, grad_scale: 8.0 2022-12-07 21:08:14,506 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=67435.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 21:08:17,934 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67439.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:08:24,082 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.180e+02 2.447e+02 3.065e+02 3.875e+02 6.583e+02, threshold=6.130e+02, percent-clipped=2.0 2022-12-07 21:09:09,668 INFO [train.py:873] (3/4) Epoch 9, batch 7000, loss[loss=0.1591, simple_loss=0.1735, pruned_loss=0.07238, over 5991.00 frames. ], tot_loss[loss=0.1419, simple_loss=0.1666, pruned_loss=0.05861, over 1952452.46 frames. ], batch size: 100, lr: 8.60e-03, grad_scale: 8.0 2022-12-07 21:09:20,564 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.11 vs. limit=2.0 2022-12-07 21:09:24,644 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67513.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:09:53,897 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.263e+02 2.336e+02 2.835e+02 3.380e+02 6.445e+02, threshold=5.670e+02, percent-clipped=1.0 2022-12-07 21:10:01,523 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1357, 3.9473, 3.9043, 4.2677, 3.8294, 3.4965, 4.2409, 4.1198], device='cuda:3'), covar=tensor([0.0789, 0.0872, 0.0803, 0.0561, 0.0852, 0.0726, 0.0665, 0.0773], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0119, 0.0129, 0.0136, 0.0131, 0.0107, 0.0151, 0.0130], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 21:10:19,390 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67574.0, num_to_drop=1, layers_to_drop={3} 2022-12-07 21:10:40,044 INFO [train.py:873] (3/4) Epoch 9, batch 7100, loss[loss=0.1777, simple_loss=0.1508, pruned_loss=0.1024, over 1258.00 frames. ], tot_loss[loss=0.1418, simple_loss=0.1665, pruned_loss=0.05858, over 1977643.80 frames. ], batch size: 100, lr: 8.60e-03, grad_scale: 8.0 2022-12-07 21:10:50,344 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 2022-12-07 21:10:55,457 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67613.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:10:56,362 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.6252, 2.4914, 2.2448, 2.3634, 2.5431, 2.5051, 2.6039, 2.5645], device='cuda:3'), covar=tensor([0.0989, 0.0763, 0.2304, 0.2489, 0.0985, 0.1024, 0.1213, 0.0946], device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0247, 0.0408, 0.0526, 0.0310, 0.0394, 0.0382, 0.0346], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 21:11:27,148 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.458e+02 2.435e+02 3.463e+02 4.295e+02 1.529e+03, threshold=6.926e+02, percent-clipped=8.0 2022-12-07 21:11:42,151 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=67661.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:11:46,367 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0449, 2.6967, 3.6050, 2.5142, 2.1002, 2.9451, 1.6347, 2.9460], device='cuda:3'), covar=tensor([0.0869, 0.1704, 0.0665, 0.1817, 0.2426, 0.0912, 0.4259, 0.1075], device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0094, 0.0087, 0.0095, 0.0112, 0.0079, 0.0129, 0.0085], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0005, 0.0003], device='cuda:3') 2022-12-07 21:12:00,612 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67680.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:12:07,754 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7929, 2.8046, 2.9295, 2.8528, 2.8522, 2.6957, 1.3111, 2.6276], device='cuda:3'), covar=tensor([0.0362, 0.0394, 0.0422, 0.0388, 0.0374, 0.0691, 0.2748, 0.0307], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0159, 0.0130, 0.0129, 0.0186, 0.0128, 0.0153, 0.0175], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 21:12:11,898 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67692.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:12:16,514 INFO [train.py:873] (3/4) Epoch 9, batch 7200, loss[loss=0.1712, simple_loss=0.1814, pruned_loss=0.08054, over 10352.00 frames. ], tot_loss[loss=0.1415, simple_loss=0.1664, pruned_loss=0.05825, over 1998654.59 frames. ], batch size: 100, lr: 8.59e-03, grad_scale: 8.0 2022-12-07 21:12:16,711 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9858, 3.2698, 4.0611, 2.8865, 2.3113, 3.3181, 2.0215, 3.1485], device='cuda:3'), covar=tensor([0.1453, 0.0936, 0.0697, 0.2908, 0.2589, 0.1212, 0.3935, 0.1775], device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0093, 0.0086, 0.0094, 0.0111, 0.0079, 0.0127, 0.0084], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2022-12-07 21:12:41,590 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.22 vs. limit=5.0 2022-12-07 21:12:45,300 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=11.40 vs. limit=5.0 2022-12-07 21:12:57,777 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67739.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:12:59,668 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67741.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:13:04,117 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.489e+02 2.252e+02 2.773e+02 3.514e+02 7.822e+02, threshold=5.546e+02, percent-clipped=1.0 2022-12-07 21:13:11,058 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67753.0, num_to_drop=1, layers_to_drop={3} 2022-12-07 21:13:15,593 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9279, 1.8778, 1.5407, 1.9601, 1.8607, 1.9846, 1.8163, 1.7356], device='cuda:3'), covar=tensor([0.0921, 0.0723, 0.1806, 0.0593, 0.0674, 0.0537, 0.1155, 0.0906], device='cuda:3'), in_proj_covar=tensor([0.0263, 0.0303, 0.0276, 0.0236, 0.0302, 0.0290, 0.0258, 0.0264], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2022-12-07 21:13:29,723 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.3141, 5.1167, 4.7745, 5.3710, 4.9412, 4.6050, 5.3287, 5.2429], device='cuda:3'), covar=tensor([0.0622, 0.0536, 0.0771, 0.0448, 0.0750, 0.0543, 0.0667, 0.0628], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0119, 0.0128, 0.0135, 0.0131, 0.0106, 0.0150, 0.0129], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 21:13:43,176 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=67787.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:13:52,643 INFO [train.py:873] (3/4) Epoch 9, batch 7300, loss[loss=0.128, simple_loss=0.1539, pruned_loss=0.05107, over 6908.00 frames. ], tot_loss[loss=0.1406, simple_loss=0.1658, pruned_loss=0.05775, over 1966739.60 frames. ], batch size: 100, lr: 8.58e-03, grad_scale: 8.0 2022-12-07 21:14:19,580 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2168, 2.0473, 2.7555, 1.7684, 1.7866, 2.3967, 1.1879, 2.2689], device='cuda:3'), covar=tensor([0.1157, 0.1685, 0.0755, 0.2001, 0.2662, 0.0911, 0.4989, 0.1376], device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0091, 0.0086, 0.0093, 0.0111, 0.0078, 0.0128, 0.0084], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0005, 0.0003], device='cuda:3') 2022-12-07 21:14:39,307 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.094e+02 2.117e+02 2.726e+02 3.356e+02 6.858e+02, threshold=5.453e+02, percent-clipped=1.0 2022-12-07 21:14:56,903 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3898, 3.1594, 2.3616, 3.5868, 3.4085, 3.4307, 3.0650, 2.5499], device='cuda:3'), covar=tensor([0.1187, 0.1665, 0.4300, 0.0614, 0.0965, 0.1196, 0.1358, 0.3974], device='cuda:3'), in_proj_covar=tensor([0.0263, 0.0303, 0.0277, 0.0236, 0.0303, 0.0292, 0.0259, 0.0265], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2022-12-07 21:15:01,853 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67869.0, num_to_drop=1, layers_to_drop={3} 2022-12-07 21:15:28,020 INFO [train.py:873] (3/4) Epoch 9, batch 7400, loss[loss=0.1338, simple_loss=0.1647, pruned_loss=0.05144, over 14209.00 frames. ], tot_loss[loss=0.1416, simple_loss=0.1663, pruned_loss=0.05842, over 1959449.92 frames. ], batch size: 84, lr: 8.58e-03, grad_scale: 8.0 2022-12-07 21:16:15,149 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.327e+02 2.396e+02 2.952e+02 3.564e+02 5.471e+02, threshold=5.904e+02, percent-clipped=1.0 2022-12-07 21:16:47,261 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67981.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:17:02,498 INFO [train.py:873] (3/4) Epoch 9, batch 7500, loss[loss=0.1433, simple_loss=0.1643, pruned_loss=0.0611, over 11948.00 frames. ], tot_loss[loss=0.1422, simple_loss=0.1667, pruned_loss=0.05886, over 1955958.50 frames. ], batch size: 100, lr: 8.57e-03, grad_scale: 8.0 2022-12-07 21:17:38,161 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68036.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:17:42,609 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68042.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:17:44,527 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 5.588e+01 2.266e+02 2.777e+02 3.907e+02 7.467e+02, threshold=5.554e+02, percent-clipped=5.0 2022-12-07 21:17:45,643 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68048.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 21:18:40,274 INFO [train.py:873] (3/4) Epoch 10, batch 0, loss[loss=0.1394, simple_loss=0.1736, pruned_loss=0.05257, over 14158.00 frames. ], tot_loss[loss=0.1394, simple_loss=0.1736, pruned_loss=0.05257, over 14158.00 frames. ], batch size: 84, lr: 8.15e-03, grad_scale: 8.0 2022-12-07 21:18:40,275 INFO [train.py:896] (3/4) Computing validation loss 2022-12-07 21:18:48,290 INFO [train.py:905] (3/4) Epoch 10, validation: loss=0.1297, simple_loss=0.1728, pruned_loss=0.04327, over 857387.00 frames. 2022-12-07 21:18:48,290 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-07 21:19:16,378 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68087.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 21:19:33,572 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2022-12-07 21:19:45,428 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.59 vs. limit=2.0 2022-12-07 21:20:12,823 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.622e+01 2.492e+02 3.120e+02 3.862e+02 1.100e+03, threshold=6.240e+02, percent-clipped=9.0 2022-12-07 21:20:14,334 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68148.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 21:20:24,792 INFO [train.py:873] (3/4) Epoch 10, batch 100, loss[loss=0.1164, simple_loss=0.1551, pruned_loss=0.03883, over 14245.00 frames. ], tot_loss[loss=0.1408, simple_loss=0.1677, pruned_loss=0.05697, over 869151.61 frames. ], batch size: 31, lr: 8.14e-03, grad_scale: 4.0 2022-12-07 21:20:34,381 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68169.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:21:19,974 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68217.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:21:25,443 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.82 vs. limit=5.0 2022-12-07 21:21:27,443 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2022-12-07 21:21:33,919 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 2022-12-07 21:21:48,935 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.438e+02 2.529e+02 3.051e+02 3.928e+02 8.118e+02, threshold=6.102e+02, percent-clipped=5.0 2022-12-07 21:21:55,482 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1184, 2.3100, 2.4424, 2.4633, 2.0282, 2.4765, 2.2288, 1.1767], device='cuda:3'), covar=tensor([0.1570, 0.0831, 0.0560, 0.0491, 0.0982, 0.0475, 0.1234, 0.2647], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0071, 0.0057, 0.0060, 0.0087, 0.0067, 0.0093, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2022-12-07 21:21:59,799 INFO [train.py:873] (3/4) Epoch 10, batch 200, loss[loss=0.1253, simple_loss=0.1285, pruned_loss=0.06105, over 2635.00 frames. ], tot_loss[loss=0.1409, simple_loss=0.1668, pruned_loss=0.05752, over 1307793.37 frames. ], batch size: 100, lr: 8.14e-03, grad_scale: 4.0 2022-12-07 21:22:21,789 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4893, 1.2802, 2.7407, 1.3239, 2.7574, 2.6686, 1.9464, 2.8515], device='cuda:3'), covar=tensor([0.0284, 0.2585, 0.0362, 0.2013, 0.0381, 0.0528, 0.1097, 0.0254], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0159, 0.0159, 0.0171, 0.0171, 0.0171, 0.0137, 0.0140], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 21:23:12,802 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68336.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:23:13,667 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68337.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:23:23,111 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.156e+02 2.072e+02 2.682e+02 3.537e+02 7.107e+02, threshold=5.365e+02, percent-clipped=2.0 2022-12-07 21:23:24,289 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68348.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:23:34,541 INFO [train.py:873] (3/4) Epoch 10, batch 300, loss[loss=0.188, simple_loss=0.1738, pruned_loss=0.1011, over 1308.00 frames. ], tot_loss[loss=0.1398, simple_loss=0.1658, pruned_loss=0.0569, over 1597609.44 frames. ], batch size: 100, lr: 8.13e-03, grad_scale: 4.0 2022-12-07 21:23:57,995 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68384.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:24:10,432 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68396.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:24:13,280 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68399.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:24:15,207 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2022-12-07 21:24:22,575 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1495, 2.9920, 2.9197, 3.2711, 2.7899, 2.8336, 3.2204, 3.1932], device='cuda:3'), covar=tensor([0.0599, 0.0846, 0.0810, 0.0548, 0.0964, 0.0699, 0.0647, 0.0649], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0120, 0.0129, 0.0135, 0.0132, 0.0105, 0.0149, 0.0130], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 21:24:29,912 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3095, 1.8323, 2.3347, 1.9302, 2.3459, 2.1350, 2.0630, 2.0688], device='cuda:3'), covar=tensor([0.0411, 0.1964, 0.0371, 0.0781, 0.0318, 0.0773, 0.0358, 0.0540], device='cuda:3'), in_proj_covar=tensor([0.0336, 0.0318, 0.0399, 0.0303, 0.0371, 0.0313, 0.0361, 0.0318], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 21:24:56,069 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68443.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 21:24:59,359 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.145e+02 2.338e+02 2.858e+02 3.586e+02 7.542e+02, threshold=5.716e+02, percent-clipped=9.0 2022-12-07 21:25:04,237 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2022-12-07 21:25:10,938 INFO [train.py:873] (3/4) Epoch 10, batch 400, loss[loss=0.159, simple_loss=0.158, pruned_loss=0.08004, over 2533.00 frames. ], tot_loss[loss=0.1396, simple_loss=0.1654, pruned_loss=0.05697, over 1759326.91 frames. ], batch size: 100, lr: 8.12e-03, grad_scale: 8.0 2022-12-07 21:25:12,160 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68460.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:25:31,972 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.13 vs. limit=5.0 2022-12-07 21:26:35,457 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.366e+02 2.435e+02 2.815e+02 3.563e+02 7.495e+02, threshold=5.630e+02, percent-clipped=4.0 2022-12-07 21:26:47,613 INFO [train.py:873] (3/4) Epoch 10, batch 500, loss[loss=0.1458, simple_loss=0.1798, pruned_loss=0.05592, over 14088.00 frames. ], tot_loss[loss=0.1408, simple_loss=0.166, pruned_loss=0.05776, over 1836016.75 frames. ], batch size: 29, lr: 8.12e-03, grad_scale: 8.0 2022-12-07 21:27:05,938 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68578.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:28:01,867 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68637.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:28:03,538 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68639.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:28:10,691 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.029e+02 2.141e+02 2.717e+02 3.371e+02 6.429e+02, threshold=5.434e+02, percent-clipped=1.0 2022-12-07 21:28:13,202 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.11 vs. limit=5.0 2022-12-07 21:28:21,923 INFO [train.py:873] (3/4) Epoch 10, batch 600, loss[loss=0.1532, simple_loss=0.1695, pruned_loss=0.06848, over 6905.00 frames. ], tot_loss[loss=0.1397, simple_loss=0.1652, pruned_loss=0.05711, over 1896731.19 frames. ], batch size: 100, lr: 8.11e-03, grad_scale: 8.0 2022-12-07 21:28:46,816 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68685.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:28:46,956 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68685.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:29:41,767 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68743.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 21:29:44,925 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68746.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:29:45,571 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.363e+02 2.288e+02 3.002e+02 3.760e+02 7.437e+02, threshold=6.005e+02, percent-clipped=6.0 2022-12-07 21:29:53,320 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68755.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:29:57,181 INFO [train.py:873] (3/4) Epoch 10, batch 700, loss[loss=0.1491, simple_loss=0.165, pruned_loss=0.0666, over 5946.00 frames. ], tot_loss[loss=0.1394, simple_loss=0.165, pruned_loss=0.0569, over 1971360.69 frames. ], batch size: 100, lr: 8.11e-03, grad_scale: 8.0 2022-12-07 21:29:57,449 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6719, 3.2195, 2.4289, 3.8912, 3.7264, 3.6601, 3.2602, 2.5778], device='cuda:3'), covar=tensor([0.0980, 0.1694, 0.4521, 0.0518, 0.0901, 0.1809, 0.1349, 0.4175], device='cuda:3'), in_proj_covar=tensor([0.0259, 0.0295, 0.0271, 0.0235, 0.0296, 0.0285, 0.0256, 0.0257], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2022-12-07 21:30:27,492 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68791.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 21:31:20,804 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.536e+02 2.251e+02 2.713e+02 3.475e+02 5.744e+02, threshold=5.426e+02, percent-clipped=0.0 2022-12-07 21:31:26,138 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7550, 2.2236, 3.6618, 3.8671, 3.6832, 2.3424, 3.6196, 2.9203], device='cuda:3'), covar=tensor([0.0277, 0.0729, 0.0568, 0.0331, 0.0264, 0.1032, 0.0334, 0.0703], device='cuda:3'), in_proj_covar=tensor([0.0269, 0.0243, 0.0362, 0.0303, 0.0246, 0.0294, 0.0276, 0.0274], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 21:31:32,392 INFO [train.py:873] (3/4) Epoch 10, batch 800, loss[loss=0.1181, simple_loss=0.1558, pruned_loss=0.04017, over 14281.00 frames. ], tot_loss[loss=0.1395, simple_loss=0.1651, pruned_loss=0.05691, over 2018191.63 frames. ], batch size: 31, lr: 8.10e-03, grad_scale: 8.0 2022-12-07 21:32:21,763 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2022-12-07 21:32:43,317 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68934.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:32:50,722 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7275, 1.2336, 1.5343, 1.4646, 1.6295, 0.8964, 1.5846, 1.9348], device='cuda:3'), covar=tensor([0.0739, 0.0973, 0.1092, 0.1218, 0.1026, 0.0761, 0.1009, 0.0863], device='cuda:3'), in_proj_covar=tensor([0.0022, 0.0023, 0.0025, 0.0023, 0.0023, 0.0035, 0.0024, 0.0025], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001], device='cuda:3') 2022-12-07 21:32:55,163 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.186e+02 2.204e+02 2.859e+02 3.964e+02 7.134e+02, threshold=5.717e+02, percent-clipped=4.0 2022-12-07 21:33:06,467 INFO [train.py:873] (3/4) Epoch 10, batch 900, loss[loss=0.1278, simple_loss=0.1667, pruned_loss=0.04449, over 14295.00 frames. ], tot_loss[loss=0.1386, simple_loss=0.1645, pruned_loss=0.05632, over 1936624.34 frames. ], batch size: 28, lr: 8.09e-03, grad_scale: 8.0 2022-12-07 21:33:13,802 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.71 vs. limit=2.0 2022-12-07 21:33:59,230 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8744, 2.0889, 2.8450, 2.9613, 2.7712, 2.0943, 2.8395, 2.3058], device='cuda:3'), covar=tensor([0.0287, 0.0615, 0.0454, 0.0314, 0.0330, 0.0889, 0.0298, 0.0661], device='cuda:3'), in_proj_covar=tensor([0.0265, 0.0240, 0.0359, 0.0302, 0.0245, 0.0291, 0.0274, 0.0273], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 21:34:24,737 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69041.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:34:26,766 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69043.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:34:30,091 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.213e+02 2.392e+02 2.888e+02 3.362e+02 7.513e+02, threshold=5.776e+02, percent-clipped=4.0 2022-12-07 21:34:37,791 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69055.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:34:41,174 INFO [train.py:873] (3/4) Epoch 10, batch 1000, loss[loss=0.1326, simple_loss=0.1487, pruned_loss=0.0582, over 5958.00 frames. ], tot_loss[loss=0.1384, simple_loss=0.1645, pruned_loss=0.05612, over 1983847.00 frames. ], batch size: 100, lr: 8.09e-03, grad_scale: 8.0 2022-12-07 21:35:22,733 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69103.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:35:23,656 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69104.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:35:35,154 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1181, 3.0876, 3.8930, 2.5864, 2.2802, 3.1005, 1.6362, 3.1002], device='cuda:3'), covar=tensor([0.2102, 0.1040, 0.0568, 0.2241, 0.2659, 0.1051, 0.4690, 0.1338], device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0092, 0.0086, 0.0093, 0.0113, 0.0079, 0.0127, 0.0084], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0005, 0.0003], device='cuda:3') 2022-12-07 21:36:04,000 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.114e+02 2.200e+02 2.692e+02 3.189e+02 5.615e+02, threshold=5.384e+02, percent-clipped=0.0 2022-12-07 21:36:15,497 INFO [train.py:873] (3/4) Epoch 10, batch 1100, loss[loss=0.1581, simple_loss=0.1796, pruned_loss=0.06829, over 14171.00 frames. ], tot_loss[loss=0.1381, simple_loss=0.1644, pruned_loss=0.05592, over 2001297.32 frames. ], batch size: 99, lr: 8.08e-03, grad_scale: 8.0 2022-12-07 21:36:18,777 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69162.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:36:23,244 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.5465, 1.1107, 2.0177, 1.8460, 1.9401, 2.0809, 1.4569, 2.0577], device='cuda:3'), covar=tensor([0.0673, 0.1072, 0.0149, 0.0334, 0.0428, 0.0156, 0.0455, 0.0203], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0159, 0.0123, 0.0166, 0.0143, 0.0137, 0.0117, 0.0118], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 21:36:24,397 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69168.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:37:16,939 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69223.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:37:19,220 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69225.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:37:22,812 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69229.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:37:27,495 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69234.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:37:37,844 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69245.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 21:37:39,426 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.224e+02 2.182e+02 2.689e+02 3.231e+02 7.876e+02, threshold=5.378e+02, percent-clipped=5.0 2022-12-07 21:37:51,123 INFO [train.py:873] (3/4) Epoch 10, batch 1200, loss[loss=0.1431, simple_loss=0.1727, pruned_loss=0.05671, over 13990.00 frames. ], tot_loss[loss=0.1373, simple_loss=0.164, pruned_loss=0.05531, over 1957573.30 frames. ], batch size: 22, lr: 8.08e-03, grad_scale: 8.0 2022-12-07 21:38:04,610 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69273.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:38:13,070 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69282.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:38:16,840 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69286.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:38:36,098 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69306.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 21:38:36,410 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.42 vs. limit=5.0 2022-12-07 21:38:49,567 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69320.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:39:02,113 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69334.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 21:39:04,230 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.3106, 4.8345, 4.7506, 5.2970, 4.9883, 4.5122, 5.2995, 4.2106], device='cuda:3'), covar=tensor([0.0350, 0.0972, 0.0332, 0.0413, 0.0708, 0.0465, 0.0477, 0.0567], device='cuda:3'), in_proj_covar=tensor([0.0157, 0.0255, 0.0175, 0.0171, 0.0167, 0.0136, 0.0257, 0.0155], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-07 21:39:08,998 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69341.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:39:14,523 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.241e+02 2.336e+02 2.846e+02 3.621e+02 9.526e+02, threshold=5.693e+02, percent-clipped=8.0 2022-12-07 21:39:25,828 INFO [train.py:873] (3/4) Epoch 10, batch 1300, loss[loss=0.162, simple_loss=0.1871, pruned_loss=0.06842, over 14292.00 frames. ], tot_loss[loss=0.1386, simple_loss=0.1648, pruned_loss=0.05617, over 1948108.76 frames. ], batch size: 76, lr: 8.07e-03, grad_scale: 8.0 2022-12-07 21:39:34,585 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69368.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:39:47,002 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69381.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:39:54,596 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69389.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:39:59,633 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69394.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:40:04,016 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69399.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:40:32,796 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69429.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:40:39,086 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2022-12-07 21:40:41,511 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1489, 1.8514, 2.1842, 2.3226, 1.8938, 1.8692, 2.2414, 2.1139], device='cuda:3'), covar=tensor([0.0179, 0.0398, 0.0198, 0.0149, 0.0289, 0.0477, 0.0215, 0.0249], device='cuda:3'), in_proj_covar=tensor([0.0270, 0.0242, 0.0363, 0.0305, 0.0248, 0.0292, 0.0277, 0.0274], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 21:40:49,782 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.321e+02 2.414e+02 2.899e+02 3.828e+02 8.841e+02, threshold=5.798e+02, percent-clipped=2.0 2022-12-07 21:40:57,564 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69455.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:41:01,161 INFO [train.py:873] (3/4) Epoch 10, batch 1400, loss[loss=0.1458, simple_loss=0.133, pruned_loss=0.07928, over 1282.00 frames. ], tot_loss[loss=0.1383, simple_loss=0.1641, pruned_loss=0.05627, over 1887953.16 frames. ], batch size: 100, lr: 8.07e-03, grad_scale: 8.0 2022-12-07 21:41:40,538 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2022-12-07 21:41:40,830 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 2022-12-07 21:41:42,425 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.95 vs. limit=2.0 2022-12-07 21:41:57,157 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69518.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:42:02,615 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69524.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:42:24,550 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.449e+02 2.299e+02 2.933e+02 3.622e+02 8.157e+02, threshold=5.865e+02, percent-clipped=2.0 2022-12-07 21:42:26,331 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.18 vs. limit=2.0 2022-12-07 21:42:35,974 INFO [train.py:873] (3/4) Epoch 10, batch 1500, loss[loss=0.1349, simple_loss=0.1697, pruned_loss=0.05001, over 14511.00 frames. ], tot_loss[loss=0.1395, simple_loss=0.1647, pruned_loss=0.0571, over 1934214.88 frames. ], batch size: 49, lr: 8.06e-03, grad_scale: 8.0 2022-12-07 21:42:57,191 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69581.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:42:57,281 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0455, 3.2329, 3.0561, 3.0740, 2.4004, 3.4854, 2.9861, 1.4449], device='cuda:3'), covar=tensor([0.2112, 0.0801, 0.1458, 0.1029, 0.1158, 0.0327, 0.1647, 0.2960], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0073, 0.0058, 0.0061, 0.0089, 0.0068, 0.0093, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0006], device='cuda:3') 2022-12-07 21:43:15,918 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69601.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 21:43:42,229 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69629.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 21:43:48,066 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7413, 3.5240, 3.2391, 3.3469, 3.6320, 3.6313, 3.6852, 3.6847], device='cuda:3'), covar=tensor([0.0816, 0.0621, 0.2018, 0.2804, 0.0708, 0.0794, 0.1120, 0.0872], device='cuda:3'), in_proj_covar=tensor([0.0347, 0.0245, 0.0410, 0.0527, 0.0307, 0.0402, 0.0380, 0.0341], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 21:43:59,277 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.210e+02 2.142e+02 2.757e+02 3.564e+02 6.285e+02, threshold=5.515e+02, percent-clipped=1.0 2022-12-07 21:44:10,746 INFO [train.py:873] (3/4) Epoch 10, batch 1600, loss[loss=0.1416, simple_loss=0.1694, pruned_loss=0.05687, over 14276.00 frames. ], tot_loss[loss=0.1384, simple_loss=0.1646, pruned_loss=0.05611, over 1971561.95 frames. ], batch size: 76, lr: 8.05e-03, grad_scale: 8.0 2022-12-07 21:44:24,880 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69674.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 21:44:26,601 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69676.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:44:48,105 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69699.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:45:11,257 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69724.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:45:21,962 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69735.0, num_to_drop=1, layers_to_drop={3} 2022-12-07 21:45:33,108 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.146e+01 2.291e+02 2.751e+02 3.642e+02 1.145e+03, threshold=5.503e+02, percent-clipped=2.0 2022-12-07 21:45:33,206 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69747.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:45:36,271 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69750.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:45:45,009 INFO [train.py:873] (3/4) Epoch 10, batch 1700, loss[loss=0.1549, simple_loss=0.1474, pruned_loss=0.08126, over 2607.00 frames. ], tot_loss[loss=0.1382, simple_loss=0.1644, pruned_loss=0.05598, over 1916382.19 frames. ], batch size: 100, lr: 8.05e-03, grad_scale: 8.0 2022-12-07 21:46:32,572 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8322, 1.9055, 4.4284, 2.3848, 4.3203, 4.7563, 4.5166, 5.2599], device='cuda:3'), covar=tensor([0.0186, 0.2970, 0.0411, 0.2074, 0.0308, 0.0334, 0.0241, 0.0133], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0160, 0.0158, 0.0168, 0.0171, 0.0169, 0.0135, 0.0140], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 21:46:34,880 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2022-12-07 21:46:40,876 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69818.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:46:46,200 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69824.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:47:08,313 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.377e+02 2.221e+02 2.761e+02 3.390e+02 7.818e+02, threshold=5.522e+02, percent-clipped=1.0 2022-12-07 21:47:19,445 INFO [train.py:873] (3/4) Epoch 10, batch 1800, loss[loss=0.09315, simple_loss=0.1269, pruned_loss=0.02969, over 10776.00 frames. ], tot_loss[loss=0.1376, simple_loss=0.164, pruned_loss=0.05553, over 1948509.76 frames. ], batch size: 13, lr: 8.04e-03, grad_scale: 8.0 2022-12-07 21:47:26,413 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69866.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:47:26,891 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.65 vs. limit=2.0 2022-12-07 21:47:32,303 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69872.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:47:40,907 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69881.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:47:59,700 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69901.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 21:48:26,766 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69929.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:48:26,884 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69929.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 21:48:43,307 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.161e+02 2.361e+02 2.916e+02 3.771e+02 6.970e+02, threshold=5.833e+02, percent-clipped=3.0 2022-12-07 21:48:45,619 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69949.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 21:48:55,139 INFO [train.py:873] (3/4) Epoch 10, batch 1900, loss[loss=0.1225, simple_loss=0.1562, pruned_loss=0.04441, over 14283.00 frames. ], tot_loss[loss=0.1391, simple_loss=0.1649, pruned_loss=0.05667, over 1928725.64 frames. ], batch size: 46, lr: 8.04e-03, grad_scale: 8.0 2022-12-07 21:49:01,011 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4076, 1.8967, 2.3163, 2.4622, 2.1711, 1.8829, 2.3825, 2.0747], device='cuda:3'), covar=tensor([0.0222, 0.0517, 0.0266, 0.0254, 0.0335, 0.0622, 0.0252, 0.0383], device='cuda:3'), in_proj_covar=tensor([0.0269, 0.0240, 0.0360, 0.0303, 0.0248, 0.0292, 0.0276, 0.0273], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 21:49:11,059 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69976.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:49:11,844 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69977.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:50:02,130 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70024.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:50:02,246 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70024.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:50:08,124 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70030.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 21:50:24,079 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.053e+02 2.126e+02 2.650e+02 3.300e+02 6.481e+02, threshold=5.300e+02, percent-clipped=2.0 2022-12-07 21:50:27,008 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70050.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:50:35,179 INFO [train.py:873] (3/4) Epoch 10, batch 2000, loss[loss=0.141, simple_loss=0.17, pruned_loss=0.05597, over 13916.00 frames. ], tot_loss[loss=0.1381, simple_loss=0.1645, pruned_loss=0.05585, over 1970015.31 frames. ], batch size: 23, lr: 8.03e-03, grad_scale: 8.0 2022-12-07 21:50:48,050 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70072.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:50:48,217 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6828, 1.9747, 3.8277, 2.6071, 3.6984, 1.9437, 2.7239, 3.6800], device='cuda:3'), covar=tensor([0.0740, 0.4200, 0.0457, 0.6672, 0.0502, 0.3408, 0.1454, 0.0451], device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0220, 0.0192, 0.0298, 0.0214, 0.0225, 0.0219, 0.0201], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 21:50:49,895 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0499, 3.8113, 3.7968, 4.1529, 3.6356, 3.3835, 4.1346, 4.0217], device='cuda:3'), covar=tensor([0.0666, 0.1019, 0.0769, 0.0570, 0.0880, 0.0778, 0.0665, 0.0665], device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0122, 0.0132, 0.0138, 0.0135, 0.0111, 0.0153, 0.0132], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 21:51:12,862 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70098.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:51:24,485 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70110.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:51:27,298 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.5333, 1.1351, 2.0434, 1.8341, 1.9389, 2.0750, 1.4693, 2.0528], device='cuda:3'), covar=tensor([0.0670, 0.1114, 0.0150, 0.0391, 0.0435, 0.0182, 0.0405, 0.0211], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0158, 0.0122, 0.0166, 0.0142, 0.0134, 0.0114, 0.0117], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 21:51:52,010 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.54 vs. limit=5.0 2022-12-07 21:51:59,346 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.415e+02 2.238e+02 2.810e+02 3.599e+02 8.591e+02, threshold=5.620e+02, percent-clipped=6.0 2022-12-07 21:52:10,763 INFO [train.py:873] (3/4) Epoch 10, batch 2100, loss[loss=0.1499, simple_loss=0.1657, pruned_loss=0.06699, over 5953.00 frames. ], tot_loss[loss=0.1381, simple_loss=0.1643, pruned_loss=0.05599, over 1914591.86 frames. ], batch size: 100, lr: 8.03e-03, grad_scale: 16.0 2022-12-07 21:52:22,413 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70171.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:52:29,784 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.7153, 0.7630, 0.6220, 0.7898, 0.8050, 0.2675, 0.5837, 0.7717], device='cuda:3'), covar=tensor([0.0216, 0.0306, 0.0258, 0.0295, 0.0272, 0.0247, 0.0421, 0.0602], device='cuda:3'), in_proj_covar=tensor([0.0023, 0.0024, 0.0026, 0.0023, 0.0025, 0.0035, 0.0024, 0.0026], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001], device='cuda:3') 2022-12-07 21:52:40,097 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.4927, 5.3356, 5.0204, 5.5600, 5.0173, 5.0488, 5.5520, 5.4065], device='cuda:3'), covar=tensor([0.0570, 0.0553, 0.0782, 0.0479, 0.0678, 0.0420, 0.0537, 0.0584], device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0121, 0.0131, 0.0138, 0.0132, 0.0110, 0.0151, 0.0130], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 21:52:54,316 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.19 vs. limit=5.0 2022-12-07 21:53:17,196 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9848, 1.9768, 1.6021, 1.9897, 1.8927, 1.9766, 1.7510, 1.7841], device='cuda:3'), covar=tensor([0.0551, 0.0566, 0.1480, 0.0476, 0.0581, 0.0424, 0.1207, 0.0684], device='cuda:3'), in_proj_covar=tensor([0.0263, 0.0299, 0.0272, 0.0243, 0.0298, 0.0291, 0.0259, 0.0261], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2022-12-07 21:53:34,938 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.579e+01 2.218e+02 2.863e+02 3.700e+02 6.629e+02, threshold=5.727e+02, percent-clipped=4.0 2022-12-07 21:53:43,571 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9442, 1.6696, 1.9985, 1.7550, 2.0695, 1.7450, 1.6907, 1.9524], device='cuda:3'), covar=tensor([0.0410, 0.1087, 0.0213, 0.0295, 0.0272, 0.0635, 0.0177, 0.0277], device='cuda:3'), in_proj_covar=tensor([0.0340, 0.0322, 0.0399, 0.0309, 0.0380, 0.0316, 0.0364, 0.0319], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 21:53:45,272 INFO [train.py:873] (3/4) Epoch 10, batch 2200, loss[loss=0.1206, simple_loss=0.1505, pruned_loss=0.04535, over 13919.00 frames. ], tot_loss[loss=0.1395, simple_loss=0.1649, pruned_loss=0.05702, over 1924173.43 frames. ], batch size: 23, lr: 8.02e-03, grad_scale: 8.0 2022-12-07 21:53:46,702 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70260.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:54:41,419 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.53 vs. limit=2.0 2022-12-07 21:54:43,875 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70321.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:54:52,792 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70330.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 21:55:09,312 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.038e+02 2.253e+02 2.794e+02 3.378e+02 9.827e+02, threshold=5.588e+02, percent-clipped=3.0 2022-12-07 21:55:11,698 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1656, 0.9951, 1.1234, 1.0714, 1.2890, 0.7259, 1.0644, 1.1256], device='cuda:3'), covar=tensor([0.1037, 0.1086, 0.0634, 0.0957, 0.0609, 0.1032, 0.0791, 0.0747], device='cuda:3'), in_proj_covar=tensor([0.0023, 0.0024, 0.0026, 0.0023, 0.0025, 0.0035, 0.0024, 0.0026], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001], device='cuda:3') 2022-12-07 21:55:19,409 INFO [train.py:873] (3/4) Epoch 10, batch 2300, loss[loss=0.1186, simple_loss=0.1359, pruned_loss=0.05059, over 2573.00 frames. ], tot_loss[loss=0.1385, simple_loss=0.1642, pruned_loss=0.05647, over 1863612.26 frames. ], batch size: 100, lr: 8.01e-03, grad_scale: 8.0 2022-12-07 21:55:37,481 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70378.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 21:55:52,145 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1702, 1.8923, 2.2267, 2.3774, 1.9684, 1.9367, 2.2597, 2.1862], device='cuda:3'), covar=tensor([0.0161, 0.0342, 0.0175, 0.0134, 0.0226, 0.0383, 0.0169, 0.0155], device='cuda:3'), in_proj_covar=tensor([0.0269, 0.0240, 0.0360, 0.0304, 0.0247, 0.0292, 0.0278, 0.0273], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 21:56:43,395 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.243e+01 2.133e+02 2.592e+02 3.520e+02 8.171e+02, threshold=5.183e+02, percent-clipped=4.0 2022-12-07 21:56:53,787 INFO [train.py:873] (3/4) Epoch 10, batch 2400, loss[loss=0.1281, simple_loss=0.1559, pruned_loss=0.05016, over 14275.00 frames. ], tot_loss[loss=0.1391, simple_loss=0.1648, pruned_loss=0.05671, over 1949482.46 frames. ], batch size: 63, lr: 8.01e-03, grad_scale: 8.0 2022-12-07 21:57:00,689 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70466.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:57:17,008 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2220, 3.0239, 2.7252, 2.8896, 3.1392, 3.1052, 3.1882, 3.1536], device='cuda:3'), covar=tensor([0.0976, 0.0788, 0.2235, 0.2741, 0.0819, 0.1040, 0.1236, 0.1024], device='cuda:3'), in_proj_covar=tensor([0.0357, 0.0246, 0.0414, 0.0530, 0.0311, 0.0405, 0.0385, 0.0352], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 21:57:20,109 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0107, 1.3789, 3.1407, 1.5610, 3.2720, 3.1383, 2.2671, 3.2959], device='cuda:3'), covar=tensor([0.0236, 0.2785, 0.0327, 0.1972, 0.0294, 0.0376, 0.0799, 0.0203], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0159, 0.0155, 0.0168, 0.0168, 0.0168, 0.0135, 0.0137], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 21:57:29,830 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=7.98 vs. limit=5.0 2022-12-07 21:57:45,492 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2979, 1.4004, 3.4140, 1.5964, 3.1788, 3.4479, 2.4386, 3.6076], device='cuda:3'), covar=tensor([0.0236, 0.2980, 0.0378, 0.2234, 0.0835, 0.0365, 0.0847, 0.0175], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0158, 0.0155, 0.0167, 0.0168, 0.0167, 0.0134, 0.0137], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 21:57:53,799 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6867, 1.9997, 2.1208, 2.0986, 1.8129, 2.1300, 1.8365, 1.2893], device='cuda:3'), covar=tensor([0.1425, 0.0835, 0.0576, 0.0499, 0.1276, 0.0568, 0.1557, 0.2401], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0073, 0.0058, 0.0061, 0.0088, 0.0068, 0.0093, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0006], device='cuda:3') 2022-12-07 21:58:15,968 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2022-12-07 21:58:18,046 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.315e+02 2.395e+02 3.014e+02 3.583e+02 8.568e+02, threshold=6.028e+02, percent-clipped=4.0 2022-12-07 21:58:19,382 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3726, 2.7661, 4.0093, 3.0441, 4.1329, 4.0525, 3.9640, 3.4722], device='cuda:3'), covar=tensor([0.0841, 0.3034, 0.1306, 0.2139, 0.1146, 0.0832, 0.1635, 0.1803], device='cuda:3'), in_proj_covar=tensor([0.0342, 0.0322, 0.0400, 0.0309, 0.0379, 0.0317, 0.0366, 0.0319], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 21:58:28,704 INFO [train.py:873] (3/4) Epoch 10, batch 2500, loss[loss=0.153, simple_loss=0.1763, pruned_loss=0.06483, over 14420.00 frames. ], tot_loss[loss=0.1391, simple_loss=0.1647, pruned_loss=0.05677, over 1967315.81 frames. ], batch size: 73, lr: 8.00e-03, grad_scale: 8.0 2022-12-07 21:58:38,892 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5480, 4.5781, 4.8486, 4.0991, 4.7101, 4.9397, 1.7603, 4.3403], device='cuda:3'), covar=tensor([0.0246, 0.0290, 0.0412, 0.0433, 0.0300, 0.0185, 0.3300, 0.0294], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0160, 0.0133, 0.0133, 0.0192, 0.0131, 0.0154, 0.0178], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 21:59:06,649 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5235, 3.5912, 3.7446, 3.4472, 3.5593, 3.5101, 1.4581, 3.3711], device='cuda:3'), covar=tensor([0.0303, 0.0290, 0.0329, 0.0394, 0.0348, 0.0431, 0.3092, 0.0274], device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0158, 0.0132, 0.0132, 0.0191, 0.0130, 0.0154, 0.0177], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 21:59:21,255 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.5069, 1.4790, 1.3967, 1.5826, 1.5416, 1.2641, 1.2043, 1.0043], device='cuda:3'), covar=tensor([0.0291, 0.0491, 0.0446, 0.0383, 0.0284, 0.0329, 0.0280, 0.0718], device='cuda:3'), in_proj_covar=tensor([0.0013, 0.0015, 0.0012, 0.0013, 0.0013, 0.0021, 0.0017, 0.0022], device='cuda:3'), out_proj_covar=tensor([1.0149e-04, 1.1055e-04, 9.6217e-05, 1.0479e-04, 1.0191e-04, 1.5127e-04, 1.2840e-04, 1.4668e-04], device='cuda:3') 2022-12-07 21:59:21,276 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6462, 1.8171, 3.7718, 2.4175, 3.5837, 1.6150, 2.8338, 3.5248], device='cuda:3'), covar=tensor([0.0840, 0.5311, 0.0605, 0.7365, 0.0693, 0.5038, 0.1595, 0.0630], device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0224, 0.0194, 0.0303, 0.0216, 0.0228, 0.0221, 0.0207], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 21:59:22,823 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70616.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 21:59:53,711 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.258e+02 2.220e+02 2.772e+02 3.507e+02 5.637e+02, threshold=5.543e+02, percent-clipped=0.0 2022-12-07 22:00:03,811 INFO [train.py:873] (3/4) Epoch 10, batch 2600, loss[loss=0.1411, simple_loss=0.1724, pruned_loss=0.05488, over 14239.00 frames. ], tot_loss[loss=0.1379, simple_loss=0.1637, pruned_loss=0.05606, over 1949174.73 frames. ], batch size: 94, lr: 8.00e-03, grad_scale: 8.0 2022-12-07 22:00:35,642 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9516, 3.0427, 3.1599, 2.9884, 3.0518, 2.8997, 1.4843, 2.8190], device='cuda:3'), covar=tensor([0.0411, 0.0388, 0.0399, 0.0412, 0.0407, 0.0715, 0.2950, 0.0323], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0160, 0.0133, 0.0134, 0.0193, 0.0131, 0.0156, 0.0178], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 22:00:51,914 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2022-12-07 22:01:27,326 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.690e+02 2.257e+02 2.909e+02 3.862e+02 1.406e+03, threshold=5.817e+02, percent-clipped=9.0 2022-12-07 22:01:38,027 INFO [train.py:873] (3/4) Epoch 10, batch 2700, loss[loss=0.1368, simple_loss=0.1515, pruned_loss=0.06109, over 4924.00 frames. ], tot_loss[loss=0.1381, simple_loss=0.1639, pruned_loss=0.05612, over 1921578.27 frames. ], batch size: 100, lr: 7.99e-03, grad_scale: 8.0 2022-12-07 22:01:45,121 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70766.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:02:30,603 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70814.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:02:48,678 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7594, 2.7407, 2.6228, 2.8855, 2.4149, 2.5697, 2.8359, 2.8163], device='cuda:3'), covar=tensor([0.0837, 0.0993, 0.0846, 0.0675, 0.1170, 0.0767, 0.0863, 0.0864], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0120, 0.0128, 0.0136, 0.0130, 0.0109, 0.0150, 0.0130], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 22:02:51,423 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5098, 4.3085, 3.9255, 4.1373, 4.2628, 4.4194, 4.4537, 4.4815], device='cuda:3'), covar=tensor([0.0765, 0.0398, 0.1976, 0.2714, 0.0645, 0.0720, 0.0917, 0.0785], device='cuda:3'), in_proj_covar=tensor([0.0363, 0.0249, 0.0417, 0.0542, 0.0314, 0.0409, 0.0389, 0.0354], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 22:03:03,174 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.053e+02 2.251e+02 2.633e+02 3.197e+02 4.846e+02, threshold=5.267e+02, percent-clipped=0.0 2022-12-07 22:03:13,061 INFO [train.py:873] (3/4) Epoch 10, batch 2800, loss[loss=0.1293, simple_loss=0.1415, pruned_loss=0.05855, over 2640.00 frames. ], tot_loss[loss=0.1374, simple_loss=0.1637, pruned_loss=0.05553, over 1895552.52 frames. ], batch size: 100, lr: 7.99e-03, grad_scale: 8.0 2022-12-07 22:03:17,034 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9973, 2.1826, 4.0737, 2.6834, 3.8782, 2.0286, 3.1869, 3.8582], device='cuda:3'), covar=tensor([0.0631, 0.4578, 0.0396, 0.6941, 0.0723, 0.3818, 0.1229, 0.0400], device='cuda:3'), in_proj_covar=tensor([0.0243, 0.0223, 0.0195, 0.0300, 0.0214, 0.0226, 0.0218, 0.0205], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 22:04:08,185 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70916.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:04:27,777 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.4757, 5.2682, 4.8082, 5.0593, 4.8854, 5.4042, 5.4252, 5.4655], device='cuda:3'), covar=tensor([0.0605, 0.0341, 0.1531, 0.2036, 0.0568, 0.0579, 0.0678, 0.0704], device='cuda:3'), in_proj_covar=tensor([0.0366, 0.0251, 0.0423, 0.0546, 0.0317, 0.0412, 0.0391, 0.0356], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 22:04:38,138 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.263e+02 2.119e+02 2.647e+02 3.407e+02 7.312e+02, threshold=5.294e+02, percent-clipped=4.0 2022-12-07 22:04:48,431 INFO [train.py:873] (3/4) Epoch 10, batch 2900, loss[loss=0.1441, simple_loss=0.1672, pruned_loss=0.06051, over 13861.00 frames. ], tot_loss[loss=0.1367, simple_loss=0.1633, pruned_loss=0.05501, over 1950909.91 frames. ], batch size: 20, lr: 7.98e-03, grad_scale: 8.0 2022-12-07 22:04:52,051 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8458, 4.6103, 4.4260, 4.9326, 4.3946, 4.2515, 4.9255, 4.7468], device='cuda:3'), covar=tensor([0.0596, 0.0606, 0.0615, 0.0484, 0.0647, 0.0573, 0.0509, 0.0671], device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0118, 0.0127, 0.0134, 0.0129, 0.0108, 0.0149, 0.0129], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 22:04:52,940 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70964.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:04:55,169 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8018, 2.8276, 2.9573, 2.8906, 2.8508, 2.6671, 1.4932, 2.5783], device='cuda:3'), covar=tensor([0.0381, 0.0404, 0.0397, 0.0412, 0.0391, 0.0789, 0.2640, 0.0328], device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0158, 0.0132, 0.0133, 0.0190, 0.0130, 0.0154, 0.0176], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 22:05:19,742 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70992.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:05:34,996 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.2234, 4.2583, 4.5629, 3.6722, 4.3827, 4.5597, 1.6976, 4.0808], device='cuda:3'), covar=tensor([0.0272, 0.0337, 0.0395, 0.0742, 0.0342, 0.0215, 0.3321, 0.0265], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0160, 0.0133, 0.0135, 0.0192, 0.0131, 0.0156, 0.0178], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 22:05:55,220 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=71030.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:06:12,367 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.268e+02 2.365e+02 2.871e+02 3.545e+02 4.870e+02, threshold=5.742e+02, percent-clipped=0.0 2022-12-07 22:06:17,168 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71053.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:06:19,526 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 2022-12-07 22:06:22,389 INFO [train.py:873] (3/4) Epoch 10, batch 3000, loss[loss=0.1165, simple_loss=0.1503, pruned_loss=0.04132, over 14406.00 frames. ], tot_loss[loss=0.1362, simple_loss=0.1629, pruned_loss=0.05478, over 1894338.80 frames. ], batch size: 53, lr: 7.98e-03, grad_scale: 8.0 2022-12-07 22:06:22,389 INFO [train.py:896] (3/4) Computing validation loss 2022-12-07 22:06:39,846 INFO [train.py:905] (3/4) Epoch 10, validation: loss=0.1251, simple_loss=0.1669, pruned_loss=0.04162, over 857387.00 frames. 2022-12-07 22:06:39,847 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-07 22:07:10,059 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71091.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:07:20,768 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7283, 2.3977, 3.6197, 3.8633, 3.7434, 2.2741, 3.6924, 2.8431], device='cuda:3'), covar=tensor([0.0326, 0.0732, 0.0728, 0.0364, 0.0350, 0.1309, 0.0318, 0.0864], device='cuda:3'), in_proj_covar=tensor([0.0272, 0.0242, 0.0364, 0.0310, 0.0251, 0.0294, 0.0277, 0.0273], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 22:07:54,984 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.46 vs. limit=2.0 2022-12-07 22:08:04,099 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.301e+01 2.256e+02 2.878e+02 3.587e+02 6.087e+02, threshold=5.757e+02, percent-clipped=3.0 2022-12-07 22:08:09,883 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.36 vs. limit=5.0 2022-12-07 22:08:11,251 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4460, 1.4156, 3.6174, 1.6027, 3.3591, 3.6815, 2.6618, 3.7973], device='cuda:3'), covar=tensor([0.0361, 0.3997, 0.0480, 0.3014, 0.0858, 0.0513, 0.0984, 0.0337], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0157, 0.0156, 0.0168, 0.0169, 0.0169, 0.0134, 0.0138], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 22:08:14,732 INFO [train.py:873] (3/4) Epoch 10, batch 3100, loss[loss=0.1394, simple_loss=0.1688, pruned_loss=0.05499, over 14205.00 frames. ], tot_loss[loss=0.1367, simple_loss=0.163, pruned_loss=0.0552, over 1903889.39 frames. ], batch size: 89, lr: 7.97e-03, grad_scale: 8.0 2022-12-07 22:09:14,427 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 2022-12-07 22:09:38,773 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.252e+02 2.264e+02 3.028e+02 3.832e+02 6.668e+02, threshold=6.056e+02, percent-clipped=4.0 2022-12-07 22:09:43,858 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5073, 3.5756, 3.7548, 3.3901, 3.6090, 3.5281, 1.3904, 3.3973], device='cuda:3'), covar=tensor([0.0329, 0.0384, 0.0366, 0.0469, 0.0340, 0.0481, 0.3277, 0.0276], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0158, 0.0132, 0.0134, 0.0191, 0.0130, 0.0153, 0.0176], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 22:09:49,303 INFO [train.py:873] (3/4) Epoch 10, batch 3200, loss[loss=0.1548, simple_loss=0.1779, pruned_loss=0.06581, over 14405.00 frames. ], tot_loss[loss=0.1362, simple_loss=0.163, pruned_loss=0.0547, over 1965882.65 frames. ], batch size: 53, lr: 7.96e-03, grad_scale: 8.0 2022-12-07 22:10:57,659 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.90 vs. limit=2.0 2022-12-07 22:11:13,458 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.358e+02 2.362e+02 2.792e+02 3.327e+02 7.945e+02, threshold=5.585e+02, percent-clipped=1.0 2022-12-07 22:11:13,609 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71348.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:11:17,603 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.78 vs. limit=2.0 2022-12-07 22:11:24,236 INFO [train.py:873] (3/4) Epoch 10, batch 3300, loss[loss=0.1095, simple_loss=0.1436, pruned_loss=0.03769, over 13977.00 frames. ], tot_loss[loss=0.1362, simple_loss=0.163, pruned_loss=0.05464, over 1985545.08 frames. ], batch size: 19, lr: 7.96e-03, grad_scale: 8.0 2022-12-07 22:11:26,325 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.6542, 2.6457, 2.7317, 2.7241, 2.6690, 2.3670, 1.2676, 2.4027], device='cuda:3'), covar=tensor([0.0409, 0.0418, 0.0391, 0.0332, 0.0392, 0.1094, 0.2723, 0.0334], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0159, 0.0133, 0.0134, 0.0192, 0.0131, 0.0154, 0.0177], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 22:11:49,885 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71386.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:11:57,056 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1922, 3.6531, 2.8921, 4.4188, 4.2180, 4.1950, 3.6873, 2.9995], device='cuda:3'), covar=tensor([0.0823, 0.1248, 0.3857, 0.0498, 0.0727, 0.1589, 0.1260, 0.3439], device='cuda:3'), in_proj_covar=tensor([0.0264, 0.0299, 0.0270, 0.0246, 0.0302, 0.0292, 0.0262, 0.0257], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2022-12-07 22:12:06,457 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7190, 3.5313, 3.1687, 3.3993, 3.6090, 3.6300, 3.6893, 3.7013], device='cuda:3'), covar=tensor([0.0949, 0.0620, 0.2364, 0.2516, 0.0805, 0.0876, 0.0987, 0.0836], device='cuda:3'), in_proj_covar=tensor([0.0358, 0.0245, 0.0415, 0.0533, 0.0313, 0.0401, 0.0382, 0.0346], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 22:12:49,402 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.381e+02 2.167e+02 2.744e+02 3.343e+02 6.323e+02, threshold=5.487e+02, percent-clipped=2.0 2022-12-07 22:12:59,545 INFO [train.py:873] (3/4) Epoch 10, batch 3400, loss[loss=0.118, simple_loss=0.1509, pruned_loss=0.0426, over 11214.00 frames. ], tot_loss[loss=0.1369, simple_loss=0.1632, pruned_loss=0.05536, over 1930948.03 frames. ], batch size: 100, lr: 7.95e-03, grad_scale: 8.0 2022-12-07 22:13:39,410 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2760, 3.4876, 3.2849, 3.2791, 2.5967, 3.5871, 3.3289, 1.6699], device='cuda:3'), covar=tensor([0.2006, 0.0694, 0.1534, 0.0691, 0.0987, 0.0681, 0.1428, 0.2576], device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0072, 0.0059, 0.0061, 0.0087, 0.0070, 0.0092, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0006], device='cuda:3') 2022-12-07 22:13:40,374 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9557, 3.8375, 3.3689, 2.6161, 3.4754, 3.6711, 4.1890, 3.4891], device='cuda:3'), covar=tensor([0.0469, 0.1468, 0.0908, 0.1417, 0.0636, 0.0649, 0.0515, 0.1021], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0183, 0.0133, 0.0125, 0.0127, 0.0139, 0.0114, 0.0136], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0007, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005], device='cuda:3') 2022-12-07 22:14:22,586 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.256e+02 2.276e+02 2.825e+02 3.448e+02 6.629e+02, threshold=5.650e+02, percent-clipped=5.0 2022-12-07 22:14:28,365 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.0466, 1.0243, 0.8515, 0.9166, 1.0004, 0.4206, 0.8820, 1.0753], device='cuda:3'), covar=tensor([0.0338, 0.0600, 0.0369, 0.0459, 0.0296, 0.0419, 0.0969, 0.0472], device='cuda:3'), in_proj_covar=tensor([0.0024, 0.0024, 0.0026, 0.0023, 0.0025, 0.0035, 0.0025, 0.0026], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001], device='cuda:3') 2022-12-07 22:14:33,335 INFO [train.py:873] (3/4) Epoch 10, batch 3500, loss[loss=0.1215, simple_loss=0.1498, pruned_loss=0.04658, over 6926.00 frames. ], tot_loss[loss=0.1374, simple_loss=0.1636, pruned_loss=0.05558, over 1949927.69 frames. ], batch size: 100, lr: 7.95e-03, grad_scale: 8.0 2022-12-07 22:15:27,752 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=71616.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:15:47,782 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3199, 1.4347, 3.4879, 1.4870, 3.2824, 3.4455, 2.5780, 3.6928], device='cuda:3'), covar=tensor([0.0271, 0.3198, 0.0380, 0.2608, 0.0793, 0.0392, 0.0807, 0.0223], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0159, 0.0158, 0.0169, 0.0172, 0.0171, 0.0134, 0.0141], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 22:15:57,136 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.705e+02 2.664e+02 3.016e+02 4.035e+02 8.090e+02, threshold=6.032e+02, percent-clipped=4.0 2022-12-07 22:15:57,346 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=71648.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:15:58,132 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5244, 4.5392, 4.3210, 4.2379, 4.4898, 4.9240, 1.7468, 4.2340], device='cuda:3'), covar=tensor([0.0530, 0.0576, 0.1064, 0.0875, 0.0807, 0.0236, 0.4768, 0.0563], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0161, 0.0133, 0.0136, 0.0193, 0.0131, 0.0155, 0.0178], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 22:16:07,712 INFO [train.py:873] (3/4) Epoch 10, batch 3600, loss[loss=0.1103, simple_loss=0.1471, pruned_loss=0.03675, over 14266.00 frames. ], tot_loss[loss=0.1378, simple_loss=0.1638, pruned_loss=0.05584, over 1907807.67 frames. ], batch size: 25, lr: 7.94e-03, grad_scale: 8.0 2022-12-07 22:16:09,841 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3713, 3.1481, 2.9042, 1.9893, 2.9515, 3.2041, 3.4397, 2.6192], device='cuda:3'), covar=tensor([0.0691, 0.1252, 0.1133, 0.2073, 0.0853, 0.0595, 0.0772, 0.1536], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0182, 0.0133, 0.0125, 0.0128, 0.0137, 0.0114, 0.0136], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0007, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005], device='cuda:3') 2022-12-07 22:16:25,262 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71677.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:16:33,244 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=71686.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:16:42,662 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=71696.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:17:18,947 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=71734.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:17:31,781 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.595e+01 2.418e+02 2.819e+02 3.742e+02 8.382e+02, threshold=5.637e+02, percent-clipped=6.0 2022-12-07 22:17:42,534 INFO [train.py:873] (3/4) Epoch 10, batch 3700, loss[loss=0.1292, simple_loss=0.1634, pruned_loss=0.04754, over 14281.00 frames. ], tot_loss[loss=0.1383, simple_loss=0.1643, pruned_loss=0.05614, over 1943151.18 frames. ], batch size: 60, lr: 7.94e-03, grad_scale: 8.0 2022-12-07 22:19:05,971 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.143e+02 2.175e+02 2.639e+02 3.434e+02 5.313e+02, threshold=5.277e+02, percent-clipped=0.0 2022-12-07 22:19:16,071 INFO [train.py:873] (3/4) Epoch 10, batch 3800, loss[loss=0.1887, simple_loss=0.1718, pruned_loss=0.1028, over 1225.00 frames. ], tot_loss[loss=0.1373, simple_loss=0.1637, pruned_loss=0.05544, over 1930117.65 frames. ], batch size: 100, lr: 7.93e-03, grad_scale: 8.0 2022-12-07 22:20:12,753 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=71919.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:20:38,535 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.2481, 4.0150, 3.8644, 4.3234, 3.8258, 3.3590, 4.3117, 4.2212], device='cuda:3'), covar=tensor([0.0632, 0.0883, 0.0809, 0.0495, 0.0765, 0.0788, 0.0653, 0.0576], device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0121, 0.0132, 0.0139, 0.0133, 0.0111, 0.0155, 0.0131], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 22:20:40,197 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.242e+02 2.292e+02 2.855e+02 3.626e+02 5.894e+02, threshold=5.709e+02, percent-clipped=2.0 2022-12-07 22:20:49,794 INFO [train.py:873] (3/4) Epoch 10, batch 3900, loss[loss=0.1315, simple_loss=0.167, pruned_loss=0.04801, over 14601.00 frames. ], tot_loss[loss=0.136, simple_loss=0.1629, pruned_loss=0.05448, over 1962848.98 frames. ], batch size: 22, lr: 7.93e-03, grad_scale: 4.0 2022-12-07 22:21:01,557 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71972.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:21:09,381 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71980.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:21:33,999 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5489, 1.9024, 2.5421, 2.0643, 2.5381, 2.3072, 2.2698, 2.1679], device='cuda:3'), covar=tensor([0.0570, 0.2236, 0.0637, 0.1436, 0.0494, 0.0975, 0.0804, 0.1221], device='cuda:3'), in_proj_covar=tensor([0.0333, 0.0312, 0.0388, 0.0300, 0.0369, 0.0309, 0.0357, 0.0312], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 22:22:03,863 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4060, 0.9807, 1.2813, 0.8455, 1.0758, 1.3938, 1.1210, 1.1296], device='cuda:3'), covar=tensor([0.0363, 0.0747, 0.0544, 0.0461, 0.0830, 0.0592, 0.0302, 0.1122], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0185, 0.0133, 0.0126, 0.0128, 0.0139, 0.0114, 0.0136], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0007, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0006], device='cuda:3') 2022-12-07 22:22:14,706 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.192e+02 2.367e+02 2.840e+02 3.454e+02 7.877e+02, threshold=5.680e+02, percent-clipped=2.0 2022-12-07 22:22:23,954 INFO [train.py:873] (3/4) Epoch 10, batch 4000, loss[loss=0.1429, simple_loss=0.1523, pruned_loss=0.06678, over 6968.00 frames. ], tot_loss[loss=0.1352, simple_loss=0.1626, pruned_loss=0.05391, over 1985676.94 frames. ], batch size: 100, lr: 7.92e-03, grad_scale: 8.0 2022-12-07 22:23:11,703 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8468, 1.7679, 3.1217, 2.2360, 3.0004, 1.7090, 2.3947, 2.9485], device='cuda:3'), covar=tensor([0.0815, 0.3989, 0.0394, 0.5016, 0.0714, 0.3258, 0.1223, 0.0450], device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0220, 0.0196, 0.0299, 0.0214, 0.0223, 0.0218, 0.0202], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 22:23:29,094 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72129.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:23:48,082 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.453e+01 2.051e+02 2.497e+02 3.120e+02 6.082e+02, threshold=4.994e+02, percent-clipped=1.0 2022-12-07 22:23:57,372 INFO [train.py:873] (3/4) Epoch 10, batch 4100, loss[loss=0.1262, simple_loss=0.1545, pruned_loss=0.04897, over 13536.00 frames. ], tot_loss[loss=0.1352, simple_loss=0.1626, pruned_loss=0.05392, over 2054170.96 frames. ], batch size: 100, lr: 7.91e-03, grad_scale: 4.0 2022-12-07 22:24:26,530 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=72190.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:24:41,373 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.6121, 4.7405, 5.1689, 4.0283, 4.9151, 5.2522, 1.9803, 4.5268], device='cuda:3'), covar=tensor([0.0255, 0.0255, 0.0339, 0.0452, 0.0275, 0.0135, 0.3209, 0.0275], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0159, 0.0134, 0.0133, 0.0191, 0.0130, 0.0155, 0.0178], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 22:24:54,701 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6564, 1.6526, 1.6264, 1.3612, 1.4806, 1.2269, 0.6491, 0.9373], device='cuda:3'), covar=tensor([0.0143, 0.0150, 0.0169, 0.0246, 0.0185, 0.0279, 0.0242, 0.0383], device='cuda:3'), in_proj_covar=tensor([0.0014, 0.0015, 0.0013, 0.0014, 0.0014, 0.0022, 0.0018, 0.0023], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2022-12-07 22:25:22,726 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.251e+02 2.564e+02 3.159e+02 3.830e+02 6.918e+02, threshold=6.317e+02, percent-clipped=10.0 2022-12-07 22:25:31,387 INFO [train.py:873] (3/4) Epoch 10, batch 4200, loss[loss=0.1847, simple_loss=0.1886, pruned_loss=0.09037, over 4961.00 frames. ], tot_loss[loss=0.1345, simple_loss=0.1624, pruned_loss=0.05331, over 2026187.96 frames. ], batch size: 100, lr: 7.91e-03, grad_scale: 4.0 2022-12-07 22:25:41,292 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.78 vs. limit=5.0 2022-12-07 22:25:44,354 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72272.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:25:46,891 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=72275.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:26:27,848 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=72320.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:26:33,201 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72326.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:26:55,037 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.448e+02 2.250e+02 2.962e+02 3.836e+02 8.638e+02, threshold=5.923e+02, percent-clipped=4.0 2022-12-07 22:27:03,695 INFO [train.py:873] (3/4) Epoch 10, batch 4300, loss[loss=0.1579, simple_loss=0.1766, pruned_loss=0.06962, over 11178.00 frames. ], tot_loss[loss=0.1369, simple_loss=0.1635, pruned_loss=0.05518, over 1950687.87 frames. ], batch size: 100, lr: 7.90e-03, grad_scale: 4.0 2022-12-07 22:27:29,465 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=72387.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:28:01,103 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0 2022-12-07 22:28:26,476 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.161e+02 2.139e+02 2.736e+02 3.399e+02 6.791e+02, threshold=5.473e+02, percent-clipped=1.0 2022-12-07 22:28:27,597 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8284, 1.2396, 2.7777, 2.5500, 2.7087, 2.8058, 2.0838, 2.7696], device='cuda:3'), covar=tensor([0.0963, 0.1276, 0.0137, 0.0321, 0.0318, 0.0145, 0.0422, 0.0185], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0161, 0.0123, 0.0166, 0.0143, 0.0134, 0.0115, 0.0118], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 22:28:30,290 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0083, 1.7616, 4.2311, 3.9623, 3.9847, 4.3040, 3.6773, 4.2942], device='cuda:3'), covar=tensor([0.1348, 0.1392, 0.0095, 0.0200, 0.0183, 0.0103, 0.0244, 0.0125], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0160, 0.0123, 0.0166, 0.0143, 0.0133, 0.0115, 0.0118], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 22:28:34,602 INFO [train.py:873] (3/4) Epoch 10, batch 4400, loss[loss=0.1659, simple_loss=0.1511, pruned_loss=0.09031, over 1280.00 frames. ], tot_loss[loss=0.1371, simple_loss=0.1634, pruned_loss=0.05535, over 1905036.78 frames. ], batch size: 100, lr: 7.90e-03, grad_scale: 8.0 2022-12-07 22:28:58,349 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=72485.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:29:07,031 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1952, 1.4813, 3.9273, 1.9794, 3.9742, 4.1786, 3.3995, 4.5076], device='cuda:3'), covar=tensor([0.0199, 0.3075, 0.0440, 0.2141, 0.0456, 0.0382, 0.0593, 0.0148], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0157, 0.0158, 0.0169, 0.0173, 0.0173, 0.0135, 0.0140], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 22:29:56,358 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.072e+02 2.111e+02 2.675e+02 3.314e+02 5.959e+02, threshold=5.350e+02, percent-clipped=1.0 2022-12-07 22:30:05,239 INFO [train.py:873] (3/4) Epoch 10, batch 4500, loss[loss=0.1259, simple_loss=0.1523, pruned_loss=0.04974, over 12774.00 frames. ], tot_loss[loss=0.1355, simple_loss=0.1629, pruned_loss=0.05404, over 1992515.92 frames. ], batch size: 100, lr: 7.89e-03, grad_scale: 8.0 2022-12-07 22:30:12,778 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2022-12-07 22:30:19,475 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72575.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:31:03,242 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=72623.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:31:27,223 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.124e+02 2.389e+02 3.143e+02 3.960e+02 1.878e+03, threshold=6.285e+02, percent-clipped=10.0 2022-12-07 22:31:35,247 INFO [train.py:873] (3/4) Epoch 10, batch 4600, loss[loss=0.1629, simple_loss=0.1761, pruned_loss=0.07481, over 7803.00 frames. ], tot_loss[loss=0.1358, simple_loss=0.1628, pruned_loss=0.05439, over 1922930.41 frames. ], batch size: 100, lr: 7.89e-03, grad_scale: 8.0 2022-12-07 22:31:56,311 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=72682.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:31:58,020 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.4365, 4.9529, 4.9020, 5.4523, 5.0503, 4.4297, 5.3626, 4.5087], device='cuda:3'), covar=tensor([0.0273, 0.0969, 0.0293, 0.0341, 0.0704, 0.0431, 0.0491, 0.0440], device='cuda:3'), in_proj_covar=tensor([0.0159, 0.0254, 0.0174, 0.0172, 0.0168, 0.0139, 0.0262, 0.0154], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-07 22:32:57,785 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.181e+02 2.006e+02 2.571e+02 3.192e+02 6.224e+02, threshold=5.142e+02, percent-clipped=0.0 2022-12-07 22:33:06,364 INFO [train.py:873] (3/4) Epoch 10, batch 4700, loss[loss=0.1373, simple_loss=0.1605, pruned_loss=0.05703, over 6964.00 frames. ], tot_loss[loss=0.136, simple_loss=0.1631, pruned_loss=0.05447, over 1961895.90 frames. ], batch size: 100, lr: 7.88e-03, grad_scale: 8.0 2022-12-07 22:33:29,639 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72785.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:33:30,193 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2022-12-07 22:33:48,175 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.13 vs. limit=5.0 2022-12-07 22:34:12,941 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=72833.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:34:21,975 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.61 vs. limit=5.0 2022-12-07 22:34:27,384 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.194e+02 2.219e+02 2.922e+02 3.522e+02 5.753e+02, threshold=5.843e+02, percent-clipped=6.0 2022-12-07 22:34:35,286 INFO [train.py:873] (3/4) Epoch 10, batch 4800, loss[loss=0.1376, simple_loss=0.1642, pruned_loss=0.05549, over 14283.00 frames. ], tot_loss[loss=0.135, simple_loss=0.1624, pruned_loss=0.05383, over 1939582.77 frames. ], batch size: 63, lr: 7.88e-03, grad_scale: 8.0 2022-12-07 22:35:06,562 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72894.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:35:19,406 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9214, 2.9479, 3.0609, 3.0068, 3.0030, 2.7877, 1.4837, 2.8020], device='cuda:3'), covar=tensor([0.0368, 0.0391, 0.0401, 0.0366, 0.0364, 0.0731, 0.2703, 0.0325], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0162, 0.0136, 0.0135, 0.0195, 0.0131, 0.0156, 0.0182], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-07 22:35:56,441 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.266e+02 2.330e+02 2.761e+02 3.317e+02 8.665e+02, threshold=5.521e+02, percent-clipped=2.0 2022-12-07 22:36:01,289 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=72955.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:36:04,983 INFO [train.py:873] (3/4) Epoch 10, batch 4900, loss[loss=0.129, simple_loss=0.1571, pruned_loss=0.05048, over 13574.00 frames. ], tot_loss[loss=0.1362, simple_loss=0.163, pruned_loss=0.05467, over 1941830.19 frames. ], batch size: 100, lr: 7.87e-03, grad_scale: 8.0 2022-12-07 22:36:09,362 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72964.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 22:36:24,860 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72982.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:36:58,451 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5869, 2.8777, 2.7666, 2.8942, 2.3406, 2.9203, 2.6171, 1.4178], device='cuda:3'), covar=tensor([0.1650, 0.0778, 0.0931, 0.0624, 0.0911, 0.0433, 0.1093, 0.2346], device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0074, 0.0059, 0.0062, 0.0089, 0.0071, 0.0091, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0005], device='cuda:3') 2022-12-07 22:37:03,850 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73025.0, num_to_drop=1, layers_to_drop={3} 2022-12-07 22:37:08,114 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=73030.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:37:26,028 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.386e+02 2.275e+02 2.705e+02 3.390e+02 8.104e+02, threshold=5.410e+02, percent-clipped=1.0 2022-12-07 22:37:32,323 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4629, 2.1854, 3.3279, 3.4223, 3.4057, 2.2538, 3.2996, 2.6188], device='cuda:3'), covar=tensor([0.0341, 0.0848, 0.0727, 0.0435, 0.0332, 0.1245, 0.0339, 0.0878], device='cuda:3'), in_proj_covar=tensor([0.0274, 0.0245, 0.0364, 0.0314, 0.0253, 0.0296, 0.0287, 0.0274], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 22:37:33,744 INFO [train.py:873] (3/4) Epoch 10, batch 5000, loss[loss=0.1499, simple_loss=0.168, pruned_loss=0.06591, over 6933.00 frames. ], tot_loss[loss=0.1365, simple_loss=0.1631, pruned_loss=0.05498, over 1917667.18 frames. ], batch size: 100, lr: 7.87e-03, grad_scale: 8.0 2022-12-07 22:38:03,198 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9220, 1.4867, 3.8047, 3.4883, 3.6005, 3.8441, 3.1175, 3.8383], device='cuda:3'), covar=tensor([0.1409, 0.1548, 0.0106, 0.0227, 0.0205, 0.0116, 0.0263, 0.0119], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0159, 0.0121, 0.0165, 0.0140, 0.0132, 0.0115, 0.0116], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 22:38:20,046 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9841, 1.9669, 3.9337, 2.6138, 3.7866, 1.8705, 2.8773, 3.7165], device='cuda:3'), covar=tensor([0.0707, 0.5209, 0.0593, 0.8625, 0.0799, 0.4403, 0.1655, 0.0689], device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0218, 0.0194, 0.0295, 0.0212, 0.0221, 0.0216, 0.0199], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 22:38:26,946 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.88 vs. limit=2.0 2022-12-07 22:38:35,213 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.9594, 5.3185, 5.3044, 5.8811, 5.4578, 4.7003, 5.8142, 4.8401], device='cuda:3'), covar=tensor([0.0320, 0.1009, 0.0369, 0.0415, 0.0708, 0.0378, 0.0527, 0.0575], device='cuda:3'), in_proj_covar=tensor([0.0161, 0.0255, 0.0178, 0.0174, 0.0170, 0.0141, 0.0265, 0.0157], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-07 22:38:54,056 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.200e+01 2.088e+02 2.701e+02 3.421e+02 5.931e+02, threshold=5.402e+02, percent-clipped=1.0 2022-12-07 22:39:01,816 INFO [train.py:873] (3/4) Epoch 10, batch 5100, loss[loss=0.1206, simple_loss=0.1512, pruned_loss=0.04504, over 14200.00 frames. ], tot_loss[loss=0.1346, simple_loss=0.162, pruned_loss=0.05361, over 1971163.77 frames. ], batch size: 89, lr: 7.86e-03, grad_scale: 4.0 2022-12-07 22:39:22,825 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.08 vs. limit=5.0 2022-12-07 22:39:59,187 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.7845, 1.1187, 1.3018, 1.3202, 1.0855, 1.3668, 1.0639, 0.9333], device='cuda:3'), covar=tensor([0.2067, 0.0911, 0.0380, 0.0366, 0.1533, 0.0519, 0.1625, 0.1158], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0076, 0.0060, 0.0064, 0.0091, 0.0074, 0.0094, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0006, 0.0006], device='cuda:3') 2022-12-07 22:40:22,007 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73250.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:40:22,701 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.094e+02 2.487e+02 3.025e+02 3.884e+02 1.180e+03, threshold=6.049e+02, percent-clipped=5.0 2022-12-07 22:40:29,837 INFO [train.py:873] (3/4) Epoch 10, batch 5200, loss[loss=0.1248, simple_loss=0.1404, pruned_loss=0.05463, over 3848.00 frames. ], tot_loss[loss=0.1359, simple_loss=0.1629, pruned_loss=0.05451, over 1990327.83 frames. ], batch size: 100, lr: 7.85e-03, grad_scale: 8.0 2022-12-07 22:41:00,565 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=73293.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:41:05,678 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.6991, 4.3838, 4.1086, 4.3219, 4.4063, 4.5381, 4.6830, 4.6611], device='cuda:3'), covar=tensor([0.0822, 0.0589, 0.2445, 0.2882, 0.0786, 0.0902, 0.0949, 0.0804], device='cuda:3'), in_proj_covar=tensor([0.0359, 0.0250, 0.0418, 0.0541, 0.0315, 0.0408, 0.0382, 0.0349], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 22:41:25,135 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73320.0, num_to_drop=1, layers_to_drop={3} 2022-12-07 22:41:34,463 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8869, 0.8093, 0.7600, 0.6332, 0.6988, 0.4261, 0.5064, 0.6842], device='cuda:3'), covar=tensor([0.0109, 0.0099, 0.0082, 0.0102, 0.0136, 0.0346, 0.0177, 0.0239], device='cuda:3'), in_proj_covar=tensor([0.0014, 0.0015, 0.0013, 0.0013, 0.0014, 0.0022, 0.0018, 0.0023], device='cuda:3'), out_proj_covar=tensor([1.0630e-04, 1.1539e-04, 9.9415e-05, 1.0664e-04, 1.0626e-04, 1.5933e-04, 1.3350e-04, 1.5417e-04], device='cuda:3') 2022-12-07 22:41:42,170 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.8486, 5.2364, 5.2271, 5.8075, 5.3086, 4.5097, 5.8055, 4.9248], device='cuda:3'), covar=tensor([0.0330, 0.0930, 0.0355, 0.0395, 0.0913, 0.0434, 0.0465, 0.0455], device='cuda:3'), in_proj_covar=tensor([0.0162, 0.0256, 0.0178, 0.0174, 0.0171, 0.0142, 0.0265, 0.0158], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-07 22:41:52,767 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.271e+02 2.362e+02 2.946e+02 3.664e+02 8.798e+02, threshold=5.892e+02, percent-clipped=2.0 2022-12-07 22:41:54,623 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73354.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:41:59,542 INFO [train.py:873] (3/4) Epoch 10, batch 5300, loss[loss=0.1348, simple_loss=0.1624, pruned_loss=0.05367, over 14139.00 frames. ], tot_loss[loss=0.1346, simple_loss=0.1621, pruned_loss=0.05354, over 1999557.10 frames. ], batch size: 84, lr: 7.85e-03, grad_scale: 4.0 2022-12-07 22:42:15,644 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.82 vs. limit=2.0 2022-12-07 22:43:21,802 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.441e+02 2.215e+02 2.822e+02 3.427e+02 6.404e+02, threshold=5.643e+02, percent-clipped=2.0 2022-12-07 22:43:24,075 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2022-12-07 22:43:25,872 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.16 vs. limit=2.0 2022-12-07 22:43:27,864 INFO [train.py:873] (3/4) Epoch 10, batch 5400, loss[loss=0.185, simple_loss=0.1611, pruned_loss=0.1045, over 1311.00 frames. ], tot_loss[loss=0.1348, simple_loss=0.1625, pruned_loss=0.05351, over 1997163.50 frames. ], batch size: 100, lr: 7.84e-03, grad_scale: 4.0 2022-12-07 22:43:59,490 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4431, 2.2087, 2.4039, 1.4309, 2.0551, 2.2987, 2.5177, 2.0190], device='cuda:3'), covar=tensor([0.0715, 0.0908, 0.0955, 0.1693, 0.1367, 0.0741, 0.0516, 0.1570], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0184, 0.0134, 0.0124, 0.0131, 0.0137, 0.0112, 0.0136], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0007, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005], device='cuda:3') 2022-12-07 22:44:48,817 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=73550.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:44:50,662 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.270e+02 2.129e+02 2.539e+02 3.119e+02 4.865e+02, threshold=5.078e+02, percent-clipped=0.0 2022-12-07 22:44:55,294 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.64 vs. limit=2.0 2022-12-07 22:44:57,304 INFO [train.py:873] (3/4) Epoch 10, batch 5500, loss[loss=0.1843, simple_loss=0.1678, pruned_loss=0.1005, over 1324.00 frames. ], tot_loss[loss=0.134, simple_loss=0.1615, pruned_loss=0.0532, over 1914427.03 frames. ], batch size: 100, lr: 7.84e-03, grad_scale: 4.0 2022-12-07 22:45:03,318 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9839, 1.3761, 3.1203, 1.5887, 3.2792, 3.1738, 2.2403, 3.3141], device='cuda:3'), covar=tensor([0.0234, 0.2786, 0.0349, 0.1834, 0.0292, 0.0337, 0.0878, 0.0194], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0160, 0.0158, 0.0170, 0.0172, 0.0172, 0.0133, 0.0141], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 22:45:24,749 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=73590.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:45:31,969 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=73598.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:45:50,949 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=73620.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 22:46:17,375 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73649.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:46:19,041 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73651.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:46:19,731 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.890e+01 2.291e+02 2.969e+02 3.792e+02 1.060e+03, threshold=5.938e+02, percent-clipped=6.0 2022-12-07 22:46:25,681 INFO [train.py:873] (3/4) Epoch 10, batch 5600, loss[loss=0.1634, simple_loss=0.1809, pruned_loss=0.07301, over 10338.00 frames. ], tot_loss[loss=0.1349, simple_loss=0.1621, pruned_loss=0.0539, over 1916534.26 frames. ], batch size: 100, lr: 7.83e-03, grad_scale: 8.0 2022-12-07 22:46:33,870 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=73668.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 22:46:34,764 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5661, 4.7140, 4.9449, 4.0829, 4.8605, 5.1471, 1.9377, 4.4381], device='cuda:3'), covar=tensor([0.0274, 0.0243, 0.0376, 0.0329, 0.0236, 0.0126, 0.3007, 0.0276], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0161, 0.0134, 0.0134, 0.0193, 0.0131, 0.0156, 0.0181], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-07 22:47:18,352 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([6.1180, 5.5962, 5.5102, 6.0302, 5.6796, 4.8044, 6.0039, 5.0495], device='cuda:3'), covar=tensor([0.0261, 0.0764, 0.0290, 0.0374, 0.0632, 0.0372, 0.0424, 0.0435], device='cuda:3'), in_proj_covar=tensor([0.0160, 0.0256, 0.0180, 0.0175, 0.0171, 0.0144, 0.0265, 0.0158], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-07 22:47:20,299 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5044, 3.7074, 3.4095, 3.6692, 2.7885, 3.7486, 3.6052, 2.0426], device='cuda:3'), covar=tensor([0.2150, 0.0885, 0.1085, 0.0669, 0.0943, 0.0317, 0.0968, 0.2534], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0074, 0.0059, 0.0063, 0.0089, 0.0071, 0.0092, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0005, 0.0006], device='cuda:3') 2022-12-07 22:47:30,701 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=73732.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:47:48,148 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.384e+02 2.591e+02 3.035e+02 3.423e+02 1.031e+03, threshold=6.069e+02, percent-clipped=1.0 2022-12-07 22:47:55,243 INFO [train.py:873] (3/4) Epoch 10, batch 5700, loss[loss=0.1456, simple_loss=0.1698, pruned_loss=0.06071, over 11977.00 frames. ], tot_loss[loss=0.1346, simple_loss=0.1621, pruned_loss=0.05354, over 1946952.97 frames. ], batch size: 100, lr: 7.83e-03, grad_scale: 8.0 2022-12-07 22:48:24,580 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73793.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:49:17,101 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.176e+02 2.156e+02 2.743e+02 3.331e+02 7.800e+02, threshold=5.486e+02, percent-clipped=2.0 2022-12-07 22:49:23,294 INFO [train.py:873] (3/4) Epoch 10, batch 5800, loss[loss=0.1184, simple_loss=0.1617, pruned_loss=0.03752, over 14312.00 frames. ], tot_loss[loss=0.1343, simple_loss=0.1621, pruned_loss=0.05322, over 1958325.89 frames. ], batch size: 28, lr: 7.82e-03, grad_scale: 8.0 2022-12-07 22:49:38,232 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7728, 2.3273, 2.5407, 1.4930, 2.3737, 2.5486, 2.8622, 2.2275], device='cuda:3'), covar=tensor([0.0900, 0.1093, 0.1310, 0.2278, 0.1227, 0.0774, 0.0710, 0.1748], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0183, 0.0133, 0.0125, 0.0131, 0.0138, 0.0113, 0.0135], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0007, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0005], device='cuda:3') 2022-12-07 22:50:13,836 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.06 vs. limit=5.0 2022-12-07 22:50:20,214 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.6808, 2.4090, 4.8175, 3.0983, 4.3655, 2.1815, 3.3589, 4.4867], device='cuda:3'), covar=tensor([0.0524, 0.4705, 0.0291, 0.8415, 0.0544, 0.3890, 0.1389, 0.0275], device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0217, 0.0193, 0.0294, 0.0214, 0.0221, 0.0215, 0.0199], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 22:50:41,243 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73946.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:50:44,063 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=73949.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:50:46,443 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.240e+02 2.192e+02 2.728e+02 3.266e+02 1.115e+03, threshold=5.456e+02, percent-clipped=1.0 2022-12-07 22:50:52,943 INFO [train.py:873] (3/4) Epoch 10, batch 5900, loss[loss=0.154, simple_loss=0.1736, pruned_loss=0.06723, over 9494.00 frames. ], tot_loss[loss=0.1341, simple_loss=0.1618, pruned_loss=0.05324, over 1945167.88 frames. ], batch size: 100, lr: 7.82e-03, grad_scale: 8.0 2022-12-07 22:51:26,784 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=73997.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:51:32,494 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0471, 2.1894, 2.9975, 3.0892, 2.9696, 2.1009, 2.9212, 2.3851], device='cuda:3'), covar=tensor([0.0274, 0.0690, 0.0572, 0.0360, 0.0360, 0.1061, 0.0270, 0.0713], device='cuda:3'), in_proj_covar=tensor([0.0272, 0.0245, 0.0363, 0.0311, 0.0252, 0.0294, 0.0285, 0.0274], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 22:52:14,955 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.161e+02 2.132e+02 2.894e+02 3.484e+02 7.158e+02, threshold=5.787e+02, percent-clipped=1.0 2022-12-07 22:52:20,978 INFO [train.py:873] (3/4) Epoch 10, batch 6000, loss[loss=0.1762, simple_loss=0.1957, pruned_loss=0.07832, over 14298.00 frames. ], tot_loss[loss=0.1345, simple_loss=0.1622, pruned_loss=0.05337, over 2039691.88 frames. ], batch size: 39, lr: 7.81e-03, grad_scale: 8.0 2022-12-07 22:52:20,978 INFO [train.py:896] (3/4) Computing validation loss 2022-12-07 22:52:36,794 INFO [train.py:905] (3/4) Epoch 10, validation: loss=0.1251, simple_loss=0.167, pruned_loss=0.04163, over 857387.00 frames. 2022-12-07 22:52:36,795 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-07 22:52:37,045 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6567, 2.3088, 3.3491, 2.4679, 3.4399, 3.2862, 3.3720, 2.8567], device='cuda:3'), covar=tensor([0.0723, 0.2851, 0.1105, 0.2127, 0.0692, 0.1020, 0.1278, 0.1999], device='cuda:3'), in_proj_covar=tensor([0.0342, 0.0319, 0.0399, 0.0305, 0.0374, 0.0316, 0.0361, 0.0317], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 22:52:42,078 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74064.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:53:02,965 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74088.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:53:35,971 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74125.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:53:59,516 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5951, 3.5876, 4.2865, 3.0824, 2.6511, 3.4731, 1.9406, 3.6101], device='cuda:3'), covar=tensor([0.1169, 0.0995, 0.0571, 0.2453, 0.2492, 0.0963, 0.4841, 0.1207], device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0093, 0.0087, 0.0093, 0.0112, 0.0078, 0.0124, 0.0086], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0005, 0.0004], device='cuda:3') 2022-12-07 22:54:00,210 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.199e+02 2.430e+02 2.883e+02 3.808e+02 1.343e+03, threshold=5.765e+02, percent-clipped=6.0 2022-12-07 22:54:06,424 INFO [train.py:873] (3/4) Epoch 10, batch 6100, loss[loss=0.1247, simple_loss=0.1632, pruned_loss=0.04308, over 14234.00 frames. ], tot_loss[loss=0.135, simple_loss=0.1623, pruned_loss=0.05386, over 2015971.65 frames. ], batch size: 37, lr: 7.81e-03, grad_scale: 8.0 2022-12-07 22:54:14,623 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.91 vs. limit=5.0 2022-12-07 22:54:32,137 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2142, 2.4400, 5.2394, 4.6463, 4.4713, 5.3648, 5.0091, 5.3867], device='cuda:3'), covar=tensor([0.1301, 0.1140, 0.0065, 0.0145, 0.0169, 0.0075, 0.0082, 0.0084], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0158, 0.0122, 0.0166, 0.0141, 0.0134, 0.0116, 0.0117], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 22:54:49,378 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5842, 3.4827, 3.3220, 3.7029, 3.2287, 3.1642, 3.6771, 3.5951], device='cuda:3'), covar=tensor([0.0817, 0.0836, 0.0983, 0.0698, 0.1075, 0.0738, 0.0755, 0.0760], device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0121, 0.0131, 0.0139, 0.0133, 0.0109, 0.0155, 0.0131], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 22:55:18,913 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74240.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:55:23,991 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74246.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:55:29,020 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.083e+02 2.250e+02 2.648e+02 3.376e+02 5.030e+02, threshold=5.296e+02, percent-clipped=0.0 2022-12-07 22:55:35,298 INFO [train.py:873] (3/4) Epoch 10, batch 6200, loss[loss=0.128, simple_loss=0.1256, pruned_loss=0.06525, over 1228.00 frames. ], tot_loss[loss=0.1338, simple_loss=0.1617, pruned_loss=0.05296, over 2029170.64 frames. ], batch size: 100, lr: 7.80e-03, grad_scale: 8.0 2022-12-07 22:55:51,953 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8766, 1.7208, 2.0652, 1.6249, 2.0851, 1.8152, 1.7651, 1.9153], device='cuda:3'), covar=tensor([0.0491, 0.1267, 0.0255, 0.0381, 0.0294, 0.0494, 0.0210, 0.0311], device='cuda:3'), in_proj_covar=tensor([0.0343, 0.0319, 0.0397, 0.0305, 0.0373, 0.0315, 0.0360, 0.0315], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 22:55:57,095 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5959, 2.1815, 3.0260, 1.9363, 1.8647, 2.6217, 1.4774, 2.5090], device='cuda:3'), covar=tensor([0.1066, 0.1614, 0.0633, 0.2077, 0.2700, 0.0803, 0.3988, 0.0944], device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0094, 0.0087, 0.0094, 0.0114, 0.0080, 0.0126, 0.0087], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0005, 0.0004], device='cuda:3') 2022-12-07 22:56:06,632 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=74294.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:56:13,673 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74301.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:56:53,236 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.3665, 1.3041, 1.2872, 1.3641, 1.3893, 0.7785, 1.3272, 1.3105], device='cuda:3'), covar=tensor([0.0787, 0.0736, 0.0532, 0.0501, 0.0724, 0.0956, 0.0883, 0.0739], device='cuda:3'), in_proj_covar=tensor([0.0024, 0.0025, 0.0027, 0.0024, 0.0025, 0.0037, 0.0025, 0.0027], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001], device='cuda:3') 2022-12-07 22:56:58,218 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.014e+02 2.284e+02 2.758e+02 3.409e+02 5.517e+02, threshold=5.515e+02, percent-clipped=2.0 2022-12-07 22:57:04,253 INFO [train.py:873] (3/4) Epoch 10, batch 6300, loss[loss=0.1622, simple_loss=0.1567, pruned_loss=0.08382, over 2644.00 frames. ], tot_loss[loss=0.1335, simple_loss=0.1614, pruned_loss=0.05282, over 1988778.14 frames. ], batch size: 100, lr: 7.80e-03, grad_scale: 8.0 2022-12-07 22:57:30,074 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74388.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:57:58,123 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74420.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:58:11,970 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=74436.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 22:58:25,309 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.400e+02 2.328e+02 2.936e+02 3.672e+02 8.689e+02, threshold=5.873e+02, percent-clipped=4.0 2022-12-07 22:58:32,160 INFO [train.py:873] (3/4) Epoch 10, batch 6400, loss[loss=0.1531, simple_loss=0.1729, pruned_loss=0.06668, over 14178.00 frames. ], tot_loss[loss=0.1337, simple_loss=0.1613, pruned_loss=0.05299, over 1947126.86 frames. ], batch size: 84, lr: 7.79e-03, grad_scale: 8.0 2022-12-07 22:59:08,365 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3929, 1.5191, 4.1504, 1.8052, 4.0581, 4.3169, 3.7916, 4.6357], device='cuda:3'), covar=tensor([0.0210, 0.3508, 0.0484, 0.2594, 0.0448, 0.0348, 0.0380, 0.0210], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0159, 0.0159, 0.0171, 0.0174, 0.0174, 0.0134, 0.0141], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 22:59:20,280 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2022-12-07 22:59:34,134 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8218, 2.2306, 3.8151, 3.8891, 3.8382, 2.3199, 3.8552, 3.0380], device='cuda:3'), covar=tensor([0.0300, 0.0775, 0.0626, 0.0338, 0.0288, 0.1195, 0.0272, 0.0676], device='cuda:3'), in_proj_covar=tensor([0.0271, 0.0243, 0.0361, 0.0307, 0.0249, 0.0290, 0.0282, 0.0269], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-07 22:59:53,825 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.408e+02 2.247e+02 2.848e+02 3.548e+02 7.995e+02, threshold=5.696e+02, percent-clipped=2.0 2022-12-07 22:59:59,601 INFO [train.py:873] (3/4) Epoch 10, batch 6500, loss[loss=0.1363, simple_loss=0.1346, pruned_loss=0.06901, over 2670.00 frames. ], tot_loss[loss=0.1354, simple_loss=0.1624, pruned_loss=0.05419, over 1928497.99 frames. ], batch size: 100, lr: 7.79e-03, grad_scale: 8.0 2022-12-07 23:00:09,089 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.8684, 5.2468, 5.2660, 5.8244, 5.3238, 4.6979, 5.7118, 4.7319], device='cuda:3'), covar=tensor([0.0264, 0.1004, 0.0346, 0.0399, 0.0859, 0.0374, 0.0549, 0.0559], device='cuda:3'), in_proj_covar=tensor([0.0158, 0.0254, 0.0178, 0.0175, 0.0170, 0.0142, 0.0263, 0.0156], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-07 23:00:11,638 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.5543, 1.4755, 1.6380, 1.4819, 1.4965, 1.1971, 1.0351, 1.0414], device='cuda:3'), covar=tensor([0.0209, 0.0416, 0.0250, 0.0232, 0.0242, 0.0305, 0.0272, 0.0440], device='cuda:3'), in_proj_covar=tensor([0.0014, 0.0015, 0.0013, 0.0014, 0.0014, 0.0023, 0.0018, 0.0023], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2022-12-07 23:00:31,616 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74595.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:00:32,458 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74596.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:01:09,545 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6808, 1.6727, 1.6278, 1.4881, 1.6192, 0.9315, 1.5838, 1.7013], device='cuda:3'), covar=tensor([0.0712, 0.0641, 0.0681, 0.1280, 0.0587, 0.0778, 0.0657, 0.0502], device='cuda:3'), in_proj_covar=tensor([0.0024, 0.0025, 0.0027, 0.0024, 0.0025, 0.0037, 0.0025, 0.0027], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001], device='cuda:3') 2022-12-07 23:01:18,851 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9689, 1.4993, 3.5203, 3.2134, 3.3152, 3.4878, 2.8559, 3.5441], device='cuda:3'), covar=tensor([0.1303, 0.1460, 0.0095, 0.0257, 0.0236, 0.0126, 0.0302, 0.0108], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0157, 0.0121, 0.0164, 0.0141, 0.0133, 0.0115, 0.0115], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 23:01:22,223 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.260e+02 2.214e+02 2.798e+02 3.905e+02 7.921e+02, threshold=5.597e+02, percent-clipped=3.0 2022-12-07 23:01:25,293 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74656.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:01:28,054 INFO [train.py:873] (3/4) Epoch 10, batch 6600, loss[loss=0.113, simple_loss=0.1504, pruned_loss=0.03779, over 14567.00 frames. ], tot_loss[loss=0.1347, simple_loss=0.1616, pruned_loss=0.05394, over 1934689.50 frames. ], batch size: 22, lr: 7.78e-03, grad_scale: 4.0 2022-12-07 23:02:22,389 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74720.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:02:51,954 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.183e+02 2.052e+02 2.533e+02 3.147e+02 6.445e+02, threshold=5.065e+02, percent-clipped=3.0 2022-12-07 23:02:57,098 INFO [train.py:873] (3/4) Epoch 10, batch 6700, loss[loss=0.1573, simple_loss=0.1806, pruned_loss=0.06704, over 14261.00 frames. ], tot_loss[loss=0.135, simple_loss=0.162, pruned_loss=0.05402, over 1939024.58 frames. ], batch size: 80, lr: 7.78e-03, grad_scale: 4.0 2022-12-07 23:03:04,999 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=74768.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:03:10,770 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.79 vs. limit=5.0 2022-12-07 23:03:21,166 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3595, 2.3668, 1.8715, 2.4105, 2.2282, 2.3686, 2.0652, 1.9737], device='cuda:3'), covar=tensor([0.0801, 0.1035, 0.2001, 0.0611, 0.0872, 0.0722, 0.1161, 0.1463], device='cuda:3'), in_proj_covar=tensor([0.0264, 0.0295, 0.0265, 0.0245, 0.0303, 0.0290, 0.0254, 0.0254], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2022-12-07 23:03:44,981 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8554, 1.5276, 1.7983, 2.0540, 1.3748, 1.7919, 1.8376, 1.9698], device='cuda:3'), covar=tensor([0.0098, 0.0186, 0.0107, 0.0094, 0.0199, 0.0224, 0.0117, 0.0092], device='cuda:3'), in_proj_covar=tensor([0.0273, 0.0245, 0.0365, 0.0308, 0.0252, 0.0293, 0.0283, 0.0272], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 23:04:19,048 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.400e+02 2.393e+02 2.874e+02 3.943e+02 8.045e+02, threshold=5.748e+02, percent-clipped=8.0 2022-12-07 23:04:24,814 INFO [train.py:873] (3/4) Epoch 10, batch 6800, loss[loss=0.1182, simple_loss=0.1503, pruned_loss=0.04304, over 13949.00 frames. ], tot_loss[loss=0.135, simple_loss=0.1618, pruned_loss=0.05406, over 1918666.41 frames. ], batch size: 23, lr: 7.77e-03, grad_scale: 8.0 2022-12-07 23:04:56,631 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74896.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:05:08,609 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74909.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:05:38,793 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=74944.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:05:45,391 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74951.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:05:47,781 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.213e+02 2.283e+02 2.713e+02 3.632e+02 6.989e+02, threshold=5.426e+02, percent-clipped=3.0 2022-12-07 23:05:52,207 INFO [train.py:873] (3/4) Epoch 10, batch 6900, loss[loss=0.1277, simple_loss=0.16, pruned_loss=0.04774, over 14019.00 frames. ], tot_loss[loss=0.1336, simple_loss=0.161, pruned_loss=0.05312, over 1958830.71 frames. ], batch size: 20, lr: 7.77e-03, grad_scale: 4.0 2022-12-07 23:05:56,675 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74964.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:06:01,552 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74970.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:06:08,104 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4948, 2.7289, 2.7950, 2.8467, 2.1769, 2.8350, 2.5369, 1.4211], device='cuda:3'), covar=tensor([0.1578, 0.0924, 0.0661, 0.0649, 0.1292, 0.0520, 0.1489, 0.2651], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0074, 0.0059, 0.0063, 0.0089, 0.0073, 0.0094, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0006, 0.0006], device='cuda:3') 2022-12-07 23:06:52,501 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75025.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:07:03,919 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.01 vs. limit=2.0 2022-12-07 23:07:16,832 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.068e+02 2.278e+02 2.897e+02 3.473e+02 8.232e+02, threshold=5.794e+02, percent-clipped=1.0 2022-12-07 23:07:18,236 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5480, 3.3626, 3.1134, 2.1057, 3.0865, 3.3608, 3.6897, 2.9253], device='cuda:3'), covar=tensor([0.0675, 0.1538, 0.1008, 0.1850, 0.0779, 0.0576, 0.0732, 0.1291], device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0184, 0.0133, 0.0124, 0.0131, 0.0140, 0.0113, 0.0136], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0006], device='cuda:3') 2022-12-07 23:07:18,470 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.57 vs. limit=2.0 2022-12-07 23:07:21,371 INFO [train.py:873] (3/4) Epoch 10, batch 7000, loss[loss=0.1526, simple_loss=0.1741, pruned_loss=0.06552, over 8569.00 frames. ], tot_loss[loss=0.1333, simple_loss=0.161, pruned_loss=0.05284, over 1974940.32 frames. ], batch size: 100, lr: 7.76e-03, grad_scale: 4.0 2022-12-07 23:07:31,836 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3417, 3.4997, 3.6638, 3.4486, 3.5678, 3.4167, 1.4419, 3.3810], device='cuda:3'), covar=tensor([0.0323, 0.0358, 0.0384, 0.0408, 0.0327, 0.0521, 0.3095, 0.0321], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0162, 0.0135, 0.0135, 0.0193, 0.0132, 0.0155, 0.0181], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-07 23:08:05,360 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.67 vs. limit=2.0 2022-12-07 23:08:43,870 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9270, 1.5857, 2.0953, 1.7025, 1.9453, 1.4053, 1.7356, 1.9560], device='cuda:3'), covar=tensor([0.1846, 0.2578, 0.0416, 0.1924, 0.1285, 0.1872, 0.1031, 0.0615], device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0221, 0.0200, 0.0295, 0.0220, 0.0225, 0.0218, 0.0205], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 23:08:45,730 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.467e+02 2.185e+02 2.832e+02 3.460e+02 7.150e+02, threshold=5.664e+02, percent-clipped=2.0 2022-12-07 23:08:49,895 INFO [train.py:873] (3/4) Epoch 10, batch 7100, loss[loss=0.1225, simple_loss=0.1546, pruned_loss=0.04518, over 14239.00 frames. ], tot_loss[loss=0.1349, simple_loss=0.162, pruned_loss=0.05388, over 1922695.06 frames. ], batch size: 69, lr: 7.76e-03, grad_scale: 4.0 2022-12-07 23:09:51,384 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=75228.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:10:07,056 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5398, 1.9493, 3.6460, 2.6042, 3.4893, 1.8345, 2.7498, 3.5252], device='cuda:3'), covar=tensor([0.0832, 0.4995, 0.0513, 0.6028, 0.0752, 0.4177, 0.1415, 0.0520], device='cuda:3'), in_proj_covar=tensor([0.0249, 0.0223, 0.0202, 0.0297, 0.0221, 0.0226, 0.0221, 0.0207], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 23:10:11,167 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75251.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:10:13,566 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.083e+02 2.322e+02 2.903e+02 3.521e+02 5.847e+02, threshold=5.807e+02, percent-clipped=2.0 2022-12-07 23:10:18,337 INFO [train.py:873] (3/4) Epoch 10, batch 7200, loss[loss=0.1487, simple_loss=0.1702, pruned_loss=0.06363, over 11128.00 frames. ], tot_loss[loss=0.135, simple_loss=0.1618, pruned_loss=0.05408, over 1901808.62 frames. ], batch size: 100, lr: 7.75e-03, grad_scale: 8.0 2022-12-07 23:10:23,937 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75265.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:10:28,265 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=75270.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:10:44,749 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75289.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:10:53,493 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75299.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:11:11,949 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75320.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:11:21,432 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75331.0, num_to_drop=1, layers_to_drop={3} 2022-12-07 23:11:26,635 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5765, 3.3831, 3.3099, 3.6872, 3.2664, 2.9544, 3.6125, 3.5385], device='cuda:3'), covar=tensor([0.0608, 0.0827, 0.0850, 0.0570, 0.0812, 0.0727, 0.0737, 0.0710], device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0125, 0.0134, 0.0142, 0.0136, 0.0112, 0.0158, 0.0133], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 23:11:41,652 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.115e+02 2.194e+02 2.882e+02 3.919e+02 8.679e+02, threshold=5.764e+02, percent-clipped=4.0 2022-12-07 23:11:46,062 INFO [train.py:873] (3/4) Epoch 10, batch 7300, loss[loss=0.1278, simple_loss=0.1617, pruned_loss=0.04696, over 14258.00 frames. ], tot_loss[loss=0.1345, simple_loss=0.1616, pruned_loss=0.05375, over 1917721.55 frames. ], batch size: 63, lr: 7.75e-03, grad_scale: 8.0 2022-12-07 23:12:03,729 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.76 vs. limit=2.0 2022-12-07 23:13:09,406 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.408e+02 2.421e+02 2.835e+02 3.319e+02 6.698e+02, threshold=5.670e+02, percent-clipped=3.0 2022-12-07 23:13:14,009 INFO [train.py:873] (3/4) Epoch 10, batch 7400, loss[loss=0.1131, simple_loss=0.1545, pruned_loss=0.03587, over 14439.00 frames. ], tot_loss[loss=0.1359, simple_loss=0.1626, pruned_loss=0.05457, over 1912406.36 frames. ], batch size: 53, lr: 7.74e-03, grad_scale: 8.0 2022-12-07 23:13:39,232 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.69 vs. limit=5.0 2022-12-07 23:14:25,500 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.5582, 1.6917, 1.5873, 1.4904, 1.8068, 1.3141, 1.3686, 1.2045], device='cuda:3'), covar=tensor([0.0456, 0.0497, 0.0723, 0.0646, 0.0608, 0.0574, 0.0633, 0.0862], device='cuda:3'), in_proj_covar=tensor([0.0014, 0.0016, 0.0013, 0.0014, 0.0014, 0.0023, 0.0018, 0.0024], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2022-12-07 23:14:28,472 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.1516, 4.9312, 4.6633, 5.1695, 4.6942, 4.3975, 5.1894, 5.0009], device='cuda:3'), covar=tensor([0.0574, 0.0576, 0.0762, 0.0528, 0.0767, 0.0580, 0.0575, 0.0667], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0124, 0.0133, 0.0141, 0.0135, 0.0112, 0.0157, 0.0133], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 23:14:38,882 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.227e+02 2.214e+02 2.894e+02 3.789e+02 9.059e+02, threshold=5.789e+02, percent-clipped=5.0 2022-12-07 23:14:42,302 INFO [train.py:873] (3/4) Epoch 10, batch 7500, loss[loss=0.1711, simple_loss=0.1817, pruned_loss=0.08022, over 10383.00 frames. ], tot_loss[loss=0.1346, simple_loss=0.1618, pruned_loss=0.05367, over 1958371.99 frames. ], batch size: 100, lr: 7.73e-03, grad_scale: 4.0 2022-12-07 23:14:47,594 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75565.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:15:04,155 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75584.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:15:15,860 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4080, 1.4719, 1.6629, 1.3583, 1.4472, 1.1281, 1.2069, 0.9196], device='cuda:3'), covar=tensor([0.0200, 0.0285, 0.0230, 0.0218, 0.0244, 0.0296, 0.0227, 0.0425], device='cuda:3'), in_proj_covar=tensor([0.0014, 0.0016, 0.0013, 0.0014, 0.0014, 0.0023, 0.0019, 0.0024], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2022-12-07 23:15:24,292 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75613.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:16:08,901 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75620.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:16:09,628 INFO [train.py:873] (3/4) Epoch 11, batch 0, loss[loss=0.1526, simple_loss=0.188, pruned_loss=0.05862, over 14287.00 frames. ], tot_loss[loss=0.1526, simple_loss=0.188, pruned_loss=0.05862, over 14287.00 frames. ], batch size: 44, lr: 7.38e-03, grad_scale: 8.0 2022-12-07 23:16:09,628 INFO [train.py:896] (3/4) Computing validation loss 2022-12-07 23:16:16,873 INFO [train.py:905] (3/4) Epoch 11, validation: loss=0.1341, simple_loss=0.1756, pruned_loss=0.0463, over 857387.00 frames. 2022-12-07 23:16:16,874 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-07 23:16:21,387 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75626.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 23:16:45,828 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.36 vs. limit=5.0 2022-12-07 23:16:46,493 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6672, 1.5383, 1.7483, 1.5105, 1.6989, 0.9373, 1.6805, 1.6821], device='cuda:3'), covar=tensor([0.0572, 0.0751, 0.0824, 0.0977, 0.1036, 0.0928, 0.0849, 0.0976], device='cuda:3'), in_proj_covar=tensor([0.0024, 0.0024, 0.0027, 0.0024, 0.0025, 0.0036, 0.0025, 0.0027], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001], device='cuda:3') 2022-12-07 23:16:47,137 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 5.744e+01 1.774e+02 2.985e+02 4.009e+02 1.076e+03, threshold=5.970e+02, percent-clipped=9.0 2022-12-07 23:16:58,381 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75668.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:16:59,361 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9229, 3.6567, 3.3502, 3.5220, 3.7840, 3.8409, 3.8576, 3.8847], device='cuda:3'), covar=tensor([0.0877, 0.0600, 0.2248, 0.2686, 0.0805, 0.0798, 0.1082, 0.0844], device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0245, 0.0412, 0.0532, 0.0311, 0.0396, 0.0380, 0.0345], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 23:17:10,942 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 2022-12-07 23:17:21,882 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9301, 2.9052, 3.8846, 2.8567, 2.1330, 3.1989, 1.8797, 3.3868], device='cuda:3'), covar=tensor([0.1078, 0.1090, 0.0487, 0.1483, 0.2417, 0.0916, 0.3755, 0.0752], device='cuda:3'), in_proj_covar=tensor([0.0080, 0.0094, 0.0089, 0.0095, 0.0114, 0.0081, 0.0126, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0005, 0.0004], device='cuda:3') 2022-12-07 23:17:45,592 INFO [train.py:873] (3/4) Epoch 11, batch 100, loss[loss=0.1515, simple_loss=0.1418, pruned_loss=0.08063, over 2553.00 frames. ], tot_loss[loss=0.136, simple_loss=0.164, pruned_loss=0.054, over 873921.26 frames. ], batch size: 100, lr: 7.38e-03, grad_scale: 8.0 2022-12-07 23:18:11,411 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4083, 2.3001, 2.5166, 2.4618, 2.4521, 2.0740, 1.3837, 2.1642], device='cuda:3'), covar=tensor([0.0496, 0.0457, 0.0496, 0.0371, 0.0420, 0.1179, 0.2361, 0.0411], device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0163, 0.0137, 0.0136, 0.0194, 0.0131, 0.0155, 0.0182], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-07 23:18:14,580 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.558e+02 2.145e+02 2.643e+02 3.199e+02 5.049e+02, threshold=5.286e+02, percent-clipped=0.0 2022-12-07 23:18:32,435 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 2022-12-07 23:18:55,551 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4346, 1.3875, 1.5494, 1.4253, 1.5232, 0.9106, 1.4508, 1.5873], device='cuda:3'), covar=tensor([0.1291, 0.0984, 0.0684, 0.0720, 0.1570, 0.1021, 0.0890, 0.0994], device='cuda:3'), in_proj_covar=tensor([0.0025, 0.0024, 0.0027, 0.0024, 0.0026, 0.0037, 0.0026, 0.0028], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001], device='cuda:3') 2022-12-07 23:19:11,218 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.92 vs. limit=5.0 2022-12-07 23:19:13,453 INFO [train.py:873] (3/4) Epoch 11, batch 200, loss[loss=0.1233, simple_loss=0.1594, pruned_loss=0.04364, over 14138.00 frames. ], tot_loss[loss=0.1338, simple_loss=0.1619, pruned_loss=0.05288, over 1366237.11 frames. ], batch size: 84, lr: 7.37e-03, grad_scale: 8.0 2022-12-07 23:19:43,429 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.111e+02 2.275e+02 2.829e+02 3.715e+02 6.696e+02, threshold=5.659e+02, percent-clipped=6.0 2022-12-07 23:19:51,461 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2022-12-07 23:20:08,853 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75884.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:20:27,949 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 2022-12-07 23:20:41,221 INFO [train.py:873] (3/4) Epoch 11, batch 300, loss[loss=0.1252, simple_loss=0.1445, pruned_loss=0.05297, over 4951.00 frames. ], tot_loss[loss=0.1349, simple_loss=0.1622, pruned_loss=0.05385, over 1605859.35 frames. ], batch size: 100, lr: 7.37e-03, grad_scale: 8.0 2022-12-07 23:20:46,339 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75926.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 23:20:51,531 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75932.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:21:11,325 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.372e+02 2.182e+02 2.664e+02 3.344e+02 7.501e+02, threshold=5.329e+02, percent-clipped=3.0 2022-12-07 23:21:26,392 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7756, 2.5194, 3.6219, 2.7845, 3.6521, 3.5349, 3.3709, 2.9337], device='cuda:3'), covar=tensor([0.0703, 0.2693, 0.0927, 0.1834, 0.0872, 0.0832, 0.1318, 0.1866], device='cuda:3'), in_proj_covar=tensor([0.0344, 0.0317, 0.0395, 0.0302, 0.0375, 0.0318, 0.0356, 0.0316], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 23:21:28,768 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75974.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:22:00,785 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9204, 1.6811, 3.9781, 3.7432, 3.7974, 4.0845, 3.6075, 4.0677], device='cuda:3'), covar=tensor([0.1405, 0.1471, 0.0116, 0.0211, 0.0223, 0.0127, 0.0200, 0.0140], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0157, 0.0122, 0.0164, 0.0140, 0.0133, 0.0115, 0.0116], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 23:22:10,198 INFO [train.py:873] (3/4) Epoch 11, batch 400, loss[loss=0.1856, simple_loss=0.1649, pruned_loss=0.1032, over 1288.00 frames. ], tot_loss[loss=0.1339, simple_loss=0.1616, pruned_loss=0.05312, over 1751047.98 frames. ], batch size: 100, lr: 7.36e-03, grad_scale: 8.0 2022-12-07 23:22:25,971 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.51 vs. limit=2.0 2022-12-07 23:22:40,269 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.302e+02 2.357e+02 2.915e+02 3.546e+02 6.707e+02, threshold=5.830e+02, percent-clipped=5.0 2022-12-07 23:23:38,044 INFO [train.py:873] (3/4) Epoch 11, batch 500, loss[loss=0.1352, simple_loss=0.1442, pruned_loss=0.06303, over 3895.00 frames. ], tot_loss[loss=0.1331, simple_loss=0.1611, pruned_loss=0.0525, over 1871229.29 frames. ], batch size: 100, lr: 7.36e-03, grad_scale: 8.0 2022-12-07 23:24:07,886 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.432e+01 2.214e+02 2.601e+02 3.343e+02 5.170e+02, threshold=5.201e+02, percent-clipped=0.0 2022-12-07 23:24:13,377 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7414, 1.4029, 2.8851, 1.4855, 2.9789, 2.8869, 1.9834, 3.0525], device='cuda:3'), covar=tensor([0.0277, 0.2727, 0.0391, 0.2010, 0.0350, 0.0504, 0.1064, 0.0224], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0157, 0.0157, 0.0168, 0.0171, 0.0175, 0.0133, 0.0139], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 23:24:37,946 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5374, 3.4974, 4.4043, 3.3141, 2.5218, 3.8253, 2.4238, 3.7200], device='cuda:3'), covar=tensor([0.0928, 0.0757, 0.0537, 0.1259, 0.2268, 0.0925, 0.3230, 0.0928], device='cuda:3'), in_proj_covar=tensor([0.0080, 0.0095, 0.0091, 0.0097, 0.0115, 0.0082, 0.0126, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0005, 0.0004], device='cuda:3') 2022-12-07 23:25:05,642 INFO [train.py:873] (3/4) Epoch 11, batch 600, loss[loss=0.1638, simple_loss=0.1555, pruned_loss=0.08607, over 1252.00 frames. ], tot_loss[loss=0.1337, simple_loss=0.1615, pruned_loss=0.05295, over 1911616.63 frames. ], batch size: 100, lr: 7.35e-03, grad_scale: 8.0 2022-12-07 23:25:35,309 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.207e+02 2.199e+02 2.747e+02 3.310e+02 7.404e+02, threshold=5.494e+02, percent-clipped=4.0 2022-12-07 23:25:41,885 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.25 vs. limit=5.0 2022-12-07 23:25:47,375 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.4910, 5.1593, 5.0032, 5.5143, 5.0991, 4.7825, 5.4834, 5.3692], device='cuda:3'), covar=tensor([0.0607, 0.0602, 0.0659, 0.0448, 0.0580, 0.0458, 0.0583, 0.0565], device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0123, 0.0130, 0.0141, 0.0132, 0.0110, 0.0155, 0.0131], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-07 23:25:47,492 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=76269.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 23:25:58,172 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.7196, 1.1664, 1.2718, 1.2481, 0.9851, 1.3006, 1.1735, 0.9389], device='cuda:3'), covar=tensor([0.2369, 0.0990, 0.0347, 0.0478, 0.1514, 0.0596, 0.1255, 0.1060], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0077, 0.0060, 0.0065, 0.0093, 0.0073, 0.0095, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0006, 0.0006], device='cuda:3') 2022-12-07 23:26:09,648 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=76294.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:26:12,964 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.7942, 0.7987, 0.7417, 0.8475, 0.8203, 0.3274, 0.7384, 0.8283], device='cuda:3'), covar=tensor([0.0228, 0.0334, 0.0300, 0.0311, 0.0234, 0.0261, 0.1031, 0.0504], device='cuda:3'), in_proj_covar=tensor([0.0024, 0.0025, 0.0028, 0.0024, 0.0026, 0.0037, 0.0026, 0.0027], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001], device='cuda:3') 2022-12-07 23:26:32,274 INFO [train.py:873] (3/4) Epoch 11, batch 700, loss[loss=0.2294, simple_loss=0.1914, pruned_loss=0.1336, over 1224.00 frames. ], tot_loss[loss=0.1333, simple_loss=0.1611, pruned_loss=0.05279, over 1953826.73 frames. ], batch size: 100, lr: 7.35e-03, grad_scale: 8.0 2022-12-07 23:26:34,425 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=76323.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 23:26:37,133 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2950, 1.9805, 2.2854, 1.5102, 1.9606, 2.2916, 2.3425, 2.0238], device='cuda:3'), covar=tensor([0.0719, 0.0624, 0.0909, 0.1508, 0.1303, 0.0726, 0.0474, 0.1319], device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0183, 0.0136, 0.0125, 0.0135, 0.0140, 0.0115, 0.0138], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005, 0.0006], device='cuda:3') 2022-12-07 23:26:40,772 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=76330.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 23:27:01,830 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.209e+02 2.063e+02 2.502e+02 2.955e+02 5.307e+02, threshold=5.004e+02, percent-clipped=0.0 2022-12-07 23:27:02,037 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=76355.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:27:04,581 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8213, 3.5674, 3.5429, 3.8864, 3.6096, 3.5718, 3.8965, 3.3186], device='cuda:3'), covar=tensor([0.0597, 0.1063, 0.0493, 0.0496, 0.0894, 0.1206, 0.0666, 0.0582], device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0261, 0.0186, 0.0179, 0.0176, 0.0145, 0.0268, 0.0159], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-07 23:27:27,237 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=76384.0, num_to_drop=1, layers_to_drop={3} 2022-12-07 23:27:38,378 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2775, 1.8812, 2.2977, 1.4880, 2.0007, 2.2737, 2.3188, 2.0334], device='cuda:3'), covar=tensor([0.0772, 0.0711, 0.0896, 0.1608, 0.1193, 0.0699, 0.0561, 0.1458], device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0183, 0.0135, 0.0125, 0.0135, 0.0140, 0.0115, 0.0138], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005, 0.0006], device='cuda:3') 2022-12-07 23:27:39,849 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 2022-12-07 23:27:48,761 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0616, 2.7665, 2.6088, 2.7262, 2.9863, 2.9608, 3.0052, 3.0203], device='cuda:3'), covar=tensor([0.1031, 0.0838, 0.2347, 0.2797, 0.0932, 0.1042, 0.1436, 0.0949], device='cuda:3'), in_proj_covar=tensor([0.0355, 0.0243, 0.0412, 0.0533, 0.0311, 0.0399, 0.0378, 0.0342], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 23:27:59,960 INFO [train.py:873] (3/4) Epoch 11, batch 800, loss[loss=0.1745, simple_loss=0.1826, pruned_loss=0.08321, over 9503.00 frames. ], tot_loss[loss=0.1322, simple_loss=0.1602, pruned_loss=0.05211, over 1944253.68 frames. ], batch size: 100, lr: 7.34e-03, grad_scale: 8.0 2022-12-07 23:28:20,813 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2022-12-07 23:28:29,451 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.454e+02 2.357e+02 2.806e+02 3.598e+02 6.382e+02, threshold=5.612e+02, percent-clipped=5.0 2022-12-07 23:29:26,691 INFO [train.py:873] (3/4) Epoch 11, batch 900, loss[loss=0.1186, simple_loss=0.1482, pruned_loss=0.04448, over 14290.00 frames. ], tot_loss[loss=0.1332, simple_loss=0.1609, pruned_loss=0.05272, over 2010909.09 frames. ], batch size: 35, lr: 7.34e-03, grad_scale: 8.0 2022-12-07 23:29:56,493 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.334e+02 2.233e+02 2.846e+02 3.366e+02 5.399e+02, threshold=5.691e+02, percent-clipped=0.0 2022-12-07 23:30:54,356 INFO [train.py:873] (3/4) Epoch 11, batch 1000, loss[loss=0.1304, simple_loss=0.1617, pruned_loss=0.0496, over 14286.00 frames. ], tot_loss[loss=0.1333, simple_loss=0.161, pruned_loss=0.05277, over 1959232.95 frames. ], batch size: 39, lr: 7.33e-03, grad_scale: 8.0 2022-12-07 23:30:57,823 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=76625.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 23:31:00,097 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.55 vs. limit=5.0 2022-12-07 23:31:03,263 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7137, 1.3160, 3.2834, 3.0121, 3.1559, 3.3074, 2.4614, 3.2495], device='cuda:3'), covar=tensor([0.1931, 0.2222, 0.0215, 0.0419, 0.0386, 0.0240, 0.0530, 0.0270], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0157, 0.0122, 0.0165, 0.0140, 0.0134, 0.0115, 0.0116], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 23:31:19,959 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=76650.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:31:24,209 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.288e+02 2.175e+02 2.720e+02 3.257e+02 6.419e+02, threshold=5.440e+02, percent-clipped=2.0 2022-12-07 23:31:30,901 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7678, 1.6285, 1.8527, 1.4651, 1.8509, 1.0671, 1.8219, 1.9583], device='cuda:3'), covar=tensor([0.1314, 0.2323, 0.1072, 0.2143, 0.1330, 0.0862, 0.1156, 0.1112], device='cuda:3'), in_proj_covar=tensor([0.0024, 0.0025, 0.0028, 0.0024, 0.0026, 0.0037, 0.0026, 0.0028], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001], device='cuda:3') 2022-12-07 23:31:44,627 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=76679.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 23:32:21,332 INFO [train.py:873] (3/4) Epoch 11, batch 1100, loss[loss=0.128, simple_loss=0.1622, pruned_loss=0.04687, over 14270.00 frames. ], tot_loss[loss=0.1328, simple_loss=0.1606, pruned_loss=0.05244, over 1921702.63 frames. ], batch size: 35, lr: 7.33e-03, grad_scale: 8.0 2022-12-07 23:32:50,793 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.125e+02 2.272e+02 2.937e+02 3.615e+02 8.026e+02, threshold=5.874e+02, percent-clipped=3.0 2022-12-07 23:33:49,947 INFO [train.py:873] (3/4) Epoch 11, batch 1200, loss[loss=0.1547, simple_loss=0.1723, pruned_loss=0.06861, over 14174.00 frames. ], tot_loss[loss=0.1325, simple_loss=0.1611, pruned_loss=0.05193, over 2024063.59 frames. ], batch size: 99, lr: 7.32e-03, grad_scale: 8.0 2022-12-07 23:34:00,359 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2056, 1.8592, 2.1743, 2.3167, 1.9840, 1.9934, 2.2246, 2.1377], device='cuda:3'), covar=tensor([0.0211, 0.0436, 0.0194, 0.0175, 0.0305, 0.0504, 0.0242, 0.0235], device='cuda:3'), in_proj_covar=tensor([0.0277, 0.0249, 0.0369, 0.0314, 0.0259, 0.0299, 0.0290, 0.0278], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 23:34:05,528 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=76839.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:34:14,962 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.76 vs. limit=2.0 2022-12-07 23:34:19,397 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.367e+02 2.391e+02 2.914e+02 3.659e+02 7.427e+02, threshold=5.827e+02, percent-clipped=2.0 2022-12-07 23:34:59,331 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=76900.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:35:04,555 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.0679, 1.0278, 1.0423, 0.8642, 0.8628, 0.6573, 0.7860, 0.7666], device='cuda:3'), covar=tensor([0.0158, 0.0147, 0.0131, 0.0163, 0.0159, 0.0346, 0.0211, 0.0334], device='cuda:3'), in_proj_covar=tensor([0.0014, 0.0016, 0.0014, 0.0014, 0.0014, 0.0024, 0.0019, 0.0024], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2022-12-07 23:35:09,215 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2137, 2.1605, 3.1201, 3.1794, 3.1759, 2.1928, 3.1264, 2.5540], device='cuda:3'), covar=tensor([0.0315, 0.0755, 0.0618, 0.0434, 0.0327, 0.1097, 0.0334, 0.0718], device='cuda:3'), in_proj_covar=tensor([0.0277, 0.0249, 0.0367, 0.0314, 0.0259, 0.0298, 0.0290, 0.0277], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 23:35:17,348 INFO [train.py:873] (3/4) Epoch 11, batch 1300, loss[loss=0.1174, simple_loss=0.152, pruned_loss=0.04142, over 14167.00 frames. ], tot_loss[loss=0.1317, simple_loss=0.1606, pruned_loss=0.05139, over 2030662.18 frames. ], batch size: 29, lr: 7.32e-03, grad_scale: 8.0 2022-12-07 23:35:20,313 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2022-12-07 23:35:21,022 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=76925.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 23:35:42,572 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=76950.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:35:44,724 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9165, 3.2319, 2.9991, 3.2368, 2.4578, 3.2353, 3.1150, 1.4403], device='cuda:3'), covar=tensor([0.2099, 0.0903, 0.0821, 0.0694, 0.1026, 0.0634, 0.1131, 0.2773], device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0076, 0.0059, 0.0064, 0.0091, 0.0074, 0.0094, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005, 0.0006, 0.0006], device='cuda:3') 2022-12-07 23:35:47,102 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.371e+02 2.104e+02 2.573e+02 3.185e+02 4.865e+02, threshold=5.146e+02, percent-clipped=0.0 2022-12-07 23:35:48,135 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4478, 1.4039, 1.7259, 1.3303, 1.5094, 1.1282, 1.6635, 1.8671], device='cuda:3'), covar=tensor([0.2405, 0.1305, 0.1103, 0.1725, 0.1499, 0.1009, 0.0996, 0.1278], device='cuda:3'), in_proj_covar=tensor([0.0025, 0.0025, 0.0028, 0.0025, 0.0026, 0.0038, 0.0027, 0.0028], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001], device='cuda:3') 2022-12-07 23:35:52,680 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.14 vs. limit=5.0 2022-12-07 23:36:02,814 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=76973.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 23:36:08,405 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=76979.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 23:36:21,880 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.64 vs. limit=2.0 2022-12-07 23:36:24,629 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=76998.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:36:30,271 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4954, 1.5054, 1.4946, 1.3388, 1.5763, 1.0023, 1.5735, 1.8672], device='cuda:3'), covar=tensor([0.1679, 0.0807, 0.1254, 0.1370, 0.0969, 0.1113, 0.1011, 0.0677], device='cuda:3'), in_proj_covar=tensor([0.0025, 0.0025, 0.0028, 0.0025, 0.0026, 0.0038, 0.0026, 0.0028], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001], device='cuda:3') 2022-12-07 23:36:45,317 INFO [train.py:873] (3/4) Epoch 11, batch 1400, loss[loss=0.1302, simple_loss=0.1484, pruned_loss=0.05603, over 5015.00 frames. ], tot_loss[loss=0.1315, simple_loss=0.1605, pruned_loss=0.05127, over 2026650.24 frames. ], batch size: 100, lr: 7.31e-03, grad_scale: 8.0 2022-12-07 23:36:50,496 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77027.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 23:36:52,116 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.5428, 1.1526, 2.0747, 1.8590, 1.8734, 2.0884, 1.3902, 2.0793], device='cuda:3'), covar=tensor([0.0585, 0.0972, 0.0141, 0.0315, 0.0365, 0.0168, 0.0495, 0.0169], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0159, 0.0124, 0.0167, 0.0141, 0.0134, 0.0116, 0.0118], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 23:37:14,871 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.323e+02 2.332e+02 2.864e+02 3.537e+02 8.782e+02, threshold=5.729e+02, percent-clipped=8.0 2022-12-07 23:37:49,629 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77095.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:38:12,199 INFO [train.py:873] (3/4) Epoch 11, batch 1500, loss[loss=0.1451, simple_loss=0.1434, pruned_loss=0.07342, over 2650.00 frames. ], tot_loss[loss=0.1312, simple_loss=0.1597, pruned_loss=0.05133, over 2003814.94 frames. ], batch size: 100, lr: 7.31e-03, grad_scale: 8.0 2022-12-07 23:38:41,966 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.264e+02 2.474e+02 2.983e+02 4.107e+02 8.785e+02, threshold=5.967e+02, percent-clipped=9.0 2022-12-07 23:38:43,013 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77156.0, num_to_drop=1, layers_to_drop={1} 2022-12-07 23:39:17,239 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77195.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:39:32,240 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77211.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:39:40,747 INFO [train.py:873] (3/4) Epoch 11, batch 1600, loss[loss=0.1287, simple_loss=0.1639, pruned_loss=0.0468, over 14276.00 frames. ], tot_loss[loss=0.131, simple_loss=0.1597, pruned_loss=0.05115, over 2039206.45 frames. ], batch size: 76, lr: 7.31e-03, grad_scale: 8.0 2022-12-07 23:40:03,107 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.66 vs. limit=2.0 2022-12-07 23:40:10,413 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.174e+02 1.982e+02 2.587e+02 3.263e+02 7.799e+02, threshold=5.175e+02, percent-clipped=3.0 2022-12-07 23:40:26,211 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77272.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:40:26,995 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77273.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:40:30,516 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77277.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:40:51,553 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9782, 3.4575, 2.7802, 4.1373, 3.9690, 3.9885, 3.5717, 2.7583], device='cuda:3'), covar=tensor([0.1248, 0.1726, 0.4026, 0.0692, 0.0815, 0.1569, 0.1316, 0.4005], device='cuda:3'), in_proj_covar=tensor([0.0267, 0.0299, 0.0271, 0.0251, 0.0307, 0.0291, 0.0258, 0.0254], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2022-12-07 23:41:05,681 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9855, 2.5426, 5.0211, 3.2496, 4.6938, 2.4025, 3.8817, 4.6259], device='cuda:3'), covar=tensor([0.0391, 0.4042, 0.0304, 0.7250, 0.0437, 0.3303, 0.1064, 0.0358], device='cuda:3'), in_proj_covar=tensor([0.0247, 0.0218, 0.0201, 0.0296, 0.0219, 0.0220, 0.0216, 0.0207], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 23:41:08,997 INFO [train.py:873] (3/4) Epoch 11, batch 1700, loss[loss=0.155, simple_loss=0.1736, pruned_loss=0.06826, over 9487.00 frames. ], tot_loss[loss=0.1308, simple_loss=0.1597, pruned_loss=0.05097, over 2101353.97 frames. ], batch size: 100, lr: 7.30e-03, grad_scale: 8.0 2022-12-07 23:41:20,905 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77334.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:41:24,009 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77338.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:41:39,033 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.284e+02 2.095e+02 2.641e+02 3.320e+02 6.995e+02, threshold=5.282e+02, percent-clipped=0.0 2022-12-07 23:41:40,590 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.48 vs. limit=5.0 2022-12-07 23:42:10,147 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.79 vs. limit=2.0 2022-12-07 23:42:37,344 INFO [train.py:873] (3/4) Epoch 11, batch 1800, loss[loss=0.1453, simple_loss=0.1309, pruned_loss=0.07981, over 2631.00 frames. ], tot_loss[loss=0.132, simple_loss=0.1602, pruned_loss=0.05193, over 1992594.44 frames. ], batch size: 100, lr: 7.30e-03, grad_scale: 8.0 2022-12-07 23:42:44,533 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8484, 1.3685, 2.0022, 1.2417, 1.9733, 2.0513, 1.6812, 2.1166], device='cuda:3'), covar=tensor([0.0275, 0.1861, 0.0368, 0.1776, 0.0476, 0.0476, 0.1027, 0.0297], device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0156, 0.0155, 0.0167, 0.0169, 0.0173, 0.0131, 0.0138], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-07 23:43:04,171 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77451.0, num_to_drop=1, layers_to_drop={3} 2022-12-07 23:43:07,257 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.178e+02 2.136e+02 2.704e+02 3.283e+02 4.688e+02, threshold=5.409e+02, percent-clipped=1.0 2022-12-07 23:43:12,379 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1129, 1.9758, 2.0366, 2.1443, 2.0740, 2.0637, 2.1909, 1.8210], device='cuda:3'), covar=tensor([0.0790, 0.1204, 0.0776, 0.0749, 0.0888, 0.0624, 0.0881, 0.0725], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0264, 0.0185, 0.0180, 0.0177, 0.0144, 0.0266, 0.0159], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-07 23:43:42,429 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77495.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:44:05,124 INFO [train.py:873] (3/4) Epoch 11, batch 1900, loss[loss=0.1184, simple_loss=0.1505, pruned_loss=0.04311, over 14257.00 frames. ], tot_loss[loss=0.1322, simple_loss=0.1604, pruned_loss=0.05201, over 1973599.46 frames. ], batch size: 57, lr: 7.29e-03, grad_scale: 8.0 2022-12-07 23:44:24,704 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77543.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:44:36,351 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.774e+01 2.327e+02 2.859e+02 3.630e+02 7.780e+02, threshold=5.718e+02, percent-clipped=3.0 2022-12-07 23:44:36,572 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9640, 2.6847, 3.7172, 2.6218, 2.1728, 2.9210, 1.7673, 2.7462], device='cuda:3'), covar=tensor([0.1327, 0.1242, 0.0528, 0.2430, 0.2478, 0.1036, 0.3945, 0.1425], device='cuda:3'), in_proj_covar=tensor([0.0080, 0.0096, 0.0091, 0.0096, 0.0114, 0.0083, 0.0127, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-07 23:44:45,959 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77567.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:44:48,444 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.9359, 0.9618, 1.0280, 0.8650, 0.9538, 0.5844, 0.8803, 1.0423], device='cuda:3'), covar=tensor([0.0583, 0.0824, 0.0475, 0.0652, 0.0527, 0.0704, 0.0815, 0.0701], device='cuda:3'), in_proj_covar=tensor([0.0025, 0.0026, 0.0028, 0.0025, 0.0027, 0.0038, 0.0027, 0.0029], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001], device='cuda:3') 2022-12-07 23:45:29,334 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77616.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:45:33,422 INFO [train.py:873] (3/4) Epoch 11, batch 2000, loss[loss=0.1075, simple_loss=0.147, pruned_loss=0.03403, over 14359.00 frames. ], tot_loss[loss=0.1329, simple_loss=0.1612, pruned_loss=0.05231, over 1976075.32 frames. ], batch size: 31, lr: 7.29e-03, grad_scale: 8.0 2022-12-07 23:45:40,534 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77629.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:45:44,070 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77633.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:46:04,192 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.424e+02 2.333e+02 2.909e+02 3.438e+02 9.067e+02, threshold=5.818e+02, percent-clipped=5.0 2022-12-07 23:46:12,114 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.9607, 1.0383, 0.9578, 1.0192, 1.0036, 0.6396, 0.9694, 1.0421], device='cuda:3'), covar=tensor([0.0522, 0.0846, 0.0642, 0.0575, 0.0583, 0.0788, 0.0827, 0.0835], device='cuda:3'), in_proj_covar=tensor([0.0025, 0.0026, 0.0029, 0.0025, 0.0027, 0.0039, 0.0027, 0.0029], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001], device='cuda:3') 2022-12-07 23:46:21,857 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77677.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:46:21,876 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77677.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:46:31,629 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4721, 3.6606, 3.1695, 3.4334, 2.7022, 3.7622, 3.4766, 1.4660], device='cuda:3'), covar=tensor([0.2100, 0.0897, 0.1494, 0.1513, 0.1195, 0.0427, 0.1282, 0.2966], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0078, 0.0061, 0.0065, 0.0093, 0.0076, 0.0097, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0006, 0.0006], device='cuda:3') 2022-12-07 23:46:40,123 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2022-12-07 23:46:59,456 INFO [train.py:873] (3/4) Epoch 11, batch 2100, loss[loss=0.1299, simple_loss=0.128, pruned_loss=0.06588, over 2678.00 frames. ], tot_loss[loss=0.1329, simple_loss=0.1606, pruned_loss=0.05257, over 1911668.49 frames. ], batch size: 100, lr: 7.28e-03, grad_scale: 8.0 2022-12-07 23:47:00,857 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2022-12-07 23:47:07,548 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0960, 1.9645, 1.9907, 2.1501, 2.0775, 1.9728, 2.2077, 1.8647], device='cuda:3'), covar=tensor([0.0995, 0.1332, 0.0745, 0.0766, 0.0925, 0.0766, 0.0844, 0.0657], device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0259, 0.0183, 0.0180, 0.0174, 0.0143, 0.0265, 0.0158], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-07 23:47:14,269 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77738.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:47:25,614 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77751.0, num_to_drop=1, layers_to_drop={0} 2022-12-07 23:47:29,642 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.109e+02 2.408e+02 3.077e+02 4.325e+02 1.206e+03, threshold=6.153e+02, percent-clipped=10.0 2022-12-07 23:48:07,066 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77799.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:48:26,367 INFO [train.py:873] (3/4) Epoch 11, batch 2200, loss[loss=0.1441, simple_loss=0.1437, pruned_loss=0.07221, over 2573.00 frames. ], tot_loss[loss=0.1319, simple_loss=0.1605, pruned_loss=0.05165, over 1942503.71 frames. ], batch size: 100, lr: 7.28e-03, grad_scale: 8.0 2022-12-07 23:48:57,074 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.095e+02 2.100e+02 2.823e+02 3.472e+02 8.203e+02, threshold=5.646e+02, percent-clipped=4.0 2022-12-07 23:49:06,607 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77867.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:49:26,913 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9797, 1.8648, 3.1530, 2.3381, 2.9917, 1.9160, 2.3924, 2.9492], device='cuda:3'), covar=tensor([0.0994, 0.4411, 0.0582, 0.5191, 0.0884, 0.3303, 0.1404, 0.0785], device='cuda:3'), in_proj_covar=tensor([0.0250, 0.0223, 0.0202, 0.0297, 0.0223, 0.0221, 0.0217, 0.0209], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 23:49:48,086 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77915.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:49:53,347 INFO [train.py:873] (3/4) Epoch 11, batch 2300, loss[loss=0.1409, simple_loss=0.158, pruned_loss=0.06188, over 6908.00 frames. ], tot_loss[loss=0.131, simple_loss=0.1598, pruned_loss=0.05106, over 1954198.32 frames. ], batch size: 100, lr: 7.27e-03, grad_scale: 8.0 2022-12-07 23:50:00,472 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77929.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:50:04,135 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77933.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:50:21,944 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.77 vs. limit=5.0 2022-12-07 23:50:23,859 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.278e+02 2.313e+02 2.906e+02 3.798e+02 8.346e+02, threshold=5.811e+02, percent-clipped=6.0 2022-12-07 23:50:38,037 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77972.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:50:42,322 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77977.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:50:45,705 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77981.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:51:14,622 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.60 vs. limit=5.0 2022-12-07 23:51:21,250 INFO [train.py:873] (3/4) Epoch 11, batch 2400, loss[loss=0.1156, simple_loss=0.149, pruned_loss=0.04106, over 14275.00 frames. ], tot_loss[loss=0.1308, simple_loss=0.1596, pruned_loss=0.05099, over 1972796.54 frames. ], batch size: 63, lr: 7.27e-03, grad_scale: 8.0 2022-12-07 23:51:32,065 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78033.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:51:51,987 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.111e+02 2.203e+02 2.586e+02 3.154e+02 7.524e+02, threshold=5.173e+02, percent-clipped=2.0 2022-12-07 23:52:47,590 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9448, 3.2255, 3.0820, 3.2386, 2.4716, 3.1812, 2.9733, 1.4412], device='cuda:3'), covar=tensor([0.1971, 0.0753, 0.1252, 0.0715, 0.1046, 0.0638, 0.1273, 0.2780], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0077, 0.0061, 0.0066, 0.0093, 0.0075, 0.0096, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0006, 0.0006], device='cuda:3') 2022-12-07 23:52:49,056 INFO [train.py:873] (3/4) Epoch 11, batch 2500, loss[loss=0.1597, simple_loss=0.1472, pruned_loss=0.08609, over 1239.00 frames. ], tot_loss[loss=0.1298, simple_loss=0.1587, pruned_loss=0.05051, over 1916146.01 frames. ], batch size: 100, lr: 7.26e-03, grad_scale: 8.0 2022-12-07 23:53:19,543 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.400e+02 2.052e+02 2.774e+02 3.543e+02 6.822e+02, threshold=5.549e+02, percent-clipped=2.0 2022-12-07 23:53:23,354 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4606, 1.8820, 2.3913, 2.1510, 2.5047, 2.3621, 2.2639, 2.2029], device='cuda:3'), covar=tensor([0.0568, 0.2351, 0.0780, 0.1370, 0.0582, 0.0985, 0.0815, 0.1137], device='cuda:3'), in_proj_covar=tensor([0.0342, 0.0311, 0.0390, 0.0297, 0.0374, 0.0315, 0.0358, 0.0311], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 23:53:26,997 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.70 vs. limit=2.0 2022-12-07 23:53:31,278 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.3810, 0.8302, 1.2362, 0.8384, 1.0730, 1.3267, 1.1004, 1.1012], device='cuda:3'), covar=tensor([0.0454, 0.1051, 0.0640, 0.0524, 0.1224, 0.0710, 0.0411, 0.1240], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0182, 0.0134, 0.0124, 0.0134, 0.0139, 0.0116, 0.0138], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0007, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005, 0.0006], device='cuda:3') 2022-12-07 23:54:16,600 INFO [train.py:873] (3/4) Epoch 11, batch 2600, loss[loss=0.1263, simple_loss=0.1582, pruned_loss=0.04724, over 12752.00 frames. ], tot_loss[loss=0.1301, simple_loss=0.1584, pruned_loss=0.05096, over 1898344.17 frames. ], batch size: 100, lr: 7.26e-03, grad_scale: 8.0 2022-12-07 23:54:25,378 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5828, 3.3955, 2.9870, 2.2906, 3.0139, 3.4096, 3.7528, 2.8626], device='cuda:3'), covar=tensor([0.0616, 0.1421, 0.0985, 0.1620, 0.1216, 0.0654, 0.0761, 0.1361], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0180, 0.0132, 0.0123, 0.0132, 0.0138, 0.0114, 0.0136], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0007, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0006], device='cuda:3') 2022-12-07 23:54:26,495 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78232.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:54:47,974 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.357e+02 2.197e+02 2.874e+02 3.504e+02 7.274e+02, threshold=5.748e+02, percent-clipped=5.0 2022-12-07 23:55:00,964 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78272.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:55:19,436 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78293.0, num_to_drop=1, layers_to_drop={2} 2022-12-07 23:55:36,635 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2022-12-07 23:55:42,573 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=78320.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:55:43,478 INFO [train.py:873] (3/4) Epoch 11, batch 2700, loss[loss=0.1551, simple_loss=0.1809, pruned_loss=0.06462, over 14387.00 frames. ], tot_loss[loss=0.1322, simple_loss=0.1598, pruned_loss=0.05226, over 1930925.86 frames. ], batch size: 53, lr: 7.25e-03, grad_scale: 4.0 2022-12-07 23:55:54,568 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78333.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:56:15,226 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.898e+01 2.288e+02 2.843e+02 3.409e+02 7.759e+02, threshold=5.687e+02, percent-clipped=1.0 2022-12-07 23:56:28,423 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.76 vs. limit=2.0 2022-12-07 23:56:36,334 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=78381.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:57:11,630 INFO [train.py:873] (3/4) Epoch 11, batch 2800, loss[loss=0.137, simple_loss=0.1639, pruned_loss=0.05504, over 14420.00 frames. ], tot_loss[loss=0.132, simple_loss=0.1599, pruned_loss=0.05204, over 1960390.27 frames. ], batch size: 73, lr: 7.25e-03, grad_scale: 8.0 2022-12-07 23:57:29,385 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2625, 3.3998, 3.5590, 3.3226, 3.4339, 2.9475, 1.4140, 3.2584], device='cuda:3'), covar=tensor([0.0353, 0.0381, 0.0327, 0.0419, 0.0343, 0.0837, 0.2930, 0.0290], device='cuda:3'), in_proj_covar=tensor([0.0157, 0.0167, 0.0138, 0.0138, 0.0196, 0.0134, 0.0156, 0.0183], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-07 23:57:43,401 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.203e+02 2.311e+02 2.686e+02 3.123e+02 7.325e+02, threshold=5.372e+02, percent-clipped=4.0 2022-12-07 23:57:46,063 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78460.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:58:13,964 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2022-12-07 23:58:40,123 INFO [train.py:873] (3/4) Epoch 11, batch 2900, loss[loss=0.1231, simple_loss=0.1542, pruned_loss=0.04601, over 6947.00 frames. ], tot_loss[loss=0.1328, simple_loss=0.1604, pruned_loss=0.05258, over 1964736.04 frames. ], batch size: 100, lr: 7.24e-03, grad_scale: 8.0 2022-12-07 23:58:40,258 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78521.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:59:10,516 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78556.0, num_to_drop=0, layers_to_drop=set() 2022-12-07 23:59:11,185 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.326e+02 2.237e+02 2.916e+02 3.680e+02 6.575e+02, threshold=5.831e+02, percent-clipped=6.0 2022-12-07 23:59:28,844 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9789, 2.5368, 3.7789, 2.8977, 3.8017, 3.6817, 3.5613, 3.0965], device='cuda:3'), covar=tensor([0.0803, 0.3156, 0.1027, 0.1993, 0.0961, 0.1040, 0.1676, 0.1914], device='cuda:3'), in_proj_covar=tensor([0.0353, 0.0320, 0.0400, 0.0306, 0.0383, 0.0324, 0.0367, 0.0320], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-07 23:59:38,766 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78588.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 00:00:05,375 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78617.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:00:08,927 INFO [train.py:873] (3/4) Epoch 11, batch 3000, loss[loss=0.1122, simple_loss=0.1467, pruned_loss=0.03889, over 14406.00 frames. ], tot_loss[loss=0.1327, simple_loss=0.1608, pruned_loss=0.05227, over 1993838.46 frames. ], batch size: 53, lr: 7.24e-03, grad_scale: 8.0 2022-12-08 00:00:08,927 INFO [train.py:896] (3/4) Computing validation loss 2022-12-08 00:00:17,382 INFO [train.py:905] (3/4) Epoch 11, validation: loss=0.1282, simple_loss=0.1681, pruned_loss=0.04413, over 857387.00 frames. 2022-12-08 00:00:17,383 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-08 00:00:48,734 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.362e+02 2.287e+02 2.909e+02 3.866e+02 8.652e+02, threshold=5.818e+02, percent-clipped=3.0 2022-12-08 00:01:31,319 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7572, 1.7508, 1.6128, 1.4998, 2.1421, 1.8380, 1.6609, 1.7273], device='cuda:3'), covar=tensor([0.0383, 0.0978, 0.0768, 0.0823, 0.0302, 0.0609, 0.0713, 0.0840], device='cuda:3'), in_proj_covar=tensor([0.0015, 0.0016, 0.0014, 0.0015, 0.0015, 0.0024, 0.0019, 0.0025], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2022-12-08 00:01:46,387 INFO [train.py:873] (3/4) Epoch 11, batch 3100, loss[loss=0.1216, simple_loss=0.1405, pruned_loss=0.05137, over 5988.00 frames. ], tot_loss[loss=0.1314, simple_loss=0.1597, pruned_loss=0.05153, over 1902366.63 frames. ], batch size: 100, lr: 7.24e-03, grad_scale: 8.0 2022-12-08 00:02:05,310 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2022-12-08 00:02:18,621 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.341e+02 2.078e+02 2.524e+02 3.137e+02 7.110e+02, threshold=5.049e+02, percent-clipped=1.0 2022-12-08 00:03:11,341 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78816.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:03:15,285 INFO [train.py:873] (3/4) Epoch 11, batch 3200, loss[loss=0.13, simple_loss=0.1645, pruned_loss=0.04778, over 14067.00 frames. ], tot_loss[loss=0.1314, simple_loss=0.1597, pruned_loss=0.05153, over 1924035.18 frames. ], batch size: 22, lr: 7.23e-03, grad_scale: 8.0 2022-12-08 00:03:46,996 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.001e+02 2.170e+02 2.756e+02 3.335e+02 6.522e+02, threshold=5.512e+02, percent-clipped=3.0 2022-12-08 00:04:14,696 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78888.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:04:35,328 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78912.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:04:43,623 INFO [train.py:873] (3/4) Epoch 11, batch 3300, loss[loss=0.1054, simple_loss=0.1468, pruned_loss=0.03203, over 14321.00 frames. ], tot_loss[loss=0.1308, simple_loss=0.159, pruned_loss=0.0513, over 1868587.54 frames. ], batch size: 25, lr: 7.23e-03, grad_scale: 8.0 2022-12-08 00:04:56,440 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=78936.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:05:15,456 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.234e+02 2.245e+02 2.789e+02 3.616e+02 5.775e+02, threshold=5.578e+02, percent-clipped=1.0 2022-12-08 00:05:22,659 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78965.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:05:24,410 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9565, 3.1998, 3.0928, 3.2115, 2.4642, 3.1324, 2.9776, 1.6748], device='cuda:3'), covar=tensor([0.1796, 0.0686, 0.0906, 0.0709, 0.0981, 0.0531, 0.1145, 0.2308], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0079, 0.0062, 0.0066, 0.0094, 0.0076, 0.0098, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0006, 0.0006], device='cuda:3') 2022-12-08 00:05:37,243 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3080, 1.9105, 2.2985, 2.3985, 2.1810, 1.9796, 2.4063, 2.1185], device='cuda:3'), covar=tensor([0.0284, 0.0656, 0.0337, 0.0304, 0.0388, 0.0753, 0.0354, 0.0410], device='cuda:3'), in_proj_covar=tensor([0.0279, 0.0252, 0.0372, 0.0319, 0.0259, 0.0301, 0.0293, 0.0279], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 00:06:10,657 INFO [train.py:873] (3/4) Epoch 11, batch 3400, loss[loss=0.1212, simple_loss=0.1556, pruned_loss=0.04337, over 14404.00 frames. ], tot_loss[loss=0.1316, simple_loss=0.1593, pruned_loss=0.05193, over 1887163.45 frames. ], batch size: 53, lr: 7.22e-03, grad_scale: 4.0 2022-12-08 00:06:11,068 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.12 vs. limit=2.0 2022-12-08 00:06:12,146 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.07 vs. limit=5.0 2022-12-08 00:06:15,711 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79026.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:06:43,352 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.385e+02 2.208e+02 2.821e+02 3.585e+02 7.488e+02, threshold=5.641e+02, percent-clipped=3.0 2022-12-08 00:07:02,110 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5970, 1.9962, 3.7028, 2.6398, 3.5508, 1.9083, 2.8216, 3.5226], device='cuda:3'), covar=tensor([0.0751, 0.4605, 0.0490, 0.6173, 0.0734, 0.3975, 0.1439, 0.0481], device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0217, 0.0199, 0.0292, 0.0220, 0.0221, 0.0215, 0.0204], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 00:07:28,821 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2022-12-08 00:07:34,181 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79116.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:07:38,505 INFO [train.py:873] (3/4) Epoch 11, batch 3500, loss[loss=0.1755, simple_loss=0.1548, pruned_loss=0.09811, over 1293.00 frames. ], tot_loss[loss=0.1311, simple_loss=0.1595, pruned_loss=0.05134, over 1965258.71 frames. ], batch size: 100, lr: 7.22e-03, grad_scale: 4.0 2022-12-08 00:08:10,356 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.248e+02 2.398e+02 2.883e+02 3.497e+02 6.738e+02, threshold=5.767e+02, percent-clipped=3.0 2022-12-08 00:08:15,412 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79164.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:08:57,383 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79212.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:09:04,926 INFO [train.py:873] (3/4) Epoch 11, batch 3600, loss[loss=0.1332, simple_loss=0.1565, pruned_loss=0.05494, over 10311.00 frames. ], tot_loss[loss=0.132, simple_loss=0.16, pruned_loss=0.05197, over 1936668.39 frames. ], batch size: 100, lr: 7.21e-03, grad_scale: 8.0 2022-12-08 00:09:30,354 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=8.16 vs. limit=5.0 2022-12-08 00:09:37,406 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.384e+02 2.170e+02 2.534e+02 3.229e+02 7.657e+02, threshold=5.069e+02, percent-clipped=4.0 2022-12-08 00:09:39,213 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79260.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:09:44,354 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2022-12-08 00:10:01,682 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79285.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 00:10:32,864 INFO [train.py:873] (3/4) Epoch 11, batch 3700, loss[loss=0.1342, simple_loss=0.159, pruned_loss=0.0547, over 10373.00 frames. ], tot_loss[loss=0.1325, simple_loss=0.1604, pruned_loss=0.05233, over 1945607.53 frames. ], batch size: 100, lr: 7.21e-03, grad_scale: 8.0 2022-12-08 00:10:32,977 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79321.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:10:43,131 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2266, 2.0232, 2.1605, 2.2279, 2.1610, 2.1472, 2.2958, 1.8949], device='cuda:3'), covar=tensor([0.0772, 0.1515, 0.0727, 0.0848, 0.0898, 0.0613, 0.0902, 0.0789], device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0261, 0.0181, 0.0179, 0.0174, 0.0142, 0.0265, 0.0161], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 00:10:43,193 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79333.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:10:54,250 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79346.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 00:11:04,340 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.257e+02 2.300e+02 2.713e+02 3.691e+02 7.855e+02, threshold=5.426e+02, percent-clipped=7.0 2022-12-08 00:11:09,835 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79364.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:11:19,320 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79375.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 00:11:35,725 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79394.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:11:58,467 INFO [train.py:873] (3/4) Epoch 11, batch 3800, loss[loss=0.114, simple_loss=0.147, pruned_loss=0.04049, over 14347.00 frames. ], tot_loss[loss=0.1327, simple_loss=0.1606, pruned_loss=0.05239, over 1945456.87 frames. ], batch size: 73, lr: 7.20e-03, grad_scale: 8.0 2022-12-08 00:12:02,436 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79425.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:12:11,689 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79436.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 00:12:30,563 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.086e+02 2.258e+02 2.922e+02 3.679e+02 8.434e+02, threshold=5.844e+02, percent-clipped=6.0 2022-12-08 00:12:37,023 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8281, 1.3658, 3.1521, 1.4560, 2.9705, 3.0320, 2.3383, 3.1864], device='cuda:3'), covar=tensor([0.0421, 0.4117, 0.0468, 0.2816, 0.1123, 0.0612, 0.1066, 0.0359], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0157, 0.0157, 0.0168, 0.0170, 0.0174, 0.0133, 0.0142], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-08 00:12:47,874 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3758, 1.6913, 2.6665, 2.1379, 2.4633, 1.7447, 2.1095, 2.2519], device='cuda:3'), covar=tensor([0.1376, 0.3397, 0.0423, 0.3330, 0.0920, 0.2602, 0.1044, 0.0667], device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0214, 0.0196, 0.0291, 0.0216, 0.0218, 0.0212, 0.0202], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 00:13:17,930 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1069, 1.3840, 4.0654, 1.8228, 3.9093, 4.0963, 3.3883, 4.3904], device='cuda:3'), covar=tensor([0.0223, 0.3522, 0.0380, 0.2340, 0.0507, 0.0430, 0.0575, 0.0225], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0158, 0.0158, 0.0169, 0.0171, 0.0175, 0.0134, 0.0143], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-08 00:13:26,571 INFO [train.py:873] (3/4) Epoch 11, batch 3900, loss[loss=0.1095, simple_loss=0.1539, pruned_loss=0.03255, over 14305.00 frames. ], tot_loss[loss=0.1312, simple_loss=0.1596, pruned_loss=0.05143, over 1965365.18 frames. ], batch size: 37, lr: 7.20e-03, grad_scale: 8.0 2022-12-08 00:13:33,926 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.31 vs. limit=2.0 2022-12-08 00:13:59,194 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.032e+02 2.132e+02 2.623e+02 3.165e+02 6.507e+02, threshold=5.247e+02, percent-clipped=1.0 2022-12-08 00:14:23,101 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79585.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:14:54,983 INFO [train.py:873] (3/4) Epoch 11, batch 4000, loss[loss=0.1261, simple_loss=0.1482, pruned_loss=0.052, over 6006.00 frames. ], tot_loss[loss=0.131, simple_loss=0.1594, pruned_loss=0.05125, over 1964657.41 frames. ], batch size: 100, lr: 7.19e-03, grad_scale: 8.0 2022-12-08 00:14:55,174 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79621.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:15:12,968 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79641.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 00:15:17,574 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79646.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:15:26,984 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79657.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:15:27,677 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.890e+01 2.190e+02 2.797e+02 3.644e+02 7.365e+02, threshold=5.595e+02, percent-clipped=3.0 2022-12-08 00:15:37,276 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79669.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:15:40,589 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.25 vs. limit=5.0 2022-12-08 00:15:55,385 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79689.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:16:21,100 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79718.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:16:22,997 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79720.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:16:23,756 INFO [train.py:873] (3/4) Epoch 11, batch 4100, loss[loss=0.1592, simple_loss=0.1549, pruned_loss=0.08171, over 1246.00 frames. ], tot_loss[loss=0.1303, simple_loss=0.1589, pruned_loss=0.05081, over 1943891.69 frames. ], batch size: 100, lr: 7.19e-03, grad_scale: 8.0 2022-12-08 00:16:32,682 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79731.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 00:16:56,454 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.197e+02 2.218e+02 2.886e+02 3.615e+02 6.897e+02, threshold=5.771e+02, percent-clipped=4.0 2022-12-08 00:17:00,596 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79762.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:17:24,009 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1604, 2.6616, 3.6090, 2.4064, 2.1585, 3.0355, 1.6397, 2.9559], device='cuda:3'), covar=tensor([0.1023, 0.1058, 0.0525, 0.2331, 0.2444, 0.1076, 0.4178, 0.0931], device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0096, 0.0092, 0.0095, 0.0116, 0.0083, 0.0126, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 00:17:26,072 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1998, 2.8658, 2.8731, 1.9762, 2.7698, 2.8953, 3.2513, 2.6730], device='cuda:3'), covar=tensor([0.0702, 0.1105, 0.1167, 0.1834, 0.0914, 0.0861, 0.0643, 0.1518], device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0181, 0.0136, 0.0125, 0.0132, 0.0141, 0.0117, 0.0138], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0006], device='cuda:3') 2022-12-08 00:17:52,402 INFO [train.py:873] (3/4) Epoch 11, batch 4200, loss[loss=0.1938, simple_loss=0.1807, pruned_loss=0.1035, over 1211.00 frames. ], tot_loss[loss=0.1309, simple_loss=0.1594, pruned_loss=0.05125, over 1957090.36 frames. ], batch size: 100, lr: 7.19e-03, grad_scale: 8.0 2022-12-08 00:17:54,366 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79823.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:18:25,211 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.418e+02 2.219e+02 2.651e+02 3.356e+02 1.038e+03, threshold=5.303e+02, percent-clipped=1.0 2022-12-08 00:18:50,317 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.2518, 5.0257, 4.8307, 5.2700, 4.9491, 4.6078, 5.3220, 5.1676], device='cuda:3'), covar=tensor([0.0581, 0.0583, 0.0644, 0.0430, 0.0665, 0.0492, 0.0501, 0.0529], device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0125, 0.0131, 0.0141, 0.0132, 0.0110, 0.0152, 0.0130], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 00:19:05,841 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8185, 3.3002, 3.0984, 3.1862, 2.2832, 3.3332, 3.0432, 1.4283], device='cuda:3'), covar=tensor([0.2937, 0.0954, 0.1206, 0.1037, 0.1350, 0.0584, 0.1270, 0.3193], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0075, 0.0060, 0.0062, 0.0091, 0.0072, 0.0093, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005, 0.0006, 0.0006], device='cuda:3') 2022-12-08 00:19:06,678 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8312, 1.8330, 1.6007, 1.9923, 1.8065, 1.8875, 1.8391, 1.6603], device='cuda:3'), covar=tensor([0.1097, 0.0949, 0.1973, 0.0586, 0.0993, 0.0681, 0.1275, 0.0888], device='cuda:3'), in_proj_covar=tensor([0.0265, 0.0291, 0.0267, 0.0246, 0.0303, 0.0289, 0.0250, 0.0251], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2022-12-08 00:19:21,013 INFO [train.py:873] (3/4) Epoch 11, batch 4300, loss[loss=0.152, simple_loss=0.177, pruned_loss=0.06348, over 12751.00 frames. ], tot_loss[loss=0.1311, simple_loss=0.1594, pruned_loss=0.05138, over 1910048.47 frames. ], batch size: 100, lr: 7.18e-03, grad_scale: 8.0 2022-12-08 00:19:38,665 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79941.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:19:38,726 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79941.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 00:19:53,915 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.285e+02 2.309e+02 2.745e+02 3.460e+02 6.233e+02, threshold=5.491e+02, percent-clipped=1.0 2022-12-08 00:20:13,543 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4188, 3.8442, 4.1033, 4.4599, 4.1260, 3.9471, 4.3639, 3.6513], device='cuda:3'), covar=tensor([0.1147, 0.2034, 0.0942, 0.0965, 0.1326, 0.1114, 0.1188, 0.1258], device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0255, 0.0178, 0.0176, 0.0171, 0.0141, 0.0261, 0.0157], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 00:20:15,327 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.0609, 1.0099, 1.0415, 0.9760, 1.1724, 0.5829, 0.8482, 0.9944], device='cuda:3'), covar=tensor([0.0808, 0.0941, 0.0623, 0.0867, 0.0614, 0.0845, 0.1311, 0.0879], device='cuda:3'), in_proj_covar=tensor([0.0025, 0.0026, 0.0029, 0.0025, 0.0027, 0.0039, 0.0027, 0.0029], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2022-12-08 00:20:21,853 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79989.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 00:20:21,931 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79989.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:20:47,216 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80013.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:20:52,319 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9008, 1.4911, 2.9718, 2.6437, 2.8745, 2.9644, 2.1894, 2.9378], device='cuda:3'), covar=tensor([0.1065, 0.1208, 0.0153, 0.0385, 0.0290, 0.0158, 0.0485, 0.0172], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0159, 0.0125, 0.0167, 0.0142, 0.0135, 0.0117, 0.0117], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 00:20:53,268 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80020.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:20:53,977 INFO [train.py:873] (3/4) Epoch 11, batch 4400, loss[loss=0.1065, simple_loss=0.1474, pruned_loss=0.03281, over 14338.00 frames. ], tot_loss[loss=0.1304, simple_loss=0.1592, pruned_loss=0.05079, over 1960901.23 frames. ], batch size: 39, lr: 7.18e-03, grad_scale: 8.0 2022-12-08 00:21:03,149 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80031.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 00:21:08,517 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80037.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:21:26,666 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.889e+01 2.388e+02 2.902e+02 3.759e+02 8.737e+02, threshold=5.804e+02, percent-clipped=8.0 2022-12-08 00:21:35,390 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80068.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:21:44,981 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80079.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 00:21:51,730 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3931, 3.1460, 2.9763, 1.9954, 2.8264, 3.0879, 3.3880, 2.6342], device='cuda:3'), covar=tensor([0.0698, 0.1453, 0.1053, 0.1886, 0.1079, 0.0702, 0.0787, 0.1627], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0180, 0.0134, 0.0123, 0.0131, 0.0139, 0.0117, 0.0137], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0007, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0006], device='cuda:3') 2022-12-08 00:22:19,556 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80118.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:22:21,959 INFO [train.py:873] (3/4) Epoch 11, batch 4500, loss[loss=0.1315, simple_loss=0.1334, pruned_loss=0.06481, over 2600.00 frames. ], tot_loss[loss=0.1287, simple_loss=0.1584, pruned_loss=0.04952, over 2037390.23 frames. ], batch size: 100, lr: 7.17e-03, grad_scale: 8.0 2022-12-08 00:22:24,929 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1157, 3.0094, 2.9512, 3.2615, 2.8274, 2.8234, 3.1994, 3.1333], device='cuda:3'), covar=tensor([0.0888, 0.1096, 0.0911, 0.0715, 0.1109, 0.0903, 0.0831, 0.0799], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0128, 0.0134, 0.0145, 0.0136, 0.0113, 0.0157, 0.0133], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 00:22:54,367 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.548e+02 2.254e+02 2.921e+02 3.671e+02 7.258e+02, threshold=5.842e+02, percent-clipped=4.0 2022-12-08 00:23:41,769 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5437, 1.5104, 2.8562, 1.4826, 2.8511, 2.7464, 2.0413, 2.9189], device='cuda:3'), covar=tensor([0.0261, 0.2303, 0.0316, 0.1734, 0.0364, 0.0431, 0.0907, 0.0216], device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0158, 0.0160, 0.0168, 0.0173, 0.0175, 0.0135, 0.0144], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-08 00:23:48,495 INFO [train.py:873] (3/4) Epoch 11, batch 4600, loss[loss=0.1111, simple_loss=0.1382, pruned_loss=0.04194, over 4986.00 frames. ], tot_loss[loss=0.1293, simple_loss=0.1587, pruned_loss=0.04997, over 1986806.25 frames. ], batch size: 100, lr: 7.17e-03, grad_scale: 8.0 2022-12-08 00:23:48,650 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7395, 1.4273, 3.7166, 1.6418, 3.7460, 3.8972, 2.8311, 4.1522], device='cuda:3'), covar=tensor([0.0218, 0.3093, 0.0358, 0.2186, 0.0449, 0.0351, 0.0695, 0.0146], device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0158, 0.0160, 0.0168, 0.0172, 0.0174, 0.0136, 0.0144], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-08 00:24:06,587 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80241.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:24:15,563 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.3162, 1.7148, 1.8172, 1.7779, 1.6784, 1.7408, 1.4615, 1.2551], device='cuda:3'), covar=tensor([0.1341, 0.1003, 0.0478, 0.0601, 0.1384, 0.0710, 0.1515, 0.2025], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0076, 0.0060, 0.0063, 0.0091, 0.0074, 0.0094, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0005, 0.0005, 0.0004, 0.0006, 0.0005, 0.0006, 0.0006], device='cuda:3') 2022-12-08 00:24:21,385 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.285e+02 2.172e+02 2.626e+02 3.163e+02 7.453e+02, threshold=5.252e+02, percent-clipped=2.0 2022-12-08 00:24:41,756 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6354, 3.3472, 3.1941, 3.2924, 3.5797, 3.5454, 3.5960, 3.6052], device='cuda:3'), covar=tensor([0.0865, 0.0742, 0.1877, 0.2820, 0.0764, 0.0851, 0.1104, 0.0869], device='cuda:3'), in_proj_covar=tensor([0.0362, 0.0247, 0.0420, 0.0531, 0.0322, 0.0404, 0.0386, 0.0350], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 00:24:42,910 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80283.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:24:48,267 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80289.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:24:56,523 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2022-12-08 00:25:09,105 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80313.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:25:16,061 INFO [train.py:873] (3/4) Epoch 11, batch 4700, loss[loss=0.1979, simple_loss=0.1972, pruned_loss=0.09927, over 8599.00 frames. ], tot_loss[loss=0.1299, simple_loss=0.1588, pruned_loss=0.05051, over 1933170.49 frames. ], batch size: 100, lr: 7.16e-03, grad_scale: 8.0 2022-12-08 00:25:36,799 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80344.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:25:49,170 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.156e+02 2.137e+02 2.805e+02 3.644e+02 8.749e+02, threshold=5.610e+02, percent-clipped=7.0 2022-12-08 00:25:51,863 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80361.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:26:20,282 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80393.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:26:42,686 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80418.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:26:45,212 INFO [train.py:873] (3/4) Epoch 11, batch 4800, loss[loss=0.1007, simple_loss=0.1432, pruned_loss=0.02913, over 14643.00 frames. ], tot_loss[loss=0.1285, simple_loss=0.1573, pruned_loss=0.0498, over 1917236.04 frames. ], batch size: 23, lr: 7.16e-03, grad_scale: 8.0 2022-12-08 00:26:58,357 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7759, 1.3740, 2.5509, 2.3372, 2.4433, 2.5162, 1.8926, 2.5291], device='cuda:3'), covar=tensor([0.0855, 0.1150, 0.0160, 0.0321, 0.0349, 0.0172, 0.0548, 0.0210], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0158, 0.0125, 0.0166, 0.0141, 0.0135, 0.0117, 0.0117], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 00:27:14,137 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80454.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:27:17,205 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.361e+02 2.171e+02 2.910e+02 3.468e+02 6.443e+02, threshold=5.821e+02, percent-clipped=1.0 2022-12-08 00:27:21,575 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80463.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:27:23,924 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80466.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:27:25,815 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7842, 3.4317, 2.6921, 3.9818, 3.8744, 3.7507, 3.3037, 2.7302], device='cuda:3'), covar=tensor([0.0900, 0.1421, 0.3981, 0.0567, 0.0883, 0.1366, 0.1294, 0.3884], device='cuda:3'), in_proj_covar=tensor([0.0275, 0.0302, 0.0275, 0.0255, 0.0310, 0.0300, 0.0256, 0.0257], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2022-12-08 00:27:37,286 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.73 vs. limit=5.0 2022-12-08 00:27:38,491 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1324, 2.0233, 1.8230, 1.8518, 2.0907, 2.1141, 2.0809, 2.0660], device='cuda:3'), covar=tensor([0.1357, 0.0979, 0.3202, 0.3247, 0.1171, 0.1120, 0.1935, 0.1227], device='cuda:3'), in_proj_covar=tensor([0.0369, 0.0253, 0.0429, 0.0542, 0.0330, 0.0414, 0.0393, 0.0356], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 00:27:49,387 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80494.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:28:12,375 INFO [train.py:873] (3/4) Epoch 11, batch 4900, loss[loss=0.1267, simple_loss=0.1559, pruned_loss=0.04873, over 13574.00 frames. ], tot_loss[loss=0.1296, simple_loss=0.1585, pruned_loss=0.05039, over 1924840.77 frames. ], batch size: 100, lr: 7.15e-03, grad_scale: 8.0 2022-12-08 00:28:12,558 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.3145, 1.3719, 1.4934, 1.0905, 1.1517, 0.9376, 0.9930, 0.9781], device='cuda:3'), covar=tensor([0.0168, 0.0219, 0.0172, 0.0242, 0.0208, 0.0360, 0.0223, 0.0356], device='cuda:3'), in_proj_covar=tensor([0.0015, 0.0016, 0.0014, 0.0015, 0.0015, 0.0025, 0.0020, 0.0025], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2022-12-08 00:28:15,361 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80524.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:28:24,138 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4501, 3.1873, 3.1643, 3.4626, 3.2762, 3.4719, 3.5090, 2.9237], device='cuda:3'), covar=tensor([0.0529, 0.1035, 0.0507, 0.0535, 0.0761, 0.0382, 0.0610, 0.0639], device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0254, 0.0178, 0.0176, 0.0171, 0.0141, 0.0259, 0.0157], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 00:28:42,404 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80555.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:28:44,856 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.308e+02 2.165e+02 2.757e+02 3.267e+02 6.600e+02, threshold=5.514e+02, percent-clipped=1.0 2022-12-08 00:29:37,812 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.33 vs. limit=2.0 2022-12-08 00:29:41,616 INFO [train.py:873] (3/4) Epoch 11, batch 5000, loss[loss=0.1278, simple_loss=0.1527, pruned_loss=0.0514, over 14156.00 frames. ], tot_loss[loss=0.1291, simple_loss=0.1578, pruned_loss=0.05016, over 1891110.94 frames. ], batch size: 99, lr: 7.15e-03, grad_scale: 8.0 2022-12-08 00:29:57,610 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80639.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:30:14,651 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.349e+02 2.220e+02 2.746e+02 3.480e+02 5.243e+02, threshold=5.491e+02, percent-clipped=0.0 2022-12-08 00:30:40,853 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.2948, 4.9458, 4.8588, 5.2107, 4.8907, 4.4659, 5.3258, 5.1214], device='cuda:3'), covar=tensor([0.0528, 0.0638, 0.0738, 0.0604, 0.0664, 0.0543, 0.0571, 0.0631], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0127, 0.0133, 0.0143, 0.0135, 0.0112, 0.0155, 0.0133], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 00:31:10,694 INFO [train.py:873] (3/4) Epoch 11, batch 5100, loss[loss=0.1167, simple_loss=0.1516, pruned_loss=0.04085, over 13964.00 frames. ], tot_loss[loss=0.1272, simple_loss=0.1568, pruned_loss=0.04877, over 1941897.12 frames. ], batch size: 26, lr: 7.15e-03, grad_scale: 8.0 2022-12-08 00:31:12,795 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.57 vs. limit=5.0 2022-12-08 00:31:24,298 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80736.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:31:35,048 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80749.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:31:43,012 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.337e+02 2.114e+02 2.657e+02 3.455e+02 5.035e+02, threshold=5.315e+02, percent-clipped=0.0 2022-12-08 00:31:55,185 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6608, 1.9425, 1.8969, 1.7373, 1.8748, 1.6552, 1.5440, 1.4122], device='cuda:3'), covar=tensor([0.0611, 0.0409, 0.0348, 0.0415, 0.0347, 0.0299, 0.0277, 0.0440], device='cuda:3'), in_proj_covar=tensor([0.0015, 0.0016, 0.0014, 0.0015, 0.0015, 0.0025, 0.0020, 0.0025], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2022-12-08 00:31:56,780 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80773.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:32:17,364 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80797.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:32:36,969 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80819.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:32:38,684 INFO [train.py:873] (3/4) Epoch 11, batch 5200, loss[loss=0.1603, simple_loss=0.1565, pruned_loss=0.08204, over 2614.00 frames. ], tot_loss[loss=0.1287, simple_loss=0.1578, pruned_loss=0.04982, over 1911806.62 frames. ], batch size: 100, lr: 7.14e-03, grad_scale: 8.0 2022-12-08 00:32:44,917 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9797, 2.6883, 3.8208, 2.9538, 3.9345, 3.8178, 3.6472, 3.2255], device='cuda:3'), covar=tensor([0.0918, 0.2798, 0.0997, 0.1718, 0.0715, 0.0861, 0.1556, 0.1671], device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0316, 0.0402, 0.0307, 0.0378, 0.0321, 0.0363, 0.0316], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 00:32:46,016 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.70 vs. limit=5.0 2022-12-08 00:32:50,356 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80834.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:33:01,733 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1925, 2.9847, 2.7280, 2.8973, 3.1139, 3.0998, 3.1837, 3.1586], device='cuda:3'), covar=tensor([0.1116, 0.0949, 0.2771, 0.3117, 0.1044, 0.1111, 0.1649, 0.0984], device='cuda:3'), in_proj_covar=tensor([0.0364, 0.0252, 0.0425, 0.0537, 0.0323, 0.0410, 0.0384, 0.0355], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 00:33:04,596 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80850.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:33:11,236 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.384e+02 2.222e+02 2.775e+02 3.597e+02 5.956e+02, threshold=5.550e+02, percent-clipped=3.0 2022-12-08 00:34:04,554 INFO [train.py:873] (3/4) Epoch 11, batch 5300, loss[loss=0.1583, simple_loss=0.1532, pruned_loss=0.08175, over 2579.00 frames. ], tot_loss[loss=0.1288, simple_loss=0.1578, pruned_loss=0.04987, over 1931566.83 frames. ], batch size: 100, lr: 7.14e-03, grad_scale: 8.0 2022-12-08 00:34:20,732 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80939.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:34:37,142 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.222e+02 2.266e+02 2.683e+02 3.445e+02 6.231e+02, threshold=5.367e+02, percent-clipped=3.0 2022-12-08 00:35:00,017 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2762, 1.9606, 2.2377, 1.5815, 2.1212, 0.9197, 1.5478, 2.1281], device='cuda:3'), covar=tensor([0.1255, 0.0966, 0.0628, 0.2864, 0.1706, 0.1139, 0.2080, 0.1170], device='cuda:3'), in_proj_covar=tensor([0.0025, 0.0026, 0.0028, 0.0025, 0.0027, 0.0039, 0.0027, 0.0029], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:3') 2022-12-08 00:35:02,417 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80987.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:35:24,607 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2022-12-08 00:35:28,563 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4540, 1.1042, 1.2051, 0.8430, 1.0741, 1.4677, 1.1607, 1.1795], device='cuda:3'), covar=tensor([0.0308, 0.0664, 0.0550, 0.0432, 0.0781, 0.0493, 0.0277, 0.0886], device='cuda:3'), in_proj_covar=tensor([0.0138, 0.0184, 0.0137, 0.0126, 0.0135, 0.0144, 0.0120, 0.0140], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0005, 0.0006], device='cuda:3') 2022-12-08 00:35:32,524 INFO [train.py:873] (3/4) Epoch 11, batch 5400, loss[loss=0.1271, simple_loss=0.1289, pruned_loss=0.06263, over 2666.00 frames. ], tot_loss[loss=0.1295, simple_loss=0.1585, pruned_loss=0.05024, over 1938621.18 frames. ], batch size: 100, lr: 7.13e-03, grad_scale: 16.0 2022-12-08 00:35:58,317 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81049.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:36:05,567 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.933e+01 1.955e+02 2.740e+02 3.056e+02 5.675e+02, threshold=5.479e+02, percent-clipped=1.0 2022-12-08 00:36:35,751 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81092.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:36:39,870 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81097.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:36:56,407 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81116.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:36:58,936 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81119.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:37:00,420 INFO [train.py:873] (3/4) Epoch 11, batch 5500, loss[loss=0.1337, simple_loss=0.1539, pruned_loss=0.05673, over 6975.00 frames. ], tot_loss[loss=0.1281, simple_loss=0.1575, pruned_loss=0.04934, over 1949306.78 frames. ], batch size: 100, lr: 7.13e-03, grad_scale: 8.0 2022-12-08 00:37:07,732 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81129.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:37:07,870 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8595, 2.6652, 4.9414, 3.2957, 4.6380, 2.4497, 3.8512, 4.6056], device='cuda:3'), covar=tensor([0.0419, 0.3806, 0.0379, 0.6836, 0.0514, 0.3212, 0.1087, 0.0240], device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0214, 0.0198, 0.0289, 0.0217, 0.0218, 0.0214, 0.0202], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 00:37:25,841 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81150.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:37:33,797 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.607e+02 2.115e+02 2.510e+02 3.295e+02 5.828e+02, threshold=5.019e+02, percent-clipped=2.0 2022-12-08 00:37:40,232 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1279, 1.5555, 1.7539, 1.6866, 1.5294, 1.6168, 1.3662, 1.2570], device='cuda:3'), covar=tensor([0.1655, 0.0845, 0.0475, 0.0416, 0.1258, 0.0761, 0.1748, 0.1550], device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0076, 0.0061, 0.0065, 0.0093, 0.0075, 0.0094, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0006, 0.0006], device='cuda:3') 2022-12-08 00:37:40,965 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81167.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:37:49,588 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81177.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:38:07,658 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81198.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:38:28,044 INFO [train.py:873] (3/4) Epoch 11, batch 5600, loss[loss=0.1087, simple_loss=0.1513, pruned_loss=0.03298, over 14214.00 frames. ], tot_loss[loss=0.1284, simple_loss=0.1582, pruned_loss=0.0493, over 2006105.28 frames. ], batch size: 32, lr: 7.12e-03, grad_scale: 8.0 2022-12-08 00:39:02,406 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.421e+02 2.253e+02 2.769e+02 3.640e+02 8.621e+02, threshold=5.537e+02, percent-clipped=7.0 2022-12-08 00:39:04,635 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.93 vs. limit=5.0 2022-12-08 00:39:12,847 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2856, 3.0999, 2.8267, 2.9702, 3.2705, 3.1939, 3.2547, 3.2541], device='cuda:3'), covar=tensor([0.1119, 0.0837, 0.2529, 0.3402, 0.0808, 0.1136, 0.1292, 0.1092], device='cuda:3'), in_proj_covar=tensor([0.0367, 0.0257, 0.0429, 0.0536, 0.0320, 0.0415, 0.0384, 0.0360], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 00:39:37,264 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81299.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:39:38,966 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81301.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:39:55,989 INFO [train.py:873] (3/4) Epoch 11, batch 5700, loss[loss=0.1271, simple_loss=0.1647, pruned_loss=0.04481, over 14506.00 frames. ], tot_loss[loss=0.1282, simple_loss=0.1581, pruned_loss=0.04911, over 2020131.45 frames. ], batch size: 49, lr: 7.12e-03, grad_scale: 8.0 2022-12-08 00:40:09,457 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5678, 2.3969, 2.5585, 1.6562, 2.1413, 2.4944, 2.6302, 2.2454], device='cuda:3'), covar=tensor([0.0750, 0.0920, 0.0946, 0.1689, 0.1088, 0.0660, 0.0644, 0.1356], device='cuda:3'), in_proj_covar=tensor([0.0137, 0.0182, 0.0136, 0.0126, 0.0134, 0.0143, 0.0118, 0.0139], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0005, 0.0006], device='cuda:3') 2022-12-08 00:40:22,453 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.72 vs. limit=5.0 2022-12-08 00:40:28,773 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.149e+01 2.189e+02 2.731e+02 3.467e+02 4.923e+02, threshold=5.462e+02, percent-clipped=0.0 2022-12-08 00:40:29,847 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81360.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:40:31,284 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81362.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:40:55,131 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8672, 1.9198, 1.6007, 1.9949, 1.9081, 1.8964, 1.7988, 1.7847], device='cuda:3'), covar=tensor([0.0980, 0.0777, 0.1846, 0.0428, 0.0905, 0.0528, 0.1326, 0.0756], device='cuda:3'), in_proj_covar=tensor([0.0269, 0.0296, 0.0270, 0.0252, 0.0308, 0.0294, 0.0255, 0.0255], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2022-12-08 00:40:57,889 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81392.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:41:23,789 INFO [train.py:873] (3/4) Epoch 11, batch 5800, loss[loss=0.09741, simple_loss=0.145, pruned_loss=0.02489, over 14098.00 frames. ], tot_loss[loss=0.1278, simple_loss=0.1574, pruned_loss=0.04913, over 1962300.75 frames. ], batch size: 29, lr: 7.11e-03, grad_scale: 8.0 2022-12-08 00:41:30,577 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81429.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:41:31,021 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.75 vs. limit=2.0 2022-12-08 00:41:40,067 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81440.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:41:56,860 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.164e+02 2.147e+02 2.711e+02 3.541e+02 7.030e+02, threshold=5.421e+02, percent-clipped=5.0 2022-12-08 00:42:08,743 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81472.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:42:12,695 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81477.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:42:51,272 INFO [train.py:873] (3/4) Epoch 11, batch 5900, loss[loss=0.1626, simple_loss=0.145, pruned_loss=0.09008, over 1245.00 frames. ], tot_loss[loss=0.1281, simple_loss=0.1575, pruned_loss=0.04932, over 1929540.00 frames. ], batch size: 100, lr: 7.11e-03, grad_scale: 8.0 2022-12-08 00:43:14,126 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7590, 2.7209, 2.6767, 2.9076, 2.4551, 2.5762, 2.8433, 2.7817], device='cuda:3'), covar=tensor([0.0778, 0.1023, 0.0917, 0.0648, 0.1304, 0.0759, 0.0818, 0.0791], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0128, 0.0135, 0.0145, 0.0136, 0.0112, 0.0156, 0.0133], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 00:43:24,407 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.261e+02 2.246e+02 2.674e+02 3.531e+02 7.780e+02, threshold=5.349e+02, percent-clipped=2.0 2022-12-08 00:44:19,735 INFO [train.py:873] (3/4) Epoch 11, batch 6000, loss[loss=0.1735, simple_loss=0.1851, pruned_loss=0.08098, over 9499.00 frames. ], tot_loss[loss=0.1281, simple_loss=0.1578, pruned_loss=0.04919, over 2025381.88 frames. ], batch size: 100, lr: 7.11e-03, grad_scale: 8.0 2022-12-08 00:44:19,736 INFO [train.py:896] (3/4) Computing validation loss 2022-12-08 00:44:28,212 INFO [train.py:905] (3/4) Epoch 11, validation: loss=0.1286, simple_loss=0.169, pruned_loss=0.04409, over 857387.00 frames. 2022-12-08 00:44:28,213 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-08 00:44:57,702 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2022-12-08 00:44:58,266 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81655.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:45:00,107 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81657.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:45:01,818 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.338e+02 2.152e+02 2.716e+02 3.284e+02 6.077e+02, threshold=5.431e+02, percent-clipped=1.0 2022-12-08 00:45:11,811 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81670.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:45:55,907 INFO [train.py:873] (3/4) Epoch 11, batch 6100, loss[loss=0.104, simple_loss=0.1451, pruned_loss=0.03143, over 14261.00 frames. ], tot_loss[loss=0.1286, simple_loss=0.1582, pruned_loss=0.04952, over 2003500.09 frames. ], batch size: 35, lr: 7.10e-03, grad_scale: 8.0 2022-12-08 00:45:56,418 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.76 vs. limit=2.0 2022-12-08 00:46:04,804 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81731.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 00:46:29,253 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.111e+02 2.193e+02 2.528e+02 3.178e+02 6.202e+02, threshold=5.055e+02, percent-clipped=2.0 2022-12-08 00:46:29,519 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3814, 3.6956, 4.3551, 2.8852, 2.6322, 3.6324, 1.7349, 3.7224], device='cuda:3'), covar=tensor([0.1038, 0.0817, 0.0472, 0.1830, 0.2063, 0.0955, 0.4059, 0.0890], device='cuda:3'), in_proj_covar=tensor([0.0081, 0.0096, 0.0091, 0.0096, 0.0114, 0.0083, 0.0125, 0.0087], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 00:46:40,612 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81772.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:47:23,438 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81820.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:47:24,229 INFO [train.py:873] (3/4) Epoch 11, batch 6200, loss[loss=0.182, simple_loss=0.1573, pruned_loss=0.1033, over 1289.00 frames. ], tot_loss[loss=0.1281, simple_loss=0.1581, pruned_loss=0.04901, over 2024038.13 frames. ], batch size: 100, lr: 7.10e-03, grad_scale: 8.0 2022-12-08 00:47:41,688 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1328, 1.5065, 1.7236, 1.7110, 1.5493, 1.6949, 1.3387, 1.2435], device='cuda:3'), covar=tensor([0.1869, 0.1127, 0.0434, 0.0462, 0.1351, 0.0649, 0.1811, 0.1428], device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0077, 0.0061, 0.0065, 0.0092, 0.0075, 0.0094, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0006, 0.0006], device='cuda:3') 2022-12-08 00:47:58,181 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.426e+02 2.324e+02 2.708e+02 3.471e+02 1.552e+03, threshold=5.417e+02, percent-clipped=7.0 2022-12-08 00:48:15,586 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.35 vs. limit=5.0 2022-12-08 00:48:52,507 INFO [train.py:873] (3/4) Epoch 11, batch 6300, loss[loss=0.1484, simple_loss=0.1725, pruned_loss=0.06213, over 14235.00 frames. ], tot_loss[loss=0.1274, simple_loss=0.1574, pruned_loss=0.04868, over 1977041.27 frames. ], batch size: 94, lr: 7.09e-03, grad_scale: 8.0 2022-12-08 00:49:02,090 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9240, 4.9816, 5.3511, 4.3962, 5.1475, 5.4326, 2.2413, 4.7756], device='cuda:3'), covar=tensor([0.0239, 0.0224, 0.0271, 0.0382, 0.0218, 0.0091, 0.2592, 0.0253], device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0165, 0.0136, 0.0135, 0.0197, 0.0131, 0.0154, 0.0181], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 00:49:22,574 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81955.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:49:24,285 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81957.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:49:25,809 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.231e+02 2.360e+02 2.936e+02 3.429e+02 7.128e+02, threshold=5.872e+02, percent-clipped=5.0 2022-12-08 00:49:50,576 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81987.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:50:04,806 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82003.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:50:06,454 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82005.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:50:20,251 INFO [train.py:873] (3/4) Epoch 11, batch 6400, loss[loss=0.1336, simple_loss=0.1344, pruned_loss=0.06638, over 2655.00 frames. ], tot_loss[loss=0.1275, simple_loss=0.1575, pruned_loss=0.0488, over 1986973.03 frames. ], batch size: 100, lr: 7.09e-03, grad_scale: 8.0 2022-12-08 00:50:24,741 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82026.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 00:50:43,908 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82048.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:50:50,009 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82055.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:50:53,436 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.418e+02 2.188e+02 2.645e+02 3.155e+02 5.407e+02, threshold=5.290e+02, percent-clipped=0.0 2022-12-08 00:51:24,979 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4971, 2.5144, 4.4789, 4.5272, 4.5231, 2.6396, 4.5604, 3.6182], device='cuda:3'), covar=tensor([0.0295, 0.0901, 0.0735, 0.0308, 0.0270, 0.1432, 0.0329, 0.0757], device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0251, 0.0369, 0.0316, 0.0259, 0.0299, 0.0295, 0.0280], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 00:51:43,683 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82116.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:51:44,090 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2022-12-08 00:51:47,738 INFO [train.py:873] (3/4) Epoch 11, batch 6500, loss[loss=0.1294, simple_loss=0.1581, pruned_loss=0.05031, over 14576.00 frames. ], tot_loss[loss=0.1279, simple_loss=0.1577, pruned_loss=0.04904, over 1998608.05 frames. ], batch size: 49, lr: 7.08e-03, grad_scale: 8.0 2022-12-08 00:52:14,995 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.46 vs. limit=5.0 2022-12-08 00:52:19,817 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.9208, 5.7656, 5.3217, 5.9362, 5.5017, 5.1414, 5.9800, 5.8956], device='cuda:3'), covar=tensor([0.0465, 0.0395, 0.0586, 0.0327, 0.0518, 0.0351, 0.0360, 0.0383], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0128, 0.0134, 0.0144, 0.0135, 0.0111, 0.0157, 0.0133], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 00:52:20,599 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.664e+01 2.301e+02 3.022e+02 3.856e+02 8.544e+02, threshold=6.044e+02, percent-clipped=9.0 2022-12-08 00:52:32,022 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3277, 2.2029, 3.1199, 2.5690, 3.0715, 3.0666, 2.9367, 2.6504], device='cuda:3'), covar=tensor([0.0875, 0.2842, 0.0981, 0.1711, 0.0766, 0.1014, 0.1136, 0.1783], device='cuda:3'), in_proj_covar=tensor([0.0354, 0.0317, 0.0404, 0.0311, 0.0382, 0.0325, 0.0370, 0.0314], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 00:53:14,953 INFO [train.py:873] (3/4) Epoch 11, batch 6600, loss[loss=0.1112, simple_loss=0.1491, pruned_loss=0.03663, over 14186.00 frames. ], tot_loss[loss=0.1291, simple_loss=0.1581, pruned_loss=0.05008, over 1949207.45 frames. ], batch size: 84, lr: 7.08e-03, grad_scale: 8.0 2022-12-08 00:53:41,423 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6385, 1.4408, 3.5831, 1.6125, 3.4885, 3.6719, 2.6723, 3.9446], device='cuda:3'), covar=tensor([0.0217, 0.3154, 0.0430, 0.2242, 0.0702, 0.0384, 0.0906, 0.0178], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0158, 0.0162, 0.0169, 0.0170, 0.0173, 0.0133, 0.0143], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-08 00:53:48,294 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.365e+02 2.109e+02 2.550e+02 3.112e+02 5.386e+02, threshold=5.100e+02, percent-clipped=0.0 2022-12-08 00:54:11,674 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.0502, 4.8582, 4.5195, 4.6836, 4.7278, 4.9740, 5.0617, 5.0426], device='cuda:3'), covar=tensor([0.0764, 0.0390, 0.1907, 0.2413, 0.0664, 0.0694, 0.0732, 0.0846], device='cuda:3'), in_proj_covar=tensor([0.0373, 0.0259, 0.0431, 0.0546, 0.0325, 0.0418, 0.0390, 0.0367], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 00:54:17,811 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.6636, 2.6415, 2.7774, 2.7274, 2.7103, 2.4610, 1.5370, 2.4883], device='cuda:3'), covar=tensor([0.0452, 0.0453, 0.0407, 0.0357, 0.0390, 0.0962, 0.2414, 0.0382], device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0166, 0.0137, 0.0135, 0.0196, 0.0131, 0.0155, 0.0182], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 00:54:26,626 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9698, 1.3996, 3.8827, 1.8294, 3.7582, 4.0265, 2.9894, 4.2547], device='cuda:3'), covar=tensor([0.0233, 0.3441, 0.0506, 0.2343, 0.0556, 0.0398, 0.0728, 0.0205], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0157, 0.0161, 0.0169, 0.0169, 0.0172, 0.0132, 0.0143], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-08 00:54:43,679 INFO [train.py:873] (3/4) Epoch 11, batch 6700, loss[loss=0.1171, simple_loss=0.1521, pruned_loss=0.041, over 14275.00 frames. ], tot_loss[loss=0.1305, simple_loss=0.159, pruned_loss=0.05103, over 1968261.27 frames. ], batch size: 44, lr: 7.08e-03, grad_scale: 8.0 2022-12-08 00:54:48,101 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82326.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:55:02,419 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82343.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:55:03,952 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2022-12-08 00:55:05,109 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82346.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:55:08,755 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8405, 1.3937, 3.1290, 2.8680, 3.0018, 3.0980, 2.3340, 3.0888], device='cuda:3'), covar=tensor([0.1167, 0.1339, 0.0133, 0.0310, 0.0243, 0.0136, 0.0441, 0.0162], device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0157, 0.0125, 0.0164, 0.0141, 0.0135, 0.0117, 0.0118], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 00:55:16,416 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.150e+02 2.173e+02 2.670e+02 3.578e+02 7.367e+02, threshold=5.339e+02, percent-clipped=2.0 2022-12-08 00:55:29,455 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82374.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:55:59,194 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82407.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:56:02,568 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82411.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:56:11,099 INFO [train.py:873] (3/4) Epoch 11, batch 6800, loss[loss=0.11, simple_loss=0.1494, pruned_loss=0.03533, over 14285.00 frames. ], tot_loss[loss=0.1296, simple_loss=0.1584, pruned_loss=0.05038, over 1982552.85 frames. ], batch size: 63, lr: 7.07e-03, grad_scale: 8.0 2022-12-08 00:56:43,724 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.232e+02 2.471e+02 3.089e+02 4.263e+02 8.393e+02, threshold=6.178e+02, percent-clipped=10.0 2022-12-08 00:57:14,058 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82493.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 00:57:37,892 INFO [train.py:873] (3/4) Epoch 11, batch 6900, loss[loss=0.1439, simple_loss=0.1405, pruned_loss=0.07364, over 1160.00 frames. ], tot_loss[loss=0.1299, simple_loss=0.1588, pruned_loss=0.05047, over 1977410.33 frames. ], batch size: 100, lr: 7.07e-03, grad_scale: 8.0 2022-12-08 00:57:43,965 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7643, 1.5002, 1.7858, 2.0412, 1.3745, 1.7860, 1.7282, 1.8540], device='cuda:3'), covar=tensor([0.0126, 0.0249, 0.0122, 0.0107, 0.0247, 0.0272, 0.0183, 0.0119], device='cuda:3'), in_proj_covar=tensor([0.0283, 0.0248, 0.0365, 0.0313, 0.0257, 0.0297, 0.0292, 0.0276], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 00:57:53,048 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9194, 2.5684, 3.3704, 2.2358, 2.0399, 2.9113, 1.4366, 2.8403], device='cuda:3'), covar=tensor([0.0826, 0.0967, 0.0640, 0.1914, 0.2263, 0.0846, 0.3761, 0.1036], device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0093, 0.0087, 0.0093, 0.0110, 0.0081, 0.0119, 0.0085], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0005, 0.0004], device='cuda:3') 2022-12-08 00:58:06,347 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82554.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 00:58:11,136 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.611e+02 2.298e+02 3.049e+02 3.803e+02 8.603e+02, threshold=6.098e+02, percent-clipped=2.0 2022-12-08 00:58:20,048 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82569.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:59:06,537 INFO [train.py:873] (3/4) Epoch 11, batch 7000, loss[loss=0.1409, simple_loss=0.1693, pruned_loss=0.05628, over 9497.00 frames. ], tot_loss[loss=0.1281, simple_loss=0.1577, pruned_loss=0.04923, over 1986336.21 frames. ], batch size: 100, lr: 7.06e-03, grad_scale: 8.0 2022-12-08 00:59:14,671 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82630.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:59:26,132 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82643.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 00:59:39,575 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.371e+02 2.154e+02 2.641e+02 3.288e+02 6.152e+02, threshold=5.282e+02, percent-clipped=1.0 2022-12-08 01:00:08,002 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82691.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:00:17,817 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82702.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:00:25,971 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82711.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:00:34,803 INFO [train.py:873] (3/4) Epoch 11, batch 7100, loss[loss=0.1521, simple_loss=0.1677, pruned_loss=0.06822, over 7774.00 frames. ], tot_loss[loss=0.1281, simple_loss=0.1582, pruned_loss=0.04895, over 2022988.89 frames. ], batch size: 100, lr: 7.06e-03, grad_scale: 8.0 2022-12-08 01:01:07,645 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.406e+02 2.263e+02 2.729e+02 3.602e+02 7.969e+02, threshold=5.458e+02, percent-clipped=3.0 2022-12-08 01:01:07,749 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82759.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:02:03,357 INFO [train.py:873] (3/4) Epoch 11, batch 7200, loss[loss=0.1341, simple_loss=0.16, pruned_loss=0.05416, over 14163.00 frames. ], tot_loss[loss=0.1283, simple_loss=0.1583, pruned_loss=0.04913, over 2041583.99 frames. ], batch size: 99, lr: 7.05e-03, grad_scale: 8.0 2022-12-08 01:02:28,403 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82849.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 01:02:36,830 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.101e+02 2.276e+02 3.049e+02 3.669e+02 5.112e+02, threshold=6.097e+02, percent-clipped=0.0 2022-12-08 01:02:51,618 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.0215, 1.1376, 0.9817, 1.0499, 1.0717, 0.6148, 0.9561, 1.1020], device='cuda:3'), covar=tensor([0.0540, 0.0718, 0.0498, 0.0518, 0.0350, 0.0527, 0.0982, 0.0636], device='cuda:3'), in_proj_covar=tensor([0.0027, 0.0026, 0.0029, 0.0026, 0.0028, 0.0039, 0.0027, 0.0029], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 01:03:31,181 INFO [train.py:873] (3/4) Epoch 11, batch 7300, loss[loss=0.1412, simple_loss=0.1457, pruned_loss=0.06838, over 2658.00 frames. ], tot_loss[loss=0.1289, simple_loss=0.1584, pruned_loss=0.04963, over 2068996.24 frames. ], batch size: 100, lr: 7.05e-03, grad_scale: 8.0 2022-12-08 01:03:34,881 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82925.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:03:53,116 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.8323, 5.2631, 5.1639, 5.7972, 5.3885, 4.8735, 5.6934, 4.8376], device='cuda:3'), covar=tensor([0.0226, 0.0877, 0.0321, 0.0373, 0.0704, 0.0354, 0.0455, 0.0506], device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0256, 0.0180, 0.0179, 0.0174, 0.0144, 0.0267, 0.0159], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 01:04:04,502 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.016e+02 2.470e+02 3.084e+02 3.689e+02 1.024e+03, threshold=6.168e+02, percent-clipped=3.0 2022-12-08 01:04:08,523 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82963.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:04:43,203 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83002.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:04:49,695 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 2022-12-08 01:04:59,090 INFO [train.py:873] (3/4) Epoch 11, batch 7400, loss[loss=0.1857, simple_loss=0.1642, pruned_loss=0.1036, over 1295.00 frames. ], tot_loss[loss=0.1296, simple_loss=0.1586, pruned_loss=0.05031, over 2008602.72 frames. ], batch size: 100, lr: 7.05e-03, grad_scale: 8.0 2022-12-08 01:05:02,110 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83024.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:05:25,179 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=83050.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:05:32,577 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.163e+02 2.140e+02 2.552e+02 3.210e+02 1.054e+03, threshold=5.104e+02, percent-clipped=1.0 2022-12-08 01:06:26,610 INFO [train.py:873] (3/4) Epoch 11, batch 7500, loss[loss=0.1212, simple_loss=0.141, pruned_loss=0.0507, over 3921.00 frames. ], tot_loss[loss=0.1293, simple_loss=0.1586, pruned_loss=0.04998, over 2023219.94 frames. ], batch size: 100, lr: 7.04e-03, grad_scale: 16.0 2022-12-08 01:06:50,742 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83149.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 01:06:58,814 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.424e+02 2.439e+02 2.915e+02 3.570e+02 7.862e+02, threshold=5.829e+02, percent-clipped=6.0 2022-12-08 01:07:53,460 INFO [train.py:873] (3/4) Epoch 12, batch 0, loss[loss=0.1612, simple_loss=0.1876, pruned_loss=0.0674, over 14485.00 frames. ], tot_loss[loss=0.1612, simple_loss=0.1876, pruned_loss=0.0674, over 14485.00 frames. ], batch size: 49, lr: 6.74e-03, grad_scale: 16.0 2022-12-08 01:07:53,460 INFO [train.py:896] (3/4) Computing validation loss 2022-12-08 01:07:57,251 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9718, 1.4819, 3.5493, 3.3344, 3.3056, 3.6090, 3.1212, 3.5479], device='cuda:3'), covar=tensor([0.1181, 0.1382, 0.0134, 0.0248, 0.0266, 0.0123, 0.0206, 0.0144], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0161, 0.0130, 0.0168, 0.0145, 0.0139, 0.0121, 0.0121], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 01:07:59,229 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.6982, 4.2866, 4.3861, 4.6062, 4.4126, 4.3128, 4.6065, 4.3354], device='cuda:3'), covar=tensor([0.0187, 0.0496, 0.0292, 0.0310, 0.0601, 0.0192, 0.0333, 0.0285], device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0256, 0.0179, 0.0178, 0.0174, 0.0143, 0.0265, 0.0159], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 01:08:00,688 INFO [train.py:905] (3/4) Epoch 12, validation: loss=0.1326, simple_loss=0.1738, pruned_loss=0.04568, over 857387.00 frames. 2022-12-08 01:08:00,688 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-08 01:08:12,575 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83196.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:08:13,521 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=83197.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 01:08:38,638 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83225.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:08:54,783 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.64 vs. limit=2.0 2022-12-08 01:09:07,372 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83257.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:09:08,850 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 6.351e+01 1.849e+02 2.933e+02 3.942e+02 9.490e+02, threshold=5.866e+02, percent-clipped=6.0 2022-12-08 01:09:21,069 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=83273.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:09:30,011 INFO [train.py:873] (3/4) Epoch 12, batch 100, loss[loss=0.1622, simple_loss=0.1693, pruned_loss=0.0775, over 3847.00 frames. ], tot_loss[loss=0.1276, simple_loss=0.1586, pruned_loss=0.04834, over 854219.95 frames. ], batch size: 100, lr: 6.74e-03, grad_scale: 16.0 2022-12-08 01:09:30,896 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83284.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:09:34,674 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1319, 1.6865, 2.0685, 1.4196, 1.7814, 2.1105, 1.9617, 1.8264], device='cuda:3'), covar=tensor([0.1028, 0.1094, 0.1176, 0.1729, 0.1501, 0.0962, 0.0740, 0.1731], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0185, 0.0139, 0.0128, 0.0136, 0.0143, 0.0121, 0.0140], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0005, 0.0006], device='cuda:3') 2022-12-08 01:09:53,169 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7921, 1.5351, 2.9483, 2.6637, 2.8396, 2.9430, 2.1671, 2.9698], device='cuda:3'), covar=tensor([0.1241, 0.1234, 0.0145, 0.0324, 0.0314, 0.0147, 0.0390, 0.0178], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0159, 0.0129, 0.0167, 0.0144, 0.0138, 0.0120, 0.0120], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 01:10:00,912 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=83319.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:10:24,254 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83345.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:10:28,505 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1904, 1.8045, 2.1536, 1.9127, 2.2937, 2.0749, 2.0090, 2.0698], device='cuda:3'), covar=tensor([0.0412, 0.1547, 0.0400, 0.0822, 0.0364, 0.0793, 0.0360, 0.0524], device='cuda:3'), in_proj_covar=tensor([0.0347, 0.0312, 0.0393, 0.0304, 0.0373, 0.0320, 0.0361, 0.0309], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 01:10:36,278 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.323e+02 2.243e+02 2.760e+02 3.223e+02 5.001e+02, threshold=5.521e+02, percent-clipped=0.0 2022-12-08 01:10:57,433 INFO [train.py:873] (3/4) Epoch 12, batch 200, loss[loss=0.1655, simple_loss=0.1729, pruned_loss=0.0791, over 7811.00 frames. ], tot_loss[loss=0.1292, simple_loss=0.1586, pruned_loss=0.04986, over 1292817.61 frames. ], batch size: 100, lr: 6.74e-03, grad_scale: 16.0 2022-12-08 01:11:00,206 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5892, 2.8881, 2.7913, 2.8414, 2.2036, 2.8338, 2.5961, 1.4267], device='cuda:3'), covar=tensor([0.1652, 0.0781, 0.0806, 0.0624, 0.1046, 0.0560, 0.1225, 0.2529], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0076, 0.0062, 0.0064, 0.0092, 0.0075, 0.0094, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0006, 0.0006], device='cuda:3') 2022-12-08 01:11:06,827 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.2417, 4.5180, 5.0188, 5.2371, 4.8816, 4.4742, 5.1554, 4.0330], device='cuda:3'), covar=tensor([0.0952, 0.2177, 0.0773, 0.0946, 0.1314, 0.0662, 0.1035, 0.1204], device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0255, 0.0180, 0.0178, 0.0173, 0.0143, 0.0267, 0.0158], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 01:11:14,230 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.1269, 4.8803, 4.4887, 4.7459, 4.7465, 4.9987, 5.0906, 5.0788], device='cuda:3'), covar=tensor([0.0743, 0.0471, 0.2097, 0.2764, 0.0755, 0.0729, 0.1060, 0.0861], device='cuda:3'), in_proj_covar=tensor([0.0368, 0.0259, 0.0432, 0.0545, 0.0324, 0.0418, 0.0391, 0.0363], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 01:12:04,537 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.103e+02 2.214e+02 2.674e+02 3.340e+02 6.097e+02, threshold=5.348e+02, percent-clipped=3.0 2022-12-08 01:12:21,272 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.06 vs. limit=5.0 2022-12-08 01:12:24,714 INFO [train.py:873] (3/4) Epoch 12, batch 300, loss[loss=0.1149, simple_loss=0.1498, pruned_loss=0.04003, over 14550.00 frames. ], tot_loss[loss=0.1285, simple_loss=0.1576, pruned_loss=0.04965, over 1523762.82 frames. ], batch size: 49, lr: 6.73e-03, grad_scale: 8.0 2022-12-08 01:12:28,728 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4516, 3.7626, 3.4355, 3.3603, 2.5843, 3.5937, 3.4056, 1.9154], device='cuda:3'), covar=tensor([0.1764, 0.0792, 0.1211, 0.0947, 0.1031, 0.0443, 0.1250, 0.2304], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0077, 0.0062, 0.0064, 0.0092, 0.0076, 0.0094, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0006, 0.0006], device='cuda:3') 2022-12-08 01:12:59,562 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7507, 3.7438, 4.3420, 3.0511, 2.6743, 3.5310, 2.4404, 3.7217], device='cuda:3'), covar=tensor([0.0862, 0.0644, 0.0471, 0.1659, 0.2244, 0.0793, 0.3250, 0.1556], device='cuda:3'), in_proj_covar=tensor([0.0081, 0.0095, 0.0089, 0.0095, 0.0113, 0.0083, 0.0123, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 01:13:25,738 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=83552.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:13:31,462 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.5069, 4.9138, 5.0189, 5.4547, 5.0991, 4.5086, 5.3951, 4.4288], device='cuda:3'), covar=tensor([0.0343, 0.0985, 0.0332, 0.0389, 0.0686, 0.0380, 0.0511, 0.0532], device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0255, 0.0181, 0.0178, 0.0173, 0.0145, 0.0267, 0.0159], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 01:13:32,202 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.217e+02 2.254e+02 2.674e+02 3.399e+02 5.874e+02, threshold=5.349e+02, percent-clipped=4.0 2022-12-08 01:13:53,021 INFO [train.py:873] (3/4) Epoch 12, batch 400, loss[loss=0.1331, simple_loss=0.1609, pruned_loss=0.05268, over 14312.00 frames. ], tot_loss[loss=0.1264, simple_loss=0.1566, pruned_loss=0.04812, over 1762395.11 frames. ], batch size: 31, lr: 6.73e-03, grad_scale: 8.0 2022-12-08 01:14:18,490 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4859, 3.3076, 3.0003, 3.1839, 3.3992, 3.4036, 3.4873, 3.4353], device='cuda:3'), covar=tensor([0.0898, 0.0636, 0.2127, 0.2488, 0.0800, 0.0930, 0.1111, 0.0893], device='cuda:3'), in_proj_covar=tensor([0.0367, 0.0256, 0.0429, 0.0545, 0.0322, 0.0418, 0.0392, 0.0362], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 01:14:25,420 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83619.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:14:44,126 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=83640.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:14:47,755 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0499, 3.7859, 3.7836, 4.0702, 3.7624, 3.5064, 4.0608, 3.8949], device='cuda:3'), covar=tensor([0.0599, 0.0793, 0.0745, 0.0645, 0.0778, 0.0676, 0.0579, 0.0823], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0128, 0.0135, 0.0143, 0.0136, 0.0112, 0.0154, 0.0133], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 01:15:01,084 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.765e+01 2.109e+02 2.628e+02 3.451e+02 6.835e+02, threshold=5.257e+02, percent-clipped=4.0 2022-12-08 01:15:07,450 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=83667.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:15:21,385 INFO [train.py:873] (3/4) Epoch 12, batch 500, loss[loss=0.1288, simple_loss=0.123, pruned_loss=0.06725, over 1240.00 frames. ], tot_loss[loss=0.1273, simple_loss=0.1574, pruned_loss=0.04856, over 1842221.03 frames. ], batch size: 100, lr: 6.72e-03, grad_scale: 8.0 2022-12-08 01:15:25,821 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8722, 2.7399, 2.4709, 2.6229, 2.8274, 2.7898, 2.8671, 2.8717], device='cuda:3'), covar=tensor([0.1077, 0.0724, 0.2492, 0.2847, 0.0849, 0.1090, 0.1414, 0.0967], device='cuda:3'), in_proj_covar=tensor([0.0371, 0.0259, 0.0434, 0.0553, 0.0323, 0.0424, 0.0399, 0.0364], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 01:15:47,024 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.77 vs. limit=2.0 2022-12-08 01:16:01,915 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.72 vs. limit=2.0 2022-12-08 01:16:28,956 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.131e+02 2.083e+02 2.759e+02 3.343e+02 7.485e+02, threshold=5.518e+02, percent-clipped=2.0 2022-12-08 01:16:29,148 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83760.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:16:49,377 INFO [train.py:873] (3/4) Epoch 12, batch 600, loss[loss=0.1259, simple_loss=0.1535, pruned_loss=0.04913, over 11179.00 frames. ], tot_loss[loss=0.1272, simple_loss=0.1571, pruned_loss=0.04863, over 1918575.69 frames. ], batch size: 100, lr: 6.72e-03, grad_scale: 8.0 2022-12-08 01:17:18,962 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.51 vs. limit=5.0 2022-12-08 01:17:22,771 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83821.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:17:29,667 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.73 vs. limit=2.0 2022-12-08 01:17:34,707 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2022-12-08 01:17:50,244 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83852.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:17:50,292 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6942, 1.7298, 1.7644, 1.7122, 1.6473, 1.0067, 1.6221, 1.6346], device='cuda:3'), covar=tensor([0.1583, 0.0810, 0.0768, 0.1776, 0.1462, 0.0975, 0.0988, 0.1045], device='cuda:3'), in_proj_covar=tensor([0.0026, 0.0026, 0.0028, 0.0026, 0.0028, 0.0038, 0.0027, 0.0029], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 01:17:54,183 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.06 vs. limit=2.0 2022-12-08 01:17:57,120 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.264e+02 2.241e+02 2.685e+02 3.305e+02 1.014e+03, threshold=5.370e+02, percent-clipped=1.0 2022-12-08 01:18:17,202 INFO [train.py:873] (3/4) Epoch 12, batch 700, loss[loss=0.1211, simple_loss=0.1554, pruned_loss=0.04341, over 14279.00 frames. ], tot_loss[loss=0.1279, simple_loss=0.1574, pruned_loss=0.04921, over 1926327.61 frames. ], batch size: 66, lr: 6.72e-03, grad_scale: 8.0 2022-12-08 01:18:32,215 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=83900.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:18:54,102 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9799, 2.6992, 4.8266, 3.0872, 4.6214, 2.2051, 3.5700, 4.6208], device='cuda:3'), covar=tensor([0.0560, 0.4558, 0.0495, 0.8464, 0.0647, 0.4234, 0.1517, 0.0296], device='cuda:3'), in_proj_covar=tensor([0.0247, 0.0218, 0.0203, 0.0288, 0.0222, 0.0219, 0.0218, 0.0208], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 01:19:00,276 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7125, 2.9437, 3.0158, 2.9876, 2.2860, 3.0856, 2.8165, 1.3614], device='cuda:3'), covar=tensor([0.1743, 0.1122, 0.0676, 0.0656, 0.1006, 0.0400, 0.0974, 0.2639], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0078, 0.0063, 0.0064, 0.0093, 0.0076, 0.0095, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0006, 0.0006], device='cuda:3') 2022-12-08 01:19:02,033 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.3868, 1.2822, 1.3013, 1.4171, 1.4021, 0.8049, 1.2419, 1.2550], device='cuda:3'), covar=tensor([0.0523, 0.0668, 0.0802, 0.0593, 0.0503, 0.0811, 0.0733, 0.0698], device='cuda:3'), in_proj_covar=tensor([0.0026, 0.0026, 0.0028, 0.0026, 0.0028, 0.0038, 0.0027, 0.0029], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 01:19:06,826 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83940.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:19:14,789 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83949.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:19:23,842 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.357e+02 2.241e+02 2.717e+02 3.335e+02 9.146e+02, threshold=5.433e+02, percent-clipped=9.0 2022-12-08 01:19:32,927 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2022-12-08 01:19:44,264 INFO [train.py:873] (3/4) Epoch 12, batch 800, loss[loss=0.1107, simple_loss=0.1501, pruned_loss=0.03566, over 14281.00 frames. ], tot_loss[loss=0.1275, simple_loss=0.1572, pruned_loss=0.0489, over 1949569.19 frames. ], batch size: 31, lr: 6.71e-03, grad_scale: 8.0 2022-12-08 01:19:46,349 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83985.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:19:48,775 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=83988.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:20:05,874 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8387, 1.5713, 1.8182, 2.0156, 1.2279, 1.7973, 1.7247, 1.9380], device='cuda:3'), covar=tensor([0.0142, 0.0228, 0.0132, 0.0125, 0.0323, 0.0267, 0.0188, 0.0138], device='cuda:3'), in_proj_covar=tensor([0.0282, 0.0248, 0.0367, 0.0317, 0.0258, 0.0295, 0.0296, 0.0274], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 01:20:08,205 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84010.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 01:20:21,285 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9156, 1.9536, 1.5938, 1.9040, 1.8155, 1.8960, 1.8414, 1.6917], device='cuda:3'), covar=tensor([0.0944, 0.0791, 0.2264, 0.0709, 0.0841, 0.0649, 0.1243, 0.0783], device='cuda:3'), in_proj_covar=tensor([0.0268, 0.0295, 0.0266, 0.0256, 0.0309, 0.0295, 0.0253, 0.0251], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2022-12-08 01:20:28,663 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=7.81 vs. limit=5.0 2022-12-08 01:20:40,301 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84046.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:20:52,177 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.103e+02 2.165e+02 2.773e+02 3.425e+02 7.521e+02, threshold=5.546e+02, percent-clipped=4.0 2022-12-08 01:21:11,645 INFO [train.py:873] (3/4) Epoch 12, batch 900, loss[loss=0.1171, simple_loss=0.1525, pruned_loss=0.04087, over 14545.00 frames. ], tot_loss[loss=0.1278, simple_loss=0.1579, pruned_loss=0.0488, over 2035294.20 frames. ], batch size: 34, lr: 6.71e-03, grad_scale: 8.0 2022-12-08 01:21:24,385 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84097.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:21:30,855 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.3400, 1.6856, 1.8492, 1.7704, 1.6621, 1.8143, 1.4494, 1.2076], device='cuda:3'), covar=tensor([0.1600, 0.1233, 0.0578, 0.0496, 0.1247, 0.0699, 0.2146, 0.1970], device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0077, 0.0062, 0.0064, 0.0092, 0.0075, 0.0093, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0006, 0.0006], device='cuda:3') 2022-12-08 01:21:30,880 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.0118, 1.1675, 1.0772, 1.1057, 1.1120, 0.6483, 0.9259, 1.2192], device='cuda:3'), covar=tensor([0.0394, 0.0601, 0.0416, 0.0411, 0.0379, 0.0805, 0.0989, 0.0585], device='cuda:3'), in_proj_covar=tensor([0.0026, 0.0026, 0.0029, 0.0026, 0.0028, 0.0039, 0.0027, 0.0029], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 01:21:36,735 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.11 vs. limit=5.0 2022-12-08 01:21:40,366 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84116.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:22:07,800 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1414, 2.4877, 4.1193, 4.1802, 4.1112, 2.4935, 4.1224, 3.2527], device='cuda:3'), covar=tensor([0.0291, 0.0892, 0.0659, 0.0342, 0.0331, 0.1353, 0.0326, 0.0733], device='cuda:3'), in_proj_covar=tensor([0.0282, 0.0249, 0.0367, 0.0317, 0.0259, 0.0295, 0.0295, 0.0274], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 01:22:10,292 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.2737, 3.8101, 3.0880, 4.7140, 4.1954, 4.4448, 3.7927, 3.1826], device='cuda:3'), covar=tensor([0.0857, 0.1204, 0.3646, 0.0446, 0.1026, 0.1622, 0.1131, 0.3109], device='cuda:3'), in_proj_covar=tensor([0.0266, 0.0292, 0.0267, 0.0255, 0.0308, 0.0294, 0.0252, 0.0249], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2022-12-08 01:22:17,033 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84158.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:22:18,574 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.043e+02 2.201e+02 2.633e+02 3.303e+02 7.770e+02, threshold=5.265e+02, percent-clipped=5.0 2022-12-08 01:22:38,828 INFO [train.py:873] (3/4) Epoch 12, batch 1000, loss[loss=0.1324, simple_loss=0.1605, pruned_loss=0.05218, over 11171.00 frames. ], tot_loss[loss=0.1284, simple_loss=0.1577, pruned_loss=0.0495, over 1964817.92 frames. ], batch size: 100, lr: 6.70e-03, grad_scale: 8.0 2022-12-08 01:22:52,835 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84199.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:23:25,677 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84236.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:23:33,520 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5365, 4.1962, 4.1302, 4.6139, 4.2908, 4.0261, 4.6136, 3.8899], device='cuda:3'), covar=tensor([0.0434, 0.1098, 0.0404, 0.0449, 0.0782, 0.0857, 0.0528, 0.0536], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0258, 0.0182, 0.0181, 0.0175, 0.0145, 0.0269, 0.0160], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 01:23:46,259 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.153e+02 2.176e+02 2.741e+02 3.221e+02 6.313e+02, threshold=5.481e+02, percent-clipped=1.0 2022-12-08 01:23:46,509 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84260.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:24:06,156 INFO [train.py:873] (3/4) Epoch 12, batch 1100, loss[loss=0.1022, simple_loss=0.1377, pruned_loss=0.03334, over 14050.00 frames. ], tot_loss[loss=0.1294, simple_loss=0.1583, pruned_loss=0.05029, over 1972763.87 frames. ], batch size: 19, lr: 6.70e-03, grad_scale: 8.0 2022-12-08 01:24:18,022 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4542, 1.4581, 1.3875, 1.4613, 1.4269, 0.8630, 1.2618, 1.3206], device='cuda:3'), covar=tensor([0.0586, 0.0508, 0.0812, 0.0483, 0.0652, 0.0860, 0.0736, 0.0513], device='cuda:3'), in_proj_covar=tensor([0.0026, 0.0026, 0.0029, 0.0026, 0.0028, 0.0039, 0.0027, 0.0029], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 01:24:18,771 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84297.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:24:25,602 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84305.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 01:24:40,666 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2022-12-08 01:24:52,893 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84336.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:24:56,779 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84341.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:25:13,032 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.329e+02 2.219e+02 2.734e+02 3.210e+02 6.975e+02, threshold=5.467e+02, percent-clipped=2.0 2022-12-08 01:25:15,009 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5435, 2.1246, 3.4280, 3.5987, 3.5270, 2.2501, 3.4393, 2.6870], device='cuda:3'), covar=tensor([0.0353, 0.0901, 0.0721, 0.0423, 0.0381, 0.1341, 0.0380, 0.0827], device='cuda:3'), in_proj_covar=tensor([0.0286, 0.0251, 0.0372, 0.0322, 0.0261, 0.0297, 0.0299, 0.0278], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 01:25:33,228 INFO [train.py:873] (3/4) Epoch 12, batch 1200, loss[loss=0.1391, simple_loss=0.1403, pruned_loss=0.06894, over 2627.00 frames. ], tot_loss[loss=0.1278, simple_loss=0.1576, pruned_loss=0.04897, over 2031623.49 frames. ], batch size: 100, lr: 6.70e-03, grad_scale: 8.0 2022-12-08 01:25:45,552 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84397.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:25:45,609 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2692, 2.1482, 3.2118, 3.3578, 3.2773, 2.1308, 3.1711, 2.5845], device='cuda:3'), covar=tensor([0.0353, 0.0853, 0.0643, 0.0354, 0.0372, 0.1232, 0.0348, 0.0780], device='cuda:3'), in_proj_covar=tensor([0.0287, 0.0251, 0.0372, 0.0321, 0.0262, 0.0297, 0.0299, 0.0278], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 01:25:57,679 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2022-12-08 01:26:02,553 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84416.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:26:32,435 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1326, 2.1089, 4.5248, 4.1462, 4.1341, 4.5968, 4.2390, 4.5872], device='cuda:3'), covar=tensor([0.1227, 0.1231, 0.0086, 0.0144, 0.0176, 0.0084, 0.0130, 0.0092], device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0156, 0.0128, 0.0165, 0.0143, 0.0137, 0.0119, 0.0119], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 01:26:34,078 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84453.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:26:40,234 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.339e+02 2.298e+02 2.755e+02 3.792e+02 6.830e+02, threshold=5.510e+02, percent-clipped=5.0 2022-12-08 01:26:43,690 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84464.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:27:00,264 INFO [train.py:873] (3/4) Epoch 12, batch 1300, loss[loss=0.2418, simple_loss=0.2073, pruned_loss=0.1381, over 1222.00 frames. ], tot_loss[loss=0.1276, simple_loss=0.1575, pruned_loss=0.04886, over 2024966.65 frames. ], batch size: 100, lr: 6.69e-03, grad_scale: 8.0 2022-12-08 01:27:36,977 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84525.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:27:57,308 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9887, 2.1025, 2.3013, 2.2690, 1.9732, 2.2660, 1.9303, 1.4198], device='cuda:3'), covar=tensor([0.1137, 0.0937, 0.0695, 0.0625, 0.1180, 0.0627, 0.1507, 0.2360], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0078, 0.0062, 0.0064, 0.0093, 0.0076, 0.0094, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0006, 0.0006], device='cuda:3') 2022-12-08 01:28:03,460 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84555.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:28:07,761 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.298e+02 2.228e+02 2.625e+02 3.303e+02 8.483e+02, threshold=5.249e+02, percent-clipped=5.0 2022-12-08 01:28:18,116 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3046, 2.9569, 2.9173, 1.9362, 2.7341, 2.9609, 3.3100, 2.5520], device='cuda:3'), covar=tensor([0.0656, 0.1171, 0.0979, 0.1726, 0.1091, 0.0719, 0.0617, 0.1374], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0183, 0.0138, 0.0126, 0.0137, 0.0143, 0.0122, 0.0140], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0005, 0.0006], device='cuda:3') 2022-12-08 01:28:28,446 INFO [train.py:873] (3/4) Epoch 12, batch 1400, loss[loss=0.1262, simple_loss=0.1589, pruned_loss=0.04674, over 14270.00 frames. ], tot_loss[loss=0.1281, simple_loss=0.1575, pruned_loss=0.0494, over 1981857.43 frames. ], batch size: 76, lr: 6.69e-03, grad_scale: 8.0 2022-12-08 01:28:31,117 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84586.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:28:36,457 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84592.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:28:48,137 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84605.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:29:04,648 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8035, 2.1913, 3.7890, 3.8878, 3.7983, 2.2591, 3.7684, 2.9640], device='cuda:3'), covar=tensor([0.0368, 0.0899, 0.0749, 0.0375, 0.0357, 0.1333, 0.0379, 0.0790], device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0251, 0.0371, 0.0320, 0.0261, 0.0297, 0.0297, 0.0277], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 01:29:12,435 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84633.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:29:19,444 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84641.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:29:22,251 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2022-12-08 01:29:29,702 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84653.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:29:35,494 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.59 vs. limit=2.0 2022-12-08 01:29:35,866 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.687e+01 2.089e+02 2.664e+02 3.503e+02 6.178e+02, threshold=5.327e+02, percent-clipped=6.0 2022-12-08 01:29:55,408 INFO [train.py:873] (3/4) Epoch 12, batch 1500, loss[loss=0.1152, simple_loss=0.1463, pruned_loss=0.04201, over 5897.00 frames. ], tot_loss[loss=0.126, simple_loss=0.1562, pruned_loss=0.04788, over 2000003.12 frames. ], batch size: 100, lr: 6.68e-03, grad_scale: 8.0 2022-12-08 01:30:00,990 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84689.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:30:03,577 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84692.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:30:05,632 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84694.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:30:55,456 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.2931, 5.0240, 4.6702, 4.8941, 4.8111, 5.1302, 5.2305, 5.2420], device='cuda:3'), covar=tensor([0.0624, 0.0395, 0.1847, 0.2084, 0.0638, 0.0628, 0.0646, 0.0701], device='cuda:3'), in_proj_covar=tensor([0.0367, 0.0254, 0.0429, 0.0547, 0.0321, 0.0417, 0.0388, 0.0362], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 01:30:56,730 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84753.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:31:02,108 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.488e+02 2.424e+02 2.855e+02 3.432e+02 8.447e+02, threshold=5.711e+02, percent-clipped=4.0 2022-12-08 01:31:16,426 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3415, 4.0053, 3.9335, 4.3963, 4.0502, 3.8567, 4.3720, 3.6636], device='cuda:3'), covar=tensor([0.0400, 0.1006, 0.0422, 0.0419, 0.0807, 0.0922, 0.0541, 0.0550], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0260, 0.0183, 0.0181, 0.0175, 0.0147, 0.0271, 0.0160], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 01:31:22,263 INFO [train.py:873] (3/4) Epoch 12, batch 1600, loss[loss=0.1301, simple_loss=0.1623, pruned_loss=0.04902, over 14276.00 frames. ], tot_loss[loss=0.1279, simple_loss=0.1575, pruned_loss=0.04908, over 2011102.43 frames. ], batch size: 76, lr: 6.68e-03, grad_scale: 8.0 2022-12-08 01:31:36,930 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84800.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:31:37,626 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84801.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:32:13,729 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6818, 3.5307, 4.3128, 3.0721, 2.7939, 3.5867, 1.9915, 3.6060], device='cuda:3'), covar=tensor([0.1640, 0.0908, 0.0325, 0.2206, 0.1673, 0.1061, 0.3328, 0.0742], device='cuda:3'), in_proj_covar=tensor([0.0082, 0.0098, 0.0090, 0.0096, 0.0114, 0.0083, 0.0124, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 01:32:25,434 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84855.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:32:29,445 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.720e+01 2.080e+02 2.547e+02 3.159e+02 6.486e+02, threshold=5.095e+02, percent-clipped=3.0 2022-12-08 01:32:30,511 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84861.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:32:38,197 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.60 vs. limit=2.0 2022-12-08 01:32:47,203 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84881.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:32:48,865 INFO [train.py:873] (3/4) Epoch 12, batch 1700, loss[loss=0.1078, simple_loss=0.1478, pruned_loss=0.03393, over 14131.00 frames. ], tot_loss[loss=0.1275, simple_loss=0.1571, pruned_loss=0.0489, over 2029950.37 frames. ], batch size: 99, lr: 6.68e-03, grad_scale: 4.0 2022-12-08 01:32:54,795 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7077, 1.5444, 1.7976, 1.5685, 1.5762, 1.4194, 1.3880, 1.0733], device='cuda:3'), covar=tensor([0.0190, 0.0252, 0.0217, 0.0244, 0.0214, 0.0295, 0.0252, 0.0416], device='cuda:3'), in_proj_covar=tensor([0.0016, 0.0016, 0.0015, 0.0015, 0.0015, 0.0026, 0.0021, 0.0026], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 01:32:57,341 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84892.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:33:06,794 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84903.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:33:39,683 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84940.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:33:49,349 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84951.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:33:57,762 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.082e+02 2.143e+02 2.620e+02 3.286e+02 6.897e+02, threshold=5.240e+02, percent-clipped=6.0 2022-12-08 01:34:17,445 INFO [train.py:873] (3/4) Epoch 12, batch 1800, loss[loss=0.1539, simple_loss=0.1337, pruned_loss=0.08707, over 2631.00 frames. ], tot_loss[loss=0.1259, simple_loss=0.1562, pruned_loss=0.04782, over 2020309.92 frames. ], batch size: 100, lr: 6.67e-03, grad_scale: 4.0 2022-12-08 01:34:22,823 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84989.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:34:25,419 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84992.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:34:47,127 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85012.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:35:10,624 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85040.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:35:13,707 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.69 vs. limit=5.0 2022-12-08 01:35:19,291 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6454, 2.0955, 3.7717, 2.6549, 3.5878, 2.0984, 2.8928, 3.6051], device='cuda:3'), covar=tensor([0.0645, 0.4318, 0.0447, 0.6006, 0.0618, 0.3486, 0.1343, 0.0535], device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0213, 0.0199, 0.0287, 0.0222, 0.0216, 0.0214, 0.0204], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 01:35:29,534 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.171e+02 2.196e+02 2.693e+02 3.379e+02 9.036e+02, threshold=5.386e+02, percent-clipped=1.0 2022-12-08 01:35:44,868 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7805, 2.7177, 2.5996, 2.9187, 2.4162, 2.5178, 2.8367, 2.8028], device='cuda:3'), covar=tensor([0.0801, 0.1078, 0.0983, 0.0777, 0.1317, 0.0783, 0.0885, 0.0850], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0129, 0.0136, 0.0147, 0.0140, 0.0113, 0.0156, 0.0137], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 01:35:48,291 INFO [train.py:873] (3/4) Epoch 12, batch 1900, loss[loss=0.1054, simple_loss=0.1451, pruned_loss=0.03284, over 14675.00 frames. ], tot_loss[loss=0.1265, simple_loss=0.1562, pruned_loss=0.04834, over 1935705.46 frames. ], batch size: 33, lr: 6.67e-03, grad_scale: 4.0 2022-12-08 01:36:09,507 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85107.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:36:16,303 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85115.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:36:52,431 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85156.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:36:56,466 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.007e+02 2.116e+02 2.633e+02 3.281e+02 1.062e+03, threshold=5.267e+02, percent-clipped=7.0 2022-12-08 01:37:03,253 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85168.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 01:37:10,148 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.9923, 1.0020, 1.1084, 0.9063, 0.7907, 0.7670, 0.8771, 0.7974], device='cuda:3'), covar=tensor([0.0173, 0.0152, 0.0124, 0.0157, 0.0198, 0.0385, 0.0231, 0.0342], device='cuda:3'), in_proj_covar=tensor([0.0016, 0.0017, 0.0015, 0.0016, 0.0015, 0.0026, 0.0021, 0.0026], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 01:37:10,153 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85176.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:37:14,068 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85181.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:37:15,663 INFO [train.py:873] (3/4) Epoch 12, batch 2000, loss[loss=0.1207, simple_loss=0.1594, pruned_loss=0.04096, over 14162.00 frames. ], tot_loss[loss=0.1273, simple_loss=0.1567, pruned_loss=0.04895, over 1952749.06 frames. ], batch size: 35, lr: 6.66e-03, grad_scale: 8.0 2022-12-08 01:37:56,106 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85229.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:38:07,310 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.61 vs. limit=5.0 2022-12-08 01:38:25,282 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.573e+01 2.203e+02 2.703e+02 3.626e+02 1.921e+03, threshold=5.405e+02, percent-clipped=8.0 2022-12-08 01:38:43,506 INFO [train.py:873] (3/4) Epoch 12, batch 2100, loss[loss=0.1158, simple_loss=0.1473, pruned_loss=0.0421, over 14109.00 frames. ], tot_loss[loss=0.1266, simple_loss=0.1564, pruned_loss=0.04839, over 1996585.42 frames. ], batch size: 29, lr: 6.66e-03, grad_scale: 4.0 2022-12-08 01:38:48,999 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85289.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:39:04,453 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85307.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:39:30,440 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85337.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:39:39,303 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2368, 1.7763, 2.2146, 1.8846, 2.3285, 2.0624, 2.0008, 2.0713], device='cuda:3'), covar=tensor([0.0548, 0.1975, 0.0570, 0.0887, 0.0432, 0.0947, 0.0439, 0.0556], device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0311, 0.0398, 0.0303, 0.0371, 0.0320, 0.0367, 0.0308], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 01:39:44,518 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.7731, 2.3301, 4.7775, 3.1065, 4.4526, 2.2071, 3.2368, 4.4340], device='cuda:3'), covar=tensor([0.0734, 0.4490, 0.0482, 0.7882, 0.0650, 0.3867, 0.1692, 0.0525], device='cuda:3'), in_proj_covar=tensor([0.0250, 0.0213, 0.0203, 0.0287, 0.0222, 0.0217, 0.0216, 0.0206], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 01:39:51,892 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.818e+01 2.339e+02 2.945e+02 3.643e+02 8.230e+02, threshold=5.891e+02, percent-clipped=7.0 2022-12-08 01:40:10,082 INFO [train.py:873] (3/4) Epoch 12, batch 2200, loss[loss=0.1229, simple_loss=0.1582, pruned_loss=0.04381, over 14136.00 frames. ], tot_loss[loss=0.1271, simple_loss=0.157, pruned_loss=0.0486, over 2055678.77 frames. ], batch size: 29, lr: 6.66e-03, grad_scale: 4.0 2022-12-08 01:40:49,620 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85428.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:40:51,626 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2022-12-08 01:41:03,322 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.72 vs. limit=2.0 2022-12-08 01:41:14,551 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85456.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:41:19,365 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.272e+02 2.150e+02 2.815e+02 3.417e+02 5.996e+02, threshold=5.631e+02, percent-clipped=2.0 2022-12-08 01:41:20,085 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85463.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 01:41:21,785 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9750, 4.9787, 4.7394, 4.9031, 4.6383, 5.4854, 1.9674, 4.4613], device='cuda:3'), covar=tensor([0.0421, 0.0479, 0.0984, 0.0376, 0.0936, 0.0196, 0.4676, 0.0575], device='cuda:3'), in_proj_covar=tensor([0.0161, 0.0172, 0.0142, 0.0140, 0.0203, 0.0137, 0.0159, 0.0188], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 01:41:26,851 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85471.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:41:36,990 INFO [train.py:873] (3/4) Epoch 12, batch 2300, loss[loss=0.1725, simple_loss=0.167, pruned_loss=0.08898, over 1313.00 frames. ], tot_loss[loss=0.1252, simple_loss=0.1558, pruned_loss=0.0473, over 2047970.05 frames. ], batch size: 100, lr: 6.65e-03, grad_scale: 4.0 2022-12-08 01:41:37,142 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85483.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:41:42,434 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85489.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:41:55,417 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85504.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:41:58,954 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8630, 1.6360, 1.9723, 1.6914, 2.0025, 1.7679, 1.6931, 1.8857], device='cuda:3'), covar=tensor([0.0534, 0.1157, 0.0296, 0.0444, 0.0391, 0.0611, 0.0252, 0.0348], device='cuda:3'), in_proj_covar=tensor([0.0353, 0.0315, 0.0401, 0.0305, 0.0376, 0.0322, 0.0371, 0.0310], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 01:42:18,052 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85530.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:42:25,194 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.6473, 2.6821, 2.7861, 2.7917, 2.7130, 2.4417, 1.5013, 2.4398], device='cuda:3'), covar=tensor([0.0517, 0.0461, 0.0486, 0.0345, 0.0422, 0.1094, 0.2596, 0.0407], device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0173, 0.0144, 0.0141, 0.0206, 0.0139, 0.0161, 0.0190], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 01:42:30,192 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85544.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:42:45,172 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.046e+02 2.375e+02 2.779e+02 3.534e+02 5.701e+02, threshold=5.557e+02, percent-clipped=1.0 2022-12-08 01:42:46,235 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8990, 2.4907, 3.4805, 2.1570, 2.1621, 2.8141, 1.6764, 2.8906], device='cuda:3'), covar=tensor([0.0992, 0.1326, 0.0673, 0.2161, 0.2419, 0.0818, 0.3455, 0.1529], device='cuda:3'), in_proj_covar=tensor([0.0082, 0.0098, 0.0090, 0.0096, 0.0116, 0.0084, 0.0124, 0.0091], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 01:43:03,727 INFO [train.py:873] (3/4) Epoch 12, batch 2400, loss[loss=0.1233, simple_loss=0.1581, pruned_loss=0.04426, over 14041.00 frames. ], tot_loss[loss=0.1254, simple_loss=0.1561, pruned_loss=0.0474, over 2052439.09 frames. ], batch size: 26, lr: 6.65e-03, grad_scale: 8.0 2022-12-08 01:43:05,504 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.00 vs. limit=5.0 2022-12-08 01:43:10,517 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85591.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:43:24,426 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85607.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:43:35,556 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.71 vs. limit=2.0 2022-12-08 01:43:40,320 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85625.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:44:07,098 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85655.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:44:12,664 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.126e+02 2.042e+02 2.557e+02 3.044e+02 5.063e+02, threshold=5.114e+02, percent-clipped=0.0 2022-12-08 01:44:30,325 INFO [train.py:873] (3/4) Epoch 12, batch 2500, loss[loss=0.1699, simple_loss=0.1799, pruned_loss=0.07994, over 9480.00 frames. ], tot_loss[loss=0.125, simple_loss=0.1554, pruned_loss=0.04727, over 2027352.31 frames. ], batch size: 100, lr: 6.65e-03, grad_scale: 8.0 2022-12-08 01:44:33,082 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85686.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 01:45:12,499 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8992, 4.9969, 5.3361, 4.4586, 5.0403, 5.4473, 2.2109, 4.7323], device='cuda:3'), covar=tensor([0.0242, 0.0287, 0.0305, 0.0392, 0.0288, 0.0102, 0.2669, 0.0240], device='cuda:3'), in_proj_covar=tensor([0.0159, 0.0171, 0.0141, 0.0139, 0.0203, 0.0136, 0.0159, 0.0187], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 01:45:13,744 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7019, 1.4487, 3.7268, 1.5958, 3.6252, 3.7880, 2.7384, 4.0569], device='cuda:3'), covar=tensor([0.0249, 0.3234, 0.0409, 0.2411, 0.0570, 0.0368, 0.0685, 0.0166], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0158, 0.0160, 0.0169, 0.0170, 0.0174, 0.0134, 0.0144], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-08 01:45:24,013 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9909, 1.8688, 4.5153, 4.1750, 4.1928, 4.6480, 4.1923, 4.6277], device='cuda:3'), covar=tensor([0.1446, 0.1428, 0.0089, 0.0183, 0.0174, 0.0090, 0.0144, 0.0099], device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0157, 0.0127, 0.0166, 0.0143, 0.0138, 0.0118, 0.0119], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 01:45:40,095 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.041e+02 2.190e+02 2.751e+02 3.272e+02 6.413e+02, threshold=5.502e+02, percent-clipped=4.0 2022-12-08 01:45:40,237 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85763.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:45:47,684 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85771.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:45:58,011 INFO [train.py:873] (3/4) Epoch 12, batch 2600, loss[loss=0.1463, simple_loss=0.1687, pruned_loss=0.06196, over 14315.00 frames. ], tot_loss[loss=0.1259, simple_loss=0.1562, pruned_loss=0.04784, over 2056061.02 frames. ], batch size: 46, lr: 6.64e-03, grad_scale: 4.0 2022-12-08 01:45:58,988 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85784.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:46:23,054 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85811.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:46:29,928 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85819.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:46:46,870 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85839.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:47:08,283 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.414e+01 2.205e+02 2.825e+02 3.419e+02 7.121e+02, threshold=5.651e+02, percent-clipped=1.0 2022-12-08 01:47:18,516 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85875.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:47:25,018 INFO [train.py:873] (3/4) Epoch 12, batch 2700, loss[loss=0.1878, simple_loss=0.1635, pruned_loss=0.1061, over 1230.00 frames. ], tot_loss[loss=0.1259, simple_loss=0.1559, pruned_loss=0.04794, over 1981300.90 frames. ], batch size: 100, lr: 6.64e-03, grad_scale: 4.0 2022-12-08 01:47:27,745 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85886.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:48:06,423 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85930.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 01:48:11,839 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85936.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:48:29,632 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85957.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:48:34,526 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.895e+01 2.358e+02 3.006e+02 3.476e+02 6.598e+02, threshold=6.011e+02, percent-clipped=3.0 2022-12-08 01:48:36,482 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.6390, 4.1067, 3.3107, 5.0857, 4.3436, 4.8669, 4.2771, 3.3217], device='cuda:3'), covar=tensor([0.0793, 0.1215, 0.3874, 0.0413, 0.1157, 0.0885, 0.1054, 0.3285], device='cuda:3'), in_proj_covar=tensor([0.0272, 0.0297, 0.0271, 0.0264, 0.0312, 0.0299, 0.0259, 0.0255], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 01:48:51,240 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85981.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 01:48:52,816 INFO [train.py:873] (3/4) Epoch 12, batch 2800, loss[loss=0.1428, simple_loss=0.1376, pruned_loss=0.07398, over 2647.00 frames. ], tot_loss[loss=0.1255, simple_loss=0.1559, pruned_loss=0.04758, over 1987937.37 frames. ], batch size: 100, lr: 6.63e-03, grad_scale: 8.0 2022-12-08 01:48:59,938 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85991.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 01:49:03,388 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4614, 3.9598, 3.2608, 4.8516, 4.2888, 4.6298, 4.0681, 3.2985], device='cuda:3'), covar=tensor([0.0910, 0.1393, 0.3619, 0.1143, 0.1236, 0.1479, 0.1129, 0.3465], device='cuda:3'), in_proj_covar=tensor([0.0272, 0.0298, 0.0271, 0.0265, 0.0313, 0.0299, 0.0260, 0.0256], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 01:49:09,741 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86002.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:49:24,440 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86018.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:49:30,903 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.61 vs. limit=5.0 2022-12-08 01:49:53,847 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86052.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:50:03,560 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.107e+02 2.317e+02 2.627e+02 3.139e+02 9.999e+02, threshold=5.255e+02, percent-clipped=3.0 2022-12-08 01:50:03,733 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86063.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:50:08,692 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86069.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:50:20,498 INFO [train.py:873] (3/4) Epoch 12, batch 2900, loss[loss=0.1799, simple_loss=0.1544, pruned_loss=0.1027, over 1269.00 frames. ], tot_loss[loss=0.1247, simple_loss=0.1553, pruned_loss=0.04707, over 1977243.21 frames. ], batch size: 100, lr: 6.63e-03, grad_scale: 8.0 2022-12-08 01:50:21,328 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86084.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:50:33,667 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 2022-12-08 01:50:46,808 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86113.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:51:01,349 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86130.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:51:03,506 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86132.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:51:09,862 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86139.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:51:30,509 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.360e+01 2.186e+02 2.706e+02 3.386e+02 6.755e+02, threshold=5.412e+02, percent-clipped=5.0 2022-12-08 01:51:47,632 INFO [train.py:873] (3/4) Epoch 12, batch 3000, loss[loss=0.1268, simple_loss=0.156, pruned_loss=0.04878, over 12761.00 frames. ], tot_loss[loss=0.1251, simple_loss=0.1557, pruned_loss=0.0473, over 1945025.98 frames. ], batch size: 100, lr: 6.63e-03, grad_scale: 4.0 2022-12-08 01:51:47,632 INFO [train.py:896] (3/4) Computing validation loss 2022-12-08 01:51:51,875 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2836, 3.1837, 3.1719, 2.2809, 2.7817, 3.2236, 3.3951, 2.7694], device='cuda:3'), covar=tensor([0.0651, 0.0988, 0.0783, 0.1255, 0.0895, 0.0576, 0.0474, 0.1149], device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0178, 0.0137, 0.0125, 0.0134, 0.0141, 0.0119, 0.0137], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0005, 0.0006], device='cuda:3') 2022-12-08 01:51:56,041 INFO [train.py:905] (3/4) Epoch 12, validation: loss=0.1299, simple_loss=0.1698, pruned_loss=0.04501, over 857387.00 frames. 2022-12-08 01:51:56,042 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-08 01:51:58,752 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86186.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:51:59,522 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86187.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:52:38,288 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86231.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:52:41,121 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86234.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:52:52,110 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86246.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:53:03,472 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 2022-12-08 01:53:07,352 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.699e+01 2.315e+02 2.814e+02 3.910e+02 8.994e+02, threshold=5.629e+02, percent-clipped=7.0 2022-12-08 01:53:23,055 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86281.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:53:23,873 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8878, 4.9668, 5.2987, 4.5751, 5.0935, 5.4401, 1.9690, 4.7414], device='cuda:3'), covar=tensor([0.0212, 0.0279, 0.0331, 0.0448, 0.0280, 0.0175, 0.2962, 0.0256], device='cuda:3'), in_proj_covar=tensor([0.0158, 0.0169, 0.0139, 0.0137, 0.0201, 0.0134, 0.0156, 0.0185], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 01:53:24,631 INFO [train.py:873] (3/4) Epoch 12, batch 3100, loss[loss=0.09606, simple_loss=0.1371, pruned_loss=0.0275, over 13997.00 frames. ], tot_loss[loss=0.1257, simple_loss=0.1559, pruned_loss=0.04769, over 1980519.75 frames. ], batch size: 19, lr: 6.62e-03, grad_scale: 4.0 2022-12-08 01:53:25,533 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1922, 1.9852, 2.1079, 2.1973, 2.0966, 2.0128, 2.2339, 1.8980], device='cuda:3'), covar=tensor([0.0937, 0.1545, 0.0711, 0.0823, 0.1074, 0.0811, 0.1062, 0.0833], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0264, 0.0187, 0.0182, 0.0178, 0.0149, 0.0273, 0.0160], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 01:53:27,174 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86286.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 01:53:45,076 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86307.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:53:50,953 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86313.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:54:04,951 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86329.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:54:17,999 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8150, 1.7721, 2.0091, 2.1063, 1.7850, 1.6566, 1.9864, 1.7866], device='cuda:3'), covar=tensor([0.0401, 0.0557, 0.0469, 0.0490, 0.0878, 0.0506, 0.0416, 0.0588], device='cuda:3'), in_proj_covar=tensor([0.0016, 0.0017, 0.0015, 0.0016, 0.0016, 0.0027, 0.0021, 0.0026], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 01:54:30,986 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86358.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:54:36,315 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.153e+02 2.157e+02 2.639e+02 3.347e+02 7.885e+02, threshold=5.278e+02, percent-clipped=4.0 2022-12-08 01:54:40,771 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2022-12-08 01:54:51,746 INFO [train.py:873] (3/4) Epoch 12, batch 3200, loss[loss=0.1711, simple_loss=0.1627, pruned_loss=0.08978, over 2608.00 frames. ], tot_loss[loss=0.1258, simple_loss=0.1555, pruned_loss=0.04803, over 1851749.77 frames. ], batch size: 100, lr: 6.62e-03, grad_scale: 8.0 2022-12-08 01:55:14,582 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86408.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:55:28,826 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86425.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:55:38,699 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8626, 1.9487, 1.5945, 1.9822, 1.8137, 1.8771, 1.7694, 1.6558], device='cuda:3'), covar=tensor([0.0985, 0.0646, 0.1796, 0.0526, 0.1047, 0.0523, 0.1233, 0.0818], device='cuda:3'), in_proj_covar=tensor([0.0269, 0.0294, 0.0268, 0.0264, 0.0313, 0.0299, 0.0258, 0.0253], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 01:55:42,194 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86440.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:55:46,309 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86445.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:56:02,446 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.374e+02 2.245e+02 2.734e+02 3.789e+02 7.990e+02, threshold=5.469e+02, percent-clipped=5.0 2022-12-08 01:56:09,024 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.36 vs. limit=5.0 2022-12-08 01:56:20,044 INFO [train.py:873] (3/4) Epoch 12, batch 3300, loss[loss=0.1692, simple_loss=0.1807, pruned_loss=0.07889, over 8560.00 frames. ], tot_loss[loss=0.1249, simple_loss=0.1552, pruned_loss=0.04735, over 1905318.95 frames. ], batch size: 100, lr: 6.61e-03, grad_scale: 8.0 2022-12-08 01:56:35,431 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86501.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:56:39,522 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86506.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 01:56:52,424 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3595, 2.5596, 2.6533, 2.6509, 2.2185, 2.6726, 2.5132, 1.3876], device='cuda:3'), covar=tensor([0.1534, 0.0871, 0.0915, 0.0616, 0.0922, 0.0769, 0.1020, 0.2603], device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0080, 0.0063, 0.0067, 0.0092, 0.0078, 0.0095, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0006, 0.0006], device='cuda:3') 2022-12-08 01:57:01,833 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86531.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:57:30,031 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.019e+02 2.093e+02 2.507e+02 3.209e+02 6.718e+02, threshold=5.013e+02, percent-clipped=2.0 2022-12-08 01:57:42,902 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86579.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:57:43,883 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86580.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:57:46,320 INFO [train.py:873] (3/4) Epoch 12, batch 3400, loss[loss=0.1229, simple_loss=0.1597, pruned_loss=0.04309, over 13930.00 frames. ], tot_loss[loss=0.1247, simple_loss=0.1551, pruned_loss=0.04719, over 1979677.18 frames. ], batch size: 23, lr: 6.61e-03, grad_scale: 8.0 2022-12-08 01:57:48,908 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86586.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 01:58:03,528 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86602.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:58:13,194 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86613.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:58:31,347 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86634.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 01:58:38,110 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86641.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:58:52,804 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86658.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:58:55,356 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86661.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:58:57,881 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.288e+02 2.267e+02 2.858e+02 3.502e+02 8.320e+02, threshold=5.717e+02, percent-clipped=7.0 2022-12-08 01:59:07,414 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.7779, 2.1814, 4.7188, 3.0321, 4.3829, 2.1014, 3.3423, 4.4317], device='cuda:3'), covar=tensor([0.0383, 0.4207, 0.0287, 0.7095, 0.0884, 0.3566, 0.1348, 0.0296], device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0212, 0.0203, 0.0287, 0.0223, 0.0216, 0.0214, 0.0204], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 01:59:09,340 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3731, 2.3490, 2.4616, 2.5086, 2.3950, 2.0798, 1.4468, 2.1447], device='cuda:3'), covar=tensor([0.0530, 0.0520, 0.0497, 0.0396, 0.0490, 0.1131, 0.2215, 0.0469], device='cuda:3'), in_proj_covar=tensor([0.0160, 0.0170, 0.0140, 0.0138, 0.0203, 0.0135, 0.0158, 0.0187], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 01:59:14,483 INFO [train.py:873] (3/4) Epoch 12, batch 3500, loss[loss=0.1423, simple_loss=0.1502, pruned_loss=0.06724, over 3876.00 frames. ], tot_loss[loss=0.1247, simple_loss=0.1552, pruned_loss=0.04717, over 2005720.86 frames. ], batch size: 100, lr: 6.61e-03, grad_scale: 8.0 2022-12-08 01:59:26,417 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5573, 2.2688, 3.4502, 3.5905, 3.4264, 2.3479, 3.4568, 2.8091], device='cuda:3'), covar=tensor([0.0340, 0.0892, 0.0645, 0.0385, 0.0432, 0.1223, 0.0404, 0.0807], device='cuda:3'), in_proj_covar=tensor([0.0283, 0.0251, 0.0368, 0.0321, 0.0263, 0.0299, 0.0298, 0.0279], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 01:59:34,034 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86706.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:59:35,795 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86708.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 01:59:39,787 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 2022-12-08 01:59:42,568 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1115, 1.1178, 0.9154, 1.0806, 1.0519, 0.7197, 0.9440, 1.1121], device='cuda:3'), covar=tensor([0.0365, 0.0621, 0.0528, 0.0579, 0.0400, 0.0401, 0.0660, 0.0522], device='cuda:3'), in_proj_covar=tensor([0.0027, 0.0027, 0.0030, 0.0026, 0.0029, 0.0040, 0.0028, 0.0029], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 01:59:51,268 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86725.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:00:05,633 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4983, 3.6505, 3.4906, 3.6086, 2.7285, 3.5983, 3.5255, 1.8626], device='cuda:3'), covar=tensor([0.2118, 0.1694, 0.1096, 0.0972, 0.1033, 0.0538, 0.1046, 0.2548], device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0080, 0.0064, 0.0067, 0.0093, 0.0078, 0.0096, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0006, 0.0006], device='cuda:3') 2022-12-08 02:00:17,813 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86756.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:00:24,901 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.146e+02 2.023e+02 2.697e+02 3.316e+02 6.067e+02, threshold=5.393e+02, percent-clipped=1.0 2022-12-08 02:00:32,566 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86773.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:00:41,038 INFO [train.py:873] (3/4) Epoch 12, batch 3600, loss[loss=0.1275, simple_loss=0.1532, pruned_loss=0.05084, over 10311.00 frames. ], tot_loss[loss=0.1233, simple_loss=0.154, pruned_loss=0.04632, over 2001427.28 frames. ], batch size: 100, lr: 6.60e-03, grad_scale: 8.0 2022-12-08 02:00:52,477 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86796.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:00:57,625 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86801.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 02:01:08,595 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86814.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:01:51,778 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.100e+02 2.201e+02 2.694e+02 3.354e+02 5.968e+02, threshold=5.388e+02, percent-clipped=2.0 2022-12-08 02:02:01,347 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86875.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:02:04,444 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8716, 1.5975, 4.0077, 3.8223, 3.8311, 4.0905, 3.4588, 4.0653], device='cuda:3'), covar=tensor([0.1548, 0.1548, 0.0117, 0.0213, 0.0195, 0.0122, 0.0257, 0.0121], device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0157, 0.0126, 0.0165, 0.0144, 0.0139, 0.0118, 0.0119], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 02:02:04,485 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86878.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:02:08,671 INFO [train.py:873] (3/4) Epoch 12, batch 3700, loss[loss=0.1028, simple_loss=0.1468, pruned_loss=0.02943, over 14250.00 frames. ], tot_loss[loss=0.1236, simple_loss=0.1546, pruned_loss=0.04628, over 2012139.35 frames. ], batch size: 44, lr: 6.60e-03, grad_scale: 8.0 2022-12-08 02:02:12,231 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3694, 1.9142, 3.4017, 2.4934, 3.2974, 1.8991, 2.7873, 3.3149], device='cuda:3'), covar=tensor([0.0898, 0.4525, 0.0545, 0.5110, 0.0783, 0.3681, 0.1260, 0.0581], device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0212, 0.0203, 0.0285, 0.0223, 0.0216, 0.0213, 0.0207], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 02:02:22,108 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8250, 2.6135, 3.3152, 2.0994, 2.0929, 2.8069, 1.4984, 2.8077], device='cuda:3'), covar=tensor([0.0974, 0.1215, 0.0531, 0.2604, 0.2497, 0.1000, 0.3981, 0.0942], device='cuda:3'), in_proj_covar=tensor([0.0081, 0.0098, 0.0091, 0.0098, 0.0115, 0.0085, 0.0124, 0.0092], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 02:02:24,571 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86902.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:02:31,735 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2022-12-08 02:02:54,802 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86936.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:02:57,484 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86939.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:03:06,365 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86950.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:03:19,132 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.304e+02 2.133e+02 2.735e+02 3.535e+02 1.030e+03, threshold=5.470e+02, percent-clipped=7.0 2022-12-08 02:03:24,740 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.11 vs. limit=2.0 2022-12-08 02:03:35,278 INFO [train.py:873] (3/4) Epoch 12, batch 3800, loss[loss=0.1201, simple_loss=0.1528, pruned_loss=0.04369, over 14212.00 frames. ], tot_loss[loss=0.1245, simple_loss=0.1551, pruned_loss=0.04699, over 1948438.07 frames. ], batch size: 35, lr: 6.60e-03, grad_scale: 4.0 2022-12-08 02:03:48,941 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8992, 1.4623, 3.1808, 2.9580, 3.0684, 3.2217, 2.4261, 3.2050], device='cuda:3'), covar=tensor([0.1170, 0.1378, 0.0139, 0.0268, 0.0279, 0.0150, 0.0397, 0.0162], device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0156, 0.0125, 0.0163, 0.0142, 0.0138, 0.0118, 0.0118], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 02:03:53,067 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87002.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 02:04:45,888 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87063.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 02:04:47,287 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.323e+02 2.236e+02 2.550e+02 3.017e+02 4.626e+02, threshold=5.100e+02, percent-clipped=0.0 2022-12-08 02:05:03,580 INFO [train.py:873] (3/4) Epoch 12, batch 3900, loss[loss=0.1535, simple_loss=0.1463, pruned_loss=0.08041, over 1249.00 frames. ], tot_loss[loss=0.1241, simple_loss=0.1547, pruned_loss=0.04674, over 1964740.94 frames. ], batch size: 100, lr: 6.59e-03, grad_scale: 4.0 2022-12-08 02:05:14,069 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87096.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:05:18,341 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87101.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 02:05:30,888 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3458, 4.0702, 3.9672, 4.4298, 4.1022, 3.9282, 4.4152, 3.6409], device='cuda:3'), covar=tensor([0.0467, 0.0974, 0.0407, 0.0423, 0.0787, 0.0783, 0.0532, 0.0564], device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0265, 0.0187, 0.0183, 0.0177, 0.0151, 0.0272, 0.0161], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 02:05:56,018 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87144.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:06:00,185 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87149.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:06:14,161 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.297e+02 2.280e+02 2.775e+02 3.261e+02 6.407e+02, threshold=5.550e+02, percent-clipped=4.0 2022-12-08 02:06:18,669 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87170.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:06:29,721 INFO [train.py:873] (3/4) Epoch 12, batch 4000, loss[loss=0.09776, simple_loss=0.1417, pruned_loss=0.0269, over 14539.00 frames. ], tot_loss[loss=0.1237, simple_loss=0.1545, pruned_loss=0.04648, over 1947308.96 frames. ], batch size: 49, lr: 6.59e-03, grad_scale: 8.0 2022-12-08 02:07:10,522 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87230.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:07:11,561 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1937, 2.4555, 4.0354, 4.1863, 4.0903, 2.5408, 4.0524, 3.3217], device='cuda:3'), covar=tensor([0.0292, 0.0873, 0.0726, 0.0372, 0.0329, 0.1343, 0.0313, 0.0738], device='cuda:3'), in_proj_covar=tensor([0.0286, 0.0254, 0.0370, 0.0323, 0.0264, 0.0301, 0.0303, 0.0282], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 02:07:14,073 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87234.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:07:16,236 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87236.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:07:40,897 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.276e+02 2.153e+02 2.853e+02 3.504e+02 7.130e+02, threshold=5.705e+02, percent-clipped=5.0 2022-12-08 02:07:57,457 INFO [train.py:873] (3/4) Epoch 12, batch 4100, loss[loss=0.1259, simple_loss=0.1569, pruned_loss=0.04745, over 14248.00 frames. ], tot_loss[loss=0.1249, simple_loss=0.1549, pruned_loss=0.04742, over 1924630.08 frames. ], batch size: 80, lr: 6.58e-03, grad_scale: 8.0 2022-12-08 02:07:58,292 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87284.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:08:04,407 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87291.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:08:20,377 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0472, 1.9314, 2.0713, 2.0967, 2.0167, 1.6231, 1.3276, 1.7982], device='cuda:3'), covar=tensor([0.0515, 0.0538, 0.0552, 0.0328, 0.0486, 0.1389, 0.2240, 0.0464], device='cuda:3'), in_proj_covar=tensor([0.0157, 0.0167, 0.0139, 0.0136, 0.0198, 0.0135, 0.0155, 0.0185], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 02:08:58,885 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.0763, 4.8489, 4.4931, 4.6575, 4.7179, 4.9910, 5.1176, 5.0446], device='cuda:3'), covar=tensor([0.0897, 0.0437, 0.1925, 0.2788, 0.0714, 0.0739, 0.0942, 0.0839], device='cuda:3'), in_proj_covar=tensor([0.0378, 0.0260, 0.0435, 0.0557, 0.0329, 0.0425, 0.0389, 0.0363], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 02:09:02,769 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87358.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 02:09:06,586 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87362.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:09:09,284 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1487, 2.8270, 2.2748, 3.2271, 3.0800, 3.1513, 2.7671, 2.2614], device='cuda:3'), covar=tensor([0.0935, 0.1749, 0.4065, 0.0849, 0.1119, 0.1005, 0.1619, 0.3609], device='cuda:3'), in_proj_covar=tensor([0.0270, 0.0295, 0.0267, 0.0263, 0.0313, 0.0295, 0.0255, 0.0251], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2022-12-08 02:09:09,786 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.248e+02 2.165e+02 2.894e+02 3.729e+02 1.916e+03, threshold=5.789e+02, percent-clipped=7.0 2022-12-08 02:09:22,420 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5084, 4.1913, 4.1273, 4.5572, 4.1014, 3.5766, 4.5627, 4.3787], device='cuda:3'), covar=tensor([0.0594, 0.0745, 0.0800, 0.0571, 0.0913, 0.0795, 0.0594, 0.0628], device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0132, 0.0139, 0.0149, 0.0142, 0.0115, 0.0158, 0.0136], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 02:09:24,060 INFO [train.py:873] (3/4) Epoch 12, batch 4200, loss[loss=0.1104, simple_loss=0.1525, pruned_loss=0.03414, over 14266.00 frames. ], tot_loss[loss=0.1256, simple_loss=0.1558, pruned_loss=0.04765, over 1949450.89 frames. ], batch size: 60, lr: 6.58e-03, grad_scale: 4.0 2022-12-08 02:09:58,911 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87423.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:10:32,210 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87461.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:10:34,300 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=9.35 vs. limit=5.0 2022-12-08 02:10:36,120 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.802e+01 2.101e+02 2.786e+02 3.317e+02 5.801e+02, threshold=5.572e+02, percent-clipped=2.0 2022-12-08 02:10:36,249 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8947, 1.2911, 2.0279, 1.3320, 2.0113, 2.0959, 1.7504, 2.1375], device='cuda:3'), covar=tensor([0.0284, 0.1827, 0.0467, 0.1634, 0.0509, 0.0497, 0.0894, 0.0360], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0156, 0.0159, 0.0167, 0.0168, 0.0176, 0.0131, 0.0144], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-08 02:10:39,501 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87470.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:10:51,435 INFO [train.py:873] (3/4) Epoch 12, batch 4300, loss[loss=0.144, simple_loss=0.1485, pruned_loss=0.06972, over 3830.00 frames. ], tot_loss[loss=0.125, simple_loss=0.1556, pruned_loss=0.04724, over 1951102.88 frames. ], batch size: 100, lr: 6.58e-03, grad_scale: 4.0 2022-12-08 02:10:56,970 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1851, 1.2940, 1.4106, 0.9642, 0.8911, 1.2428, 0.8569, 1.2858], device='cuda:3'), covar=tensor([0.1766, 0.2771, 0.0921, 0.2981, 0.3571, 0.0965, 0.2303, 0.1213], device='cuda:3'), in_proj_covar=tensor([0.0081, 0.0097, 0.0091, 0.0096, 0.0115, 0.0085, 0.0124, 0.0091], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 02:11:03,071 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0635, 2.7176, 2.8371, 1.9564, 2.5147, 2.7962, 3.1085, 2.4583], device='cuda:3'), covar=tensor([0.0693, 0.1006, 0.0895, 0.1541, 0.1021, 0.0696, 0.0670, 0.1355], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0179, 0.0138, 0.0124, 0.0134, 0.0144, 0.0122, 0.0138], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0005, 0.0006], device='cuda:3') 2022-12-08 02:11:10,473 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87505.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:11:13,472 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.27 vs. limit=5.0 2022-12-08 02:11:21,738 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87518.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:11:25,909 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87522.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:11:36,424 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87534.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:12:04,323 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.237e+02 2.182e+02 2.709e+02 3.348e+02 1.136e+03, threshold=5.418e+02, percent-clipped=3.0 2022-12-08 02:12:04,526 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87566.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 02:12:18,398 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87582.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:12:19,184 INFO [train.py:873] (3/4) Epoch 12, batch 4400, loss[loss=0.1106, simple_loss=0.1522, pruned_loss=0.03449, over 14281.00 frames. ], tot_loss[loss=0.1247, simple_loss=0.1555, pruned_loss=0.04697, over 1960563.86 frames. ], batch size: 44, lr: 6.57e-03, grad_scale: 8.0 2022-12-08 02:12:19,955 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87584.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:12:21,509 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87586.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:12:24,124 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87589.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:12:36,340 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.57 vs. limit=2.0 2022-12-08 02:12:38,849 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87605.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:13:13,711 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87645.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 02:13:18,289 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87650.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:13:24,948 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87658.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 02:13:31,922 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87666.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 02:13:32,488 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.847e+01 2.130e+02 2.616e+02 3.250e+02 5.248e+02, threshold=5.232e+02, percent-clipped=0.0 2022-12-08 02:13:35,517 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2022-12-08 02:13:46,588 INFO [train.py:873] (3/4) Epoch 12, batch 4500, loss[loss=0.1323, simple_loss=0.1677, pruned_loss=0.04845, over 14368.00 frames. ], tot_loss[loss=0.1234, simple_loss=0.1551, pruned_loss=0.04587, over 2031491.52 frames. ], batch size: 55, lr: 6.57e-03, grad_scale: 4.0 2022-12-08 02:14:06,776 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87706.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 02:14:17,088 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87718.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:14:59,665 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.324e+02 2.389e+02 2.965e+02 3.391e+02 5.507e+02, threshold=5.931e+02, percent-clipped=1.0 2022-12-08 02:15:13,019 INFO [train.py:873] (3/4) Epoch 12, batch 4600, loss[loss=0.1147, simple_loss=0.1541, pruned_loss=0.03765, over 13903.00 frames. ], tot_loss[loss=0.1245, simple_loss=0.1556, pruned_loss=0.0467, over 1985641.86 frames. ], batch size: 20, lr: 6.57e-03, grad_scale: 4.0 2022-12-08 02:15:43,394 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87817.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:15:45,648 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.71 vs. limit=2.0 2022-12-08 02:16:06,044 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=9.30 vs. limit=5.0 2022-12-08 02:16:21,474 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87861.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 02:16:26,308 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.265e+02 2.249e+02 2.775e+02 3.473e+02 5.522e+02, threshold=5.551e+02, percent-clipped=0.0 2022-12-08 02:16:40,649 INFO [train.py:873] (3/4) Epoch 12, batch 4700, loss[loss=0.1467, simple_loss=0.1714, pruned_loss=0.061, over 14207.00 frames. ], tot_loss[loss=0.1235, simple_loss=0.155, pruned_loss=0.04596, over 2023983.08 frames. ], batch size: 89, lr: 6.56e-03, grad_scale: 4.0 2022-12-08 02:16:43,346 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87886.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:17:15,803 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7804, 1.5555, 2.9309, 1.6037, 2.9833, 2.9210, 2.2649, 3.0790], device='cuda:3'), covar=tensor([0.0277, 0.2506, 0.0382, 0.1902, 0.0352, 0.0403, 0.0892, 0.0245], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0157, 0.0159, 0.0168, 0.0169, 0.0175, 0.0131, 0.0145], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-08 02:17:25,814 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87934.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:17:31,123 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87940.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 02:17:35,294 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87945.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:17:49,001 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87961.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 02:17:54,230 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.984e+01 2.201e+02 2.736e+02 3.426e+02 5.956e+02, threshold=5.473e+02, percent-clipped=2.0 2022-12-08 02:18:08,684 INFO [train.py:873] (3/4) Epoch 12, batch 4800, loss[loss=0.1512, simple_loss=0.1627, pruned_loss=0.06983, over 5998.00 frames. ], tot_loss[loss=0.1237, simple_loss=0.1553, pruned_loss=0.04605, over 2049551.07 frames. ], batch size: 100, lr: 6.56e-03, grad_scale: 8.0 2022-12-08 02:18:39,694 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88018.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:18:48,224 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8160, 3.0212, 2.9792, 3.0088, 2.2928, 3.0243, 2.9506, 1.5223], device='cuda:3'), covar=tensor([0.1473, 0.0723, 0.0999, 0.0657, 0.1099, 0.0751, 0.0965, 0.2564], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0081, 0.0065, 0.0069, 0.0094, 0.0079, 0.0095, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0006, 0.0006], device='cuda:3') 2022-12-08 02:19:18,417 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2227, 1.9964, 3.4086, 2.5323, 3.2218, 1.9398, 2.7557, 3.2291], device='cuda:3'), covar=tensor([0.0755, 0.4091, 0.0475, 0.4526, 0.0876, 0.3049, 0.1054, 0.0535], device='cuda:3'), in_proj_covar=tensor([0.0248, 0.0211, 0.0203, 0.0284, 0.0224, 0.0211, 0.0212, 0.0206], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 02:19:20,981 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([6.1838, 5.5903, 5.6135, 6.1744, 5.6787, 4.8261, 6.0145, 5.0291], device='cuda:3'), covar=tensor([0.0167, 0.0674, 0.0245, 0.0256, 0.0605, 0.0292, 0.0391, 0.0372], device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0263, 0.0187, 0.0183, 0.0175, 0.0151, 0.0270, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 02:19:21,840 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=88066.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:19:22,466 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.123e+02 2.264e+02 2.796e+02 3.655e+02 7.619e+02, threshold=5.592e+02, percent-clipped=2.0 2022-12-08 02:19:36,609 INFO [train.py:873] (3/4) Epoch 12, batch 4900, loss[loss=0.1286, simple_loss=0.1675, pruned_loss=0.04485, over 14376.00 frames. ], tot_loss[loss=0.1249, simple_loss=0.1557, pruned_loss=0.04707, over 1998978.30 frames. ], batch size: 55, lr: 6.55e-03, grad_scale: 8.0 2022-12-08 02:19:51,996 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88100.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:20:07,118 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88117.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:20:40,987 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.68 vs. limit=5.0 2022-12-08 02:20:45,899 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88161.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 02:20:45,947 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88161.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:20:49,493 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=88165.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:20:51,095 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.528e+02 2.168e+02 2.607e+02 3.248e+02 6.908e+02, threshold=5.213e+02, percent-clipped=2.0 2022-12-08 02:21:05,440 INFO [train.py:873] (3/4) Epoch 12, batch 5000, loss[loss=0.1206, simple_loss=0.1377, pruned_loss=0.05176, over 3877.00 frames. ], tot_loss[loss=0.1247, simple_loss=0.1556, pruned_loss=0.04691, over 2047503.24 frames. ], batch size: 100, lr: 6.55e-03, grad_scale: 8.0 2022-12-08 02:21:28,451 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=88209.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:21:55,852 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88240.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:22:00,554 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88245.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:22:14,888 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88261.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 02:22:19,851 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.015e+02 2.163e+02 2.571e+02 3.278e+02 8.920e+02, threshold=5.143e+02, percent-clipped=2.0 2022-12-08 02:22:33,615 INFO [train.py:873] (3/4) Epoch 12, batch 5100, loss[loss=0.1108, simple_loss=0.1414, pruned_loss=0.04005, over 6944.00 frames. ], tot_loss[loss=0.1246, simple_loss=0.1555, pruned_loss=0.04682, over 2011954.41 frames. ], batch size: 100, lr: 6.55e-03, grad_scale: 8.0 2022-12-08 02:22:38,217 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=88288.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:22:41,238 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88291.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 02:22:42,679 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=88293.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:22:51,555 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0788, 2.8753, 2.5924, 2.7732, 2.9987, 2.9937, 3.0266, 3.0291], device='cuda:3'), covar=tensor([0.1013, 0.0701, 0.2339, 0.2778, 0.0832, 0.1055, 0.1324, 0.0942], device='cuda:3'), in_proj_covar=tensor([0.0376, 0.0262, 0.0435, 0.0559, 0.0329, 0.0429, 0.0389, 0.0368], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 02:22:56,550 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=88309.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:23:32,407 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2413, 2.9641, 2.9648, 1.9954, 2.7197, 2.9689, 3.3373, 2.6219], device='cuda:3'), covar=tensor([0.0668, 0.1142, 0.0872, 0.1604, 0.0929, 0.0707, 0.0625, 0.1275], device='cuda:3'), in_proj_covar=tensor([0.0140, 0.0179, 0.0137, 0.0125, 0.0136, 0.0145, 0.0121, 0.0138], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0005, 0.0006], device='cuda:3') 2022-12-08 02:23:34,142 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88352.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 02:23:46,914 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.222e+02 2.245e+02 2.755e+02 3.354e+02 6.298e+02, threshold=5.511e+02, percent-clipped=3.0 2022-12-08 02:24:01,890 INFO [train.py:873] (3/4) Epoch 12, batch 5200, loss[loss=0.1081, simple_loss=0.1473, pruned_loss=0.03444, over 14411.00 frames. ], tot_loss[loss=0.1253, simple_loss=0.1559, pruned_loss=0.04731, over 2032093.75 frames. ], batch size: 53, lr: 6.54e-03, grad_scale: 8.0 2022-12-08 02:24:04,189 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.09 vs. limit=5.0 2022-12-08 02:24:51,141 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0925, 2.2494, 2.4231, 2.4000, 1.9814, 2.4087, 2.2623, 1.2613], device='cuda:3'), covar=tensor([0.1176, 0.0724, 0.0642, 0.0581, 0.1014, 0.0580, 0.1092, 0.2256], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0080, 0.0065, 0.0068, 0.0094, 0.0079, 0.0095, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0006, 0.0006], device='cuda:3') 2022-12-08 02:24:55,910 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.13 vs. limit=2.0 2022-12-08 02:25:06,272 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=88456.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:25:15,465 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.239e+02 2.113e+02 2.685e+02 3.247e+02 4.954e+02, threshold=5.369e+02, percent-clipped=0.0 2022-12-08 02:25:29,580 INFO [train.py:873] (3/4) Epoch 12, batch 5300, loss[loss=0.1684, simple_loss=0.153, pruned_loss=0.09193, over 1251.00 frames. ], tot_loss[loss=0.1246, simple_loss=0.1555, pruned_loss=0.04689, over 1966437.84 frames. ], batch size: 100, lr: 6.54e-03, grad_scale: 8.0 2022-12-08 02:25:54,730 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2022-12-08 02:26:04,073 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3510, 2.3632, 1.9116, 2.4268, 2.2589, 2.3080, 2.1358, 2.0348], device='cuda:3'), covar=tensor([0.0841, 0.0981, 0.2174, 0.0636, 0.1003, 0.0750, 0.1299, 0.1143], device='cuda:3'), in_proj_covar=tensor([0.0265, 0.0296, 0.0265, 0.0263, 0.0310, 0.0295, 0.0254, 0.0248], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2022-12-08 02:26:18,850 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.78 vs. limit=5.0 2022-12-08 02:26:28,850 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.2432, 3.7743, 3.0479, 4.5054, 4.2328, 4.2946, 3.8299, 3.0408], device='cuda:3'), covar=tensor([0.0829, 0.1304, 0.3713, 0.0546, 0.0968, 0.1265, 0.1186, 0.3482], device='cuda:3'), in_proj_covar=tensor([0.0264, 0.0295, 0.0265, 0.0263, 0.0309, 0.0294, 0.0253, 0.0247], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004], device='cuda:3') 2022-12-08 02:26:43,276 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.234e+01 2.154e+02 2.570e+02 3.365e+02 7.012e+02, threshold=5.140e+02, percent-clipped=5.0 2022-12-08 02:26:57,899 INFO [train.py:873] (3/4) Epoch 12, batch 5400, loss[loss=0.163, simple_loss=0.1517, pruned_loss=0.08718, over 1311.00 frames. ], tot_loss[loss=0.1252, simple_loss=0.156, pruned_loss=0.04715, over 2011408.03 frames. ], batch size: 100, lr: 6.54e-03, grad_scale: 8.0 2022-12-08 02:27:53,981 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=88647.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 02:28:11,935 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.088e+01 2.108e+02 2.673e+02 3.283e+02 5.238e+02, threshold=5.346e+02, percent-clipped=1.0 2022-12-08 02:28:26,024 INFO [train.py:873] (3/4) Epoch 12, batch 5500, loss[loss=0.1198, simple_loss=0.143, pruned_loss=0.04832, over 5973.00 frames. ], tot_loss[loss=0.1246, simple_loss=0.1557, pruned_loss=0.04678, over 2001366.37 frames. ], batch size: 100, lr: 6.53e-03, grad_scale: 8.0 2022-12-08 02:29:29,690 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88756.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:29:32,748 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2022-12-08 02:29:38,829 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.261e+02 2.096e+02 2.766e+02 3.641e+02 8.752e+02, threshold=5.532e+02, percent-clipped=4.0 2022-12-08 02:29:45,907 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.67 vs. limit=5.0 2022-12-08 02:29:53,655 INFO [train.py:873] (3/4) Epoch 12, batch 5600, loss[loss=0.1478, simple_loss=0.1478, pruned_loss=0.07395, over 1312.00 frames. ], tot_loss[loss=0.1251, simple_loss=0.1559, pruned_loss=0.04716, over 1946121.55 frames. ], batch size: 100, lr: 6.53e-03, grad_scale: 8.0 2022-12-08 02:30:12,322 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=88804.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:30:57,063 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.58 vs. limit=2.0 2022-12-08 02:31:02,053 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.77 vs. limit=2.0 2022-12-08 02:31:06,999 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.050e+02 2.217e+02 2.679e+02 3.309e+02 6.646e+02, threshold=5.358e+02, percent-clipped=2.0 2022-12-08 02:31:07,244 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7544, 1.5417, 1.4831, 1.7246, 1.8461, 1.5277, 1.6590, 1.1635], device='cuda:3'), covar=tensor([0.0233, 0.0869, 0.0526, 0.0346, 0.0285, 0.0323, 0.0301, 0.0516], device='cuda:3'), in_proj_covar=tensor([0.0017, 0.0018, 0.0015, 0.0016, 0.0016, 0.0027, 0.0022, 0.0027], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 02:31:17,446 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4510, 2.4286, 1.9527, 2.0201, 2.3552, 2.4551, 2.4531, 2.3835], device='cuda:3'), covar=tensor([0.1344, 0.0866, 0.3069, 0.3529, 0.1532, 0.1387, 0.1628, 0.1445], device='cuda:3'), in_proj_covar=tensor([0.0377, 0.0261, 0.0434, 0.0555, 0.0327, 0.0426, 0.0385, 0.0366], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 02:31:20,848 INFO [train.py:873] (3/4) Epoch 12, batch 5700, loss[loss=0.112, simple_loss=0.1493, pruned_loss=0.03735, over 14330.00 frames. ], tot_loss[loss=0.1246, simple_loss=0.1555, pruned_loss=0.04692, over 1976896.02 frames. ], batch size: 60, lr: 6.52e-03, grad_scale: 8.0 2022-12-08 02:32:08,652 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1545, 2.0415, 1.8255, 1.9196, 2.0826, 2.1095, 2.0559, 2.0719], device='cuda:3'), covar=tensor([0.1049, 0.0839, 0.2544, 0.2385, 0.1161, 0.1014, 0.1530, 0.1073], device='cuda:3'), in_proj_covar=tensor([0.0376, 0.0261, 0.0433, 0.0553, 0.0325, 0.0425, 0.0383, 0.0365], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 02:32:16,555 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88947.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 02:32:26,393 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8030, 0.7579, 0.8411, 0.7653, 0.7901, 0.4768, 0.6001, 0.7626], device='cuda:3'), covar=tensor([0.0173, 0.0146, 0.0131, 0.0132, 0.0166, 0.0295, 0.0191, 0.0255], device='cuda:3'), in_proj_covar=tensor([0.0017, 0.0018, 0.0015, 0.0016, 0.0016, 0.0027, 0.0022, 0.0027], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 02:32:30,963 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2022-12-08 02:32:32,857 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.260e+02 2.272e+02 2.731e+02 3.384e+02 1.131e+03, threshold=5.463e+02, percent-clipped=4.0 2022-12-08 02:32:42,640 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.47 vs. limit=2.0 2022-12-08 02:32:47,398 INFO [train.py:873] (3/4) Epoch 12, batch 5800, loss[loss=0.1671, simple_loss=0.1793, pruned_loss=0.07744, over 9503.00 frames. ], tot_loss[loss=0.1256, simple_loss=0.1559, pruned_loss=0.04761, over 1964591.76 frames. ], batch size: 100, lr: 6.52e-03, grad_scale: 8.0 2022-12-08 02:32:51,348 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.3215, 5.0383, 4.6190, 4.8429, 4.8797, 5.1646, 5.2606, 5.2716], device='cuda:3'), covar=tensor([0.0632, 0.0384, 0.1923, 0.2606, 0.0662, 0.0755, 0.0725, 0.0716], device='cuda:3'), in_proj_covar=tensor([0.0377, 0.0263, 0.0435, 0.0558, 0.0327, 0.0428, 0.0386, 0.0368], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 02:32:57,689 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=88995.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 02:33:26,331 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89027.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:33:46,166 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1359, 2.0356, 1.7983, 1.8603, 2.0575, 2.1117, 2.0464, 2.0401], device='cuda:3'), covar=tensor([0.1179, 0.0968, 0.2857, 0.2722, 0.1341, 0.1214, 0.1836, 0.1314], device='cuda:3'), in_proj_covar=tensor([0.0377, 0.0263, 0.0435, 0.0556, 0.0326, 0.0430, 0.0388, 0.0368], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 02:34:01,515 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.293e+02 2.084e+02 2.490e+02 3.041e+02 5.969e+02, threshold=4.980e+02, percent-clipped=1.0 2022-12-08 02:34:03,373 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3904, 4.4635, 4.7573, 3.8903, 4.6159, 4.7583, 1.8646, 4.2908], device='cuda:3'), covar=tensor([0.0253, 0.0249, 0.0303, 0.0404, 0.0245, 0.0221, 0.2876, 0.0246], device='cuda:3'), in_proj_covar=tensor([0.0158, 0.0167, 0.0138, 0.0136, 0.0196, 0.0134, 0.0155, 0.0184], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 02:34:15,419 INFO [train.py:873] (3/4) Epoch 12, batch 5900, loss[loss=0.1262, simple_loss=0.1528, pruned_loss=0.04978, over 11968.00 frames. ], tot_loss[loss=0.1245, simple_loss=0.1552, pruned_loss=0.04686, over 1957870.25 frames. ], batch size: 100, lr: 6.52e-03, grad_scale: 8.0 2022-12-08 02:34:20,382 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89088.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:34:47,392 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2022-12-08 02:35:29,920 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.366e+02 2.297e+02 2.664e+02 3.393e+02 7.245e+02, threshold=5.328e+02, percent-clipped=3.0 2022-12-08 02:35:31,833 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89169.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:35:44,587 INFO [train.py:873] (3/4) Epoch 12, batch 6000, loss[loss=0.09714, simple_loss=0.1141, pruned_loss=0.04008, over 2650.00 frames. ], tot_loss[loss=0.1243, simple_loss=0.1552, pruned_loss=0.04671, over 1938850.90 frames. ], batch size: 100, lr: 6.51e-03, grad_scale: 8.0 2022-12-08 02:35:44,588 INFO [train.py:896] (3/4) Computing validation loss 2022-12-08 02:35:52,918 INFO [train.py:905] (3/4) Epoch 12, validation: loss=0.1296, simple_loss=0.1695, pruned_loss=0.04492, over 857387.00 frames. 2022-12-08 02:35:52,919 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-08 02:36:23,049 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.16 vs. limit=2.0 2022-12-08 02:36:34,401 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89230.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:37:06,542 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.059e+02 2.275e+02 2.670e+02 3.684e+02 9.566e+02, threshold=5.341e+02, percent-clipped=7.0 2022-12-08 02:37:21,419 INFO [train.py:873] (3/4) Epoch 12, batch 6100, loss[loss=0.1227, simple_loss=0.1618, pruned_loss=0.04186, over 14234.00 frames. ], tot_loss[loss=0.1247, simple_loss=0.1552, pruned_loss=0.04708, over 1909905.42 frames. ], batch size: 37, lr: 6.51e-03, grad_scale: 8.0 2022-12-08 02:37:21,540 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.2137, 5.1346, 4.1791, 4.5538, 4.6020, 5.1509, 5.3663, 5.1719], device='cuda:3'), covar=tensor([0.1325, 0.0504, 0.3238, 0.3668, 0.1264, 0.1140, 0.0917, 0.1311], device='cuda:3'), in_proj_covar=tensor([0.0379, 0.0263, 0.0438, 0.0555, 0.0327, 0.0431, 0.0389, 0.0366], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 02:37:46,345 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.65 vs. limit=2.0 2022-12-08 02:38:21,540 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4214, 2.2097, 2.7545, 1.6344, 1.9059, 2.4662, 1.4548, 2.3136], device='cuda:3'), covar=tensor([0.0927, 0.1368, 0.0759, 0.2616, 0.2523, 0.0918, 0.3641, 0.1121], device='cuda:3'), in_proj_covar=tensor([0.0081, 0.0096, 0.0090, 0.0096, 0.0115, 0.0083, 0.0121, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 02:38:22,682 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 2022-12-08 02:38:26,639 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2022-12-08 02:38:29,409 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.4248, 5.1535, 5.0375, 5.4527, 5.0212, 4.5464, 5.5044, 5.3181], device='cuda:3'), covar=tensor([0.0540, 0.0591, 0.0618, 0.0432, 0.0493, 0.0469, 0.0492, 0.0547], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0130, 0.0138, 0.0148, 0.0137, 0.0115, 0.0158, 0.0137], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 02:38:34,355 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.186e+02 2.240e+02 2.644e+02 3.331e+02 6.090e+02, threshold=5.289e+02, percent-clipped=8.0 2022-12-08 02:38:40,687 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89374.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:38:48,057 INFO [train.py:873] (3/4) Epoch 12, batch 6200, loss[loss=0.164, simple_loss=0.1773, pruned_loss=0.07539, over 8651.00 frames. ], tot_loss[loss=0.1257, simple_loss=0.1559, pruned_loss=0.04778, over 1964536.74 frames. ], batch size: 100, lr: 6.51e-03, grad_scale: 8.0 2022-12-08 02:38:48,131 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89383.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:39:17,773 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8586, 1.5874, 1.8319, 2.0524, 1.3852, 1.7302, 1.6929, 1.9393], device='cuda:3'), covar=tensor([0.0153, 0.0265, 0.0153, 0.0120, 0.0247, 0.0271, 0.0178, 0.0128], device='cuda:3'), in_proj_covar=tensor([0.0287, 0.0254, 0.0369, 0.0323, 0.0264, 0.0300, 0.0303, 0.0280], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 02:39:24,615 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89425.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:39:33,920 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89435.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:39:37,327 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89439.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 02:40:00,790 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.227e+02 2.324e+02 2.890e+02 3.661e+02 8.443e+02, threshold=5.779e+02, percent-clipped=4.0 2022-12-08 02:40:15,102 INFO [train.py:873] (3/4) Epoch 12, batch 6300, loss[loss=0.1136, simple_loss=0.1526, pruned_loss=0.0373, over 14332.00 frames. ], tot_loss[loss=0.1248, simple_loss=0.1554, pruned_loss=0.04713, over 1961782.34 frames. ], batch size: 66, lr: 6.50e-03, grad_scale: 8.0 2022-12-08 02:40:17,852 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89486.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:40:29,994 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89500.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 02:40:36,746 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5456, 2.5722, 2.7519, 2.7596, 2.7049, 2.4990, 1.3736, 2.4258], device='cuda:3'), covar=tensor([0.0518, 0.0540, 0.0458, 0.0372, 0.0404, 0.0886, 0.2520, 0.0375], device='cuda:3'), in_proj_covar=tensor([0.0162, 0.0169, 0.0141, 0.0139, 0.0199, 0.0136, 0.0157, 0.0187], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 02:40:37,605 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89509.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:40:51,632 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89525.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:41:27,938 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.271e+01 2.078e+02 2.565e+02 3.125e+02 6.537e+02, threshold=5.131e+02, percent-clipped=2.0 2022-12-08 02:41:30,693 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89570.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:41:41,867 INFO [train.py:873] (3/4) Epoch 12, batch 6400, loss[loss=0.125, simple_loss=0.1606, pruned_loss=0.04471, over 14254.00 frames. ], tot_loss[loss=0.1241, simple_loss=0.1553, pruned_loss=0.04647, over 2004440.47 frames. ], batch size: 80, lr: 6.50e-03, grad_scale: 8.0 2022-12-08 02:41:43,108 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.61 vs. limit=2.0 2022-12-08 02:41:44,650 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0014, 2.4150, 3.9890, 4.1190, 4.0515, 2.3961, 4.0046, 3.2127], device='cuda:3'), covar=tensor([0.0348, 0.0933, 0.0721, 0.0348, 0.0314, 0.1412, 0.0373, 0.0794], device='cuda:3'), in_proj_covar=tensor([0.0287, 0.0253, 0.0368, 0.0322, 0.0263, 0.0300, 0.0302, 0.0280], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 02:42:13,131 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3829, 3.9113, 2.9909, 4.7434, 4.0526, 4.4393, 3.8723, 3.3586], device='cuda:3'), covar=tensor([0.0765, 0.1257, 0.3926, 0.0438, 0.1236, 0.1481, 0.1175, 0.3017], device='cuda:3'), in_proj_covar=tensor([0.0266, 0.0298, 0.0267, 0.0262, 0.0311, 0.0294, 0.0256, 0.0249], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 02:42:13,866 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.5226, 1.0142, 2.0825, 1.8455, 1.8092, 2.0839, 1.3452, 2.0701], device='cuda:3'), covar=tensor([0.0773, 0.1371, 0.0179, 0.0447, 0.0515, 0.0219, 0.0660, 0.0217], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0157, 0.0126, 0.0163, 0.0143, 0.0139, 0.0119, 0.0119], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 02:42:50,345 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.5326, 1.4537, 1.5161, 1.5140, 1.6739, 0.9640, 1.4350, 1.4316], device='cuda:3'), covar=tensor([0.0741, 0.1058, 0.0689, 0.0761, 0.0450, 0.0864, 0.0608, 0.0617], device='cuda:3'), in_proj_covar=tensor([0.0028, 0.0028, 0.0030, 0.0027, 0.0029, 0.0041, 0.0029, 0.0031], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 02:42:56,147 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.041e+02 2.320e+02 2.969e+02 3.886e+02 7.379e+02, threshold=5.938e+02, percent-clipped=5.0 2022-12-08 02:43:07,230 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89680.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:43:09,690 INFO [train.py:873] (3/4) Epoch 12, batch 6500, loss[loss=0.1085, simple_loss=0.1445, pruned_loss=0.03623, over 14514.00 frames. ], tot_loss[loss=0.1245, simple_loss=0.1553, pruned_loss=0.04687, over 1954176.79 frames. ], batch size: 43, lr: 6.50e-03, grad_scale: 8.0 2022-12-08 02:43:09,881 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=89683.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:43:50,408 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89730.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:43:51,220 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=89731.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:44:00,249 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89741.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:44:23,988 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.437e+02 2.343e+02 2.899e+02 3.659e+02 1.529e+03, threshold=5.799e+02, percent-clipped=6.0 2022-12-08 02:44:35,268 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89781.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:44:36,901 INFO [train.py:873] (3/4) Epoch 12, batch 6600, loss[loss=0.1335, simple_loss=0.1616, pruned_loss=0.05271, over 11167.00 frames. ], tot_loss[loss=0.1245, simple_loss=0.155, pruned_loss=0.04702, over 1991225.65 frames. ], batch size: 100, lr: 6.49e-03, grad_scale: 8.0 2022-12-08 02:44:47,392 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89795.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 02:45:13,725 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=89825.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:45:21,593 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 2022-12-08 02:45:48,207 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89865.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:45:48,310 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89865.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:45:50,624 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.462e+01 2.267e+02 2.755e+02 3.423e+02 7.380e+02, threshold=5.509e+02, percent-clipped=4.0 2022-12-08 02:45:53,189 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.71 vs. limit=2.0 2022-12-08 02:45:55,146 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=89873.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:46:04,380 INFO [train.py:873] (3/4) Epoch 12, batch 6700, loss[loss=0.1132, simple_loss=0.1476, pruned_loss=0.03946, over 14234.00 frames. ], tot_loss[loss=0.125, simple_loss=0.1553, pruned_loss=0.04731, over 1974031.00 frames. ], batch size: 69, lr: 6.49e-03, grad_scale: 8.0 2022-12-08 02:46:41,440 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89926.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 02:47:17,826 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.103e+02 2.198e+02 2.718e+02 3.192e+02 1.007e+03, threshold=5.437e+02, percent-clipped=1.0 2022-12-08 02:47:30,901 INFO [train.py:873] (3/4) Epoch 12, batch 6800, loss[loss=0.1025, simple_loss=0.1434, pruned_loss=0.03077, over 14032.00 frames. ], tot_loss[loss=0.1243, simple_loss=0.1555, pruned_loss=0.04661, over 2008549.61 frames. ], batch size: 26, lr: 6.48e-03, grad_scale: 8.0 2022-12-08 02:48:13,591 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0536, 4.1433, 4.4359, 3.6792, 4.2349, 4.3965, 1.6678, 3.9722], device='cuda:3'), covar=tensor([0.0272, 0.0289, 0.0279, 0.0489, 0.0268, 0.0204, 0.2892, 0.0243], device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0171, 0.0142, 0.0141, 0.0202, 0.0136, 0.0157, 0.0188], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 02:48:15,277 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90030.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:48:20,879 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90036.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:48:47,746 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.338e+02 2.301e+02 2.958e+02 4.111e+02 8.301e+02, threshold=5.916e+02, percent-clipped=6.0 2022-12-08 02:48:57,172 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90078.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:48:59,897 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90081.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:49:01,448 INFO [train.py:873] (3/4) Epoch 12, batch 6900, loss[loss=0.1384, simple_loss=0.1373, pruned_loss=0.06981, over 1160.00 frames. ], tot_loss[loss=0.1243, simple_loss=0.1553, pruned_loss=0.0466, over 2011812.88 frames. ], batch size: 100, lr: 6.48e-03, grad_scale: 8.0 2022-12-08 02:49:11,909 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90095.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 02:49:34,454 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2022-12-08 02:49:41,790 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90129.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:49:42,831 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90130.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:49:53,770 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90143.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 02:50:13,234 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90165.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:50:14,010 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9646, 4.5518, 4.4353, 4.9889, 4.6195, 4.2533, 4.9118, 4.1376], device='cuda:3'), covar=tensor([0.0313, 0.0912, 0.0351, 0.0330, 0.0699, 0.0569, 0.0480, 0.0520], device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0257, 0.0185, 0.0179, 0.0173, 0.0146, 0.0265, 0.0160], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 02:50:15,737 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.569e+02 2.354e+02 2.821e+02 3.688e+02 2.263e+03, threshold=5.641e+02, percent-clipped=6.0 2022-12-08 02:50:28,849 INFO [train.py:873] (3/4) Epoch 12, batch 7000, loss[loss=0.1402, simple_loss=0.1644, pruned_loss=0.05801, over 12753.00 frames. ], tot_loss[loss=0.1239, simple_loss=0.155, pruned_loss=0.04642, over 1998814.44 frames. ], batch size: 100, lr: 6.48e-03, grad_scale: 8.0 2022-12-08 02:50:35,967 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90191.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:50:55,451 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90213.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:51:02,583 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90221.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 02:51:12,271 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90232.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:51:44,201 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.424e+02 1.997e+02 2.595e+02 3.353e+02 6.185e+02, threshold=5.189e+02, percent-clipped=2.0 2022-12-08 02:51:57,580 INFO [train.py:873] (3/4) Epoch 12, batch 7100, loss[loss=0.177, simple_loss=0.1561, pruned_loss=0.09898, over 1267.00 frames. ], tot_loss[loss=0.1251, simple_loss=0.1558, pruned_loss=0.04717, over 2029098.46 frames. ], batch size: 100, lr: 6.47e-03, grad_scale: 8.0 2022-12-08 02:52:02,294 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2022-12-08 02:52:06,131 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90293.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:52:44,215 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90336.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:52:57,305 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90351.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:53:12,653 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.266e+02 2.346e+02 2.734e+02 3.324e+02 7.118e+02, threshold=5.469e+02, percent-clipped=5.0 2022-12-08 02:53:14,622 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6548, 2.4602, 3.4518, 2.6257, 3.3674, 3.4010, 3.2893, 2.7884], device='cuda:3'), covar=tensor([0.0891, 0.2671, 0.1042, 0.2020, 0.0845, 0.0948, 0.1314, 0.1902], device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0315, 0.0400, 0.0303, 0.0376, 0.0321, 0.0365, 0.0310], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 02:53:26,145 INFO [train.py:873] (3/4) Epoch 12, batch 7200, loss[loss=0.1373, simple_loss=0.1615, pruned_loss=0.05652, over 8603.00 frames. ], tot_loss[loss=0.1245, simple_loss=0.1553, pruned_loss=0.04689, over 2025635.85 frames. ], batch size: 100, lr: 6.47e-03, grad_scale: 8.0 2022-12-08 02:53:26,994 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90384.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:53:51,996 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90412.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:54:15,914 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5082, 1.8679, 2.4067, 2.1078, 2.4344, 2.3423, 2.2599, 2.1716], device='cuda:3'), covar=tensor([0.0620, 0.1955, 0.0744, 0.1403, 0.0480, 0.0949, 0.0790, 0.1272], device='cuda:3'), in_proj_covar=tensor([0.0355, 0.0315, 0.0403, 0.0306, 0.0378, 0.0323, 0.0366, 0.0311], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 02:54:32,202 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4834, 2.0827, 2.3295, 1.3379, 2.1618, 2.4248, 2.5026, 2.0103], device='cuda:3'), covar=tensor([0.0798, 0.0889, 0.1060, 0.2032, 0.1206, 0.0732, 0.0779, 0.1668], device='cuda:3'), in_proj_covar=tensor([0.0138, 0.0176, 0.0135, 0.0122, 0.0133, 0.0144, 0.0122, 0.0137], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0005, 0.0006], device='cuda:3') 2022-12-08 02:54:32,971 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90459.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:54:35,255 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.98 vs. limit=2.0 2022-12-08 02:54:40,950 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.493e+02 2.297e+02 2.801e+02 3.599e+02 6.534e+02, threshold=5.602e+02, percent-clipped=3.0 2022-12-08 02:54:54,143 INFO [train.py:873] (3/4) Epoch 12, batch 7300, loss[loss=0.1259, simple_loss=0.1577, pruned_loss=0.04701, over 14377.00 frames. ], tot_loss[loss=0.1252, simple_loss=0.1556, pruned_loss=0.04736, over 2009061.78 frames. ], batch size: 73, lr: 6.47e-03, grad_scale: 8.0 2022-12-08 02:54:56,674 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90486.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:55:02,627 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7788, 1.3556, 2.5280, 2.2590, 2.4235, 2.5389, 1.6227, 2.5126], device='cuda:3'), covar=tensor([0.0857, 0.1172, 0.0176, 0.0409, 0.0416, 0.0193, 0.0646, 0.0243], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0157, 0.0127, 0.0165, 0.0143, 0.0139, 0.0119, 0.0119], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 02:55:05,378 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90496.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:55:26,469 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90520.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:55:27,611 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90521.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 02:55:50,766 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8890, 1.6789, 1.9798, 1.6972, 1.9771, 1.7996, 1.6973, 1.9025], device='cuda:3'), covar=tensor([0.0627, 0.1427, 0.0347, 0.0441, 0.0400, 0.0919, 0.0327, 0.0454], device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0313, 0.0398, 0.0303, 0.0375, 0.0319, 0.0362, 0.0307], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 02:55:58,939 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90557.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:56:08,321 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.346e+02 2.439e+02 2.820e+02 3.761e+02 1.640e+03, threshold=5.641e+02, percent-clipped=6.0 2022-12-08 02:56:09,303 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90569.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:56:21,035 INFO [train.py:873] (3/4) Epoch 12, batch 7400, loss[loss=0.1384, simple_loss=0.1694, pruned_loss=0.05374, over 10360.00 frames. ], tot_loss[loss=0.1253, simple_loss=0.1554, pruned_loss=0.04764, over 1982938.75 frames. ], batch size: 100, lr: 6.46e-03, grad_scale: 8.0 2022-12-08 02:56:25,652 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90588.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:57:06,763 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4117, 3.2595, 3.1918, 3.5114, 3.1143, 2.9172, 3.4681, 3.3829], device='cuda:3'), covar=tensor([0.0693, 0.0973, 0.0884, 0.0601, 0.1033, 0.0757, 0.0717, 0.0752], device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0132, 0.0140, 0.0149, 0.0140, 0.0117, 0.0159, 0.0141], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 02:57:36,485 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.166e+02 2.056e+02 2.705e+02 3.388e+02 5.902e+02, threshold=5.409e+02, percent-clipped=1.0 2022-12-08 02:57:50,345 INFO [train.py:873] (3/4) Epoch 12, batch 7500, loss[loss=0.1313, simple_loss=0.1616, pruned_loss=0.0505, over 14248.00 frames. ], tot_loss[loss=0.1238, simple_loss=0.1545, pruned_loss=0.04658, over 2002623.11 frames. ], batch size: 57, lr: 6.46e-03, grad_scale: 8.0 2022-12-08 02:58:00,647 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1519, 1.8154, 4.9962, 4.4565, 4.3231, 5.1476, 4.7941, 5.1130], device='cuda:3'), covar=tensor([0.1305, 0.1411, 0.0079, 0.0166, 0.0193, 0.0088, 0.0100, 0.0097], device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0157, 0.0126, 0.0165, 0.0143, 0.0140, 0.0120, 0.0119], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 02:58:10,893 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90707.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 02:59:13,082 INFO [train.py:873] (3/4) Epoch 13, batch 0, loss[loss=0.1663, simple_loss=0.1927, pruned_loss=0.07001, over 14228.00 frames. ], tot_loss[loss=0.1663, simple_loss=0.1927, pruned_loss=0.07001, over 14228.00 frames. ], batch size: 37, lr: 6.21e-03, grad_scale: 8.0 2022-12-08 02:59:13,082 INFO [train.py:896] (3/4) Computing validation loss 2022-12-08 02:59:20,362 INFO [train.py:905] (3/4) Epoch 13, validation: loss=0.1364, simple_loss=0.1777, pruned_loss=0.04756, over 857387.00 frames. 2022-12-08 02:59:20,363 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-08 02:59:41,454 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 6.584e+01 1.659e+02 2.772e+02 3.641e+02 1.065e+03, threshold=5.544e+02, percent-clipped=8.0 2022-12-08 02:59:57,667 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90786.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:00:16,899 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3086, 3.1372, 2.8552, 3.0376, 3.2649, 3.2590, 3.3066, 3.2800], device='cuda:3'), covar=tensor([0.0911, 0.0642, 0.2106, 0.2624, 0.0685, 0.0906, 0.1133, 0.0876], device='cuda:3'), in_proj_covar=tensor([0.0374, 0.0261, 0.0435, 0.0549, 0.0325, 0.0425, 0.0387, 0.0363], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 03:00:24,311 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90815.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:00:40,603 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90834.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:00:44,013 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0935, 2.1147, 2.2479, 2.2885, 1.9812, 2.3391, 2.1945, 1.3677], device='cuda:3'), covar=tensor([0.1237, 0.1304, 0.0919, 0.0754, 0.1103, 0.0625, 0.1203, 0.2158], device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0079, 0.0063, 0.0066, 0.0092, 0.0077, 0.0092, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0006, 0.0006], device='cuda:3') 2022-12-08 03:00:50,553 INFO [train.py:873] (3/4) Epoch 13, batch 100, loss[loss=0.1024, simple_loss=0.1487, pruned_loss=0.02805, over 13941.00 frames. ], tot_loss[loss=0.1227, simple_loss=0.1548, pruned_loss=0.0453, over 884531.26 frames. ], batch size: 23, lr: 6.20e-03, grad_scale: 8.0 2022-12-08 03:00:54,339 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90849.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:00:56,708 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90852.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:01:09,979 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.361e+02 2.335e+02 2.702e+02 3.282e+02 9.825e+02, threshold=5.404e+02, percent-clipped=3.0 2022-12-08 03:01:27,453 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90888.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:01:46,461 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90910.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:02:07,720 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7679, 2.4662, 3.5189, 2.8005, 3.6118, 3.5206, 3.4628, 3.0473], device='cuda:3'), covar=tensor([0.0703, 0.2688, 0.0956, 0.1836, 0.0842, 0.0823, 0.1380, 0.1770], device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0315, 0.0399, 0.0303, 0.0377, 0.0320, 0.0363, 0.0307], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 03:02:09,087 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90936.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:02:16,741 INFO [train.py:873] (3/4) Epoch 13, batch 200, loss[loss=0.1269, simple_loss=0.159, pruned_loss=0.0474, over 14277.00 frames. ], tot_loss[loss=0.1211, simple_loss=0.1536, pruned_loss=0.04431, over 1356696.90 frames. ], batch size: 28, lr: 6.20e-03, grad_scale: 8.0 2022-12-08 03:02:22,873 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8341, 3.8911, 4.1243, 3.5549, 3.9759, 4.0060, 1.6285, 3.7463], device='cuda:3'), covar=tensor([0.0325, 0.0342, 0.0332, 0.0640, 0.0320, 0.0356, 0.3152, 0.0277], device='cuda:3'), in_proj_covar=tensor([0.0160, 0.0168, 0.0140, 0.0137, 0.0198, 0.0133, 0.0154, 0.0185], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 03:02:36,077 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.982e+01 2.220e+02 2.675e+02 3.390e+02 6.533e+02, threshold=5.349e+02, percent-clipped=6.0 2022-12-08 03:02:54,049 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.0644, 1.3101, 1.3497, 0.9972, 0.8542, 1.1282, 0.8020, 1.1875], device='cuda:3'), covar=tensor([0.2078, 0.2793, 0.0939, 0.2482, 0.3066, 0.1031, 0.2445, 0.1196], device='cuda:3'), in_proj_covar=tensor([0.0082, 0.0098, 0.0090, 0.0097, 0.0114, 0.0084, 0.0123, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 03:03:09,677 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9490, 1.7540, 2.1864, 1.6770, 2.1718, 1.6717, 1.5886, 1.2396], device='cuda:3'), covar=tensor([0.0190, 0.0551, 0.0292, 0.0457, 0.0148, 0.0275, 0.0250, 0.0564], device='cuda:3'), in_proj_covar=tensor([0.0017, 0.0018, 0.0015, 0.0016, 0.0016, 0.0027, 0.0022, 0.0027], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 03:03:11,362 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=91007.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:03:44,491 INFO [train.py:873] (3/4) Epoch 13, batch 300, loss[loss=0.1057, simple_loss=0.1432, pruned_loss=0.03408, over 13919.00 frames. ], tot_loss[loss=0.1227, simple_loss=0.154, pruned_loss=0.04566, over 1584422.96 frames. ], batch size: 19, lr: 6.20e-03, grad_scale: 8.0 2022-12-08 03:03:53,666 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=91055.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:04:04,582 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.277e+02 2.005e+02 2.425e+02 3.069e+02 6.147e+02, threshold=4.849e+02, percent-clipped=1.0 2022-12-08 03:04:46,144 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=91115.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:04:50,700 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.57 vs. limit=5.0 2022-12-08 03:04:52,982 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.74 vs. limit=2.0 2022-12-08 03:05:05,350 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9092, 4.7209, 4.5079, 4.9811, 4.5448, 3.9790, 4.9686, 4.7463], device='cuda:3'), covar=tensor([0.0663, 0.0780, 0.0851, 0.0649, 0.0780, 0.0661, 0.0603, 0.0845], device='cuda:3'), in_proj_covar=tensor([0.0138, 0.0133, 0.0143, 0.0152, 0.0142, 0.0118, 0.0161, 0.0142], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 03:05:13,070 INFO [train.py:873] (3/4) Epoch 13, batch 400, loss[loss=0.107, simple_loss=0.1536, pruned_loss=0.03019, over 14250.00 frames. ], tot_loss[loss=0.1215, simple_loss=0.1534, pruned_loss=0.04479, over 1810385.33 frames. ], batch size: 37, lr: 6.19e-03, grad_scale: 8.0 2022-12-08 03:05:16,670 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.3933, 0.9114, 1.2658, 0.8669, 1.0585, 1.3499, 1.1628, 1.1278], device='cuda:3'), covar=tensor([0.0442, 0.1036, 0.0628, 0.0603, 0.1151, 0.0993, 0.0426, 0.1258], device='cuda:3'), in_proj_covar=tensor([0.0140, 0.0178, 0.0138, 0.0125, 0.0135, 0.0146, 0.0124, 0.0139], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0005, 0.0006], device='cuda:3') 2022-12-08 03:05:19,301 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=91152.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:05:28,729 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=91163.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:05:33,349 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.258e+02 2.284e+02 2.800e+02 3.397e+02 6.782e+02, threshold=5.601e+02, percent-clipped=4.0 2022-12-08 03:06:01,552 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=91200.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:06:06,362 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91205.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:06:30,891 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1465, 3.8733, 4.2749, 3.4363, 3.0322, 3.9549, 2.0807, 3.7575], device='cuda:3'), covar=tensor([0.1829, 0.0774, 0.0798, 0.1608, 0.1870, 0.0776, 0.3389, 0.1337], device='cuda:3'), in_proj_covar=tensor([0.0081, 0.0098, 0.0090, 0.0097, 0.0114, 0.0085, 0.0122, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 03:06:38,883 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8283, 2.0930, 2.1014, 2.1768, 1.9586, 2.2293, 1.9814, 1.3412], device='cuda:3'), covar=tensor([0.1369, 0.0847, 0.0860, 0.0703, 0.1063, 0.0546, 0.1137, 0.2235], device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0079, 0.0063, 0.0066, 0.0093, 0.0077, 0.0092, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0006, 0.0006], device='cuda:3') 2022-12-08 03:06:41,227 INFO [train.py:873] (3/4) Epoch 13, batch 500, loss[loss=0.1515, simple_loss=0.1386, pruned_loss=0.08218, over 1199.00 frames. ], tot_loss[loss=0.1221, simple_loss=0.1535, pruned_loss=0.0453, over 1887744.31 frames. ], batch size: 100, lr: 6.19e-03, grad_scale: 8.0 2022-12-08 03:07:01,832 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.375e+02 2.197e+02 2.790e+02 3.555e+02 6.703e+02, threshold=5.580e+02, percent-clipped=3.0 2022-12-08 03:07:16,623 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91285.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:07:51,099 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=7.64 vs. limit=5.0 2022-12-08 03:08:09,277 INFO [train.py:873] (3/4) Epoch 13, batch 600, loss[loss=0.0961, simple_loss=0.1383, pruned_loss=0.02695, over 11978.00 frames. ], tot_loss[loss=0.1228, simple_loss=0.154, pruned_loss=0.04585, over 1893520.95 frames. ], batch size: 100, lr: 6.19e-03, grad_scale: 8.0 2022-12-08 03:08:10,242 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=91346.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:08:29,164 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.092e+02 2.257e+02 2.792e+02 3.414e+02 7.089e+02, threshold=5.584e+02, percent-clipped=8.0 2022-12-08 03:08:44,476 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.69 vs. limit=2.0 2022-12-08 03:09:01,831 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8490, 1.3152, 2.0222, 1.2865, 2.0422, 2.0759, 1.7016, 2.1379], device='cuda:3'), covar=tensor([0.0292, 0.2027, 0.0489, 0.1793, 0.0450, 0.0485, 0.0962, 0.0313], device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0155, 0.0157, 0.0166, 0.0167, 0.0175, 0.0132, 0.0144], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-08 03:09:06,001 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.71 vs. limit=2.0 2022-12-08 03:09:36,012 INFO [train.py:873] (3/4) Epoch 13, batch 700, loss[loss=0.1, simple_loss=0.1295, pruned_loss=0.03526, over 4943.00 frames. ], tot_loss[loss=0.1235, simple_loss=0.154, pruned_loss=0.04651, over 1886836.82 frames. ], batch size: 100, lr: 6.18e-03, grad_scale: 8.0 2022-12-08 03:09:46,310 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.5193, 1.1566, 2.0606, 1.8482, 1.9024, 2.0675, 1.3692, 2.0479], device='cuda:3'), covar=tensor([0.0723, 0.1188, 0.0195, 0.0425, 0.0449, 0.0246, 0.0652, 0.0241], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0159, 0.0129, 0.0167, 0.0146, 0.0142, 0.0123, 0.0121], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 03:09:55,978 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.741e+01 2.087e+02 2.616e+02 3.337e+02 7.440e+02, threshold=5.233e+02, percent-clipped=3.0 2022-12-08 03:10:28,599 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=91505.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:10:29,742 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.50 vs. limit=5.0 2022-12-08 03:10:53,843 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.3388, 5.1065, 5.0188, 5.2636, 5.0179, 4.6522, 5.3034, 5.1806], device='cuda:3'), covar=tensor([0.0481, 0.0818, 0.0630, 0.0508, 0.0584, 0.0522, 0.0516, 0.0527], device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0132, 0.0142, 0.0152, 0.0141, 0.0118, 0.0161, 0.0141], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 03:11:01,164 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91542.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 03:11:03,621 INFO [train.py:873] (3/4) Epoch 13, batch 800, loss[loss=0.1063, simple_loss=0.144, pruned_loss=0.03428, over 14538.00 frames. ], tot_loss[loss=0.1225, simple_loss=0.1533, pruned_loss=0.04583, over 1948993.04 frames. ], batch size: 34, lr: 6.18e-03, grad_scale: 8.0 2022-12-08 03:11:10,472 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=91553.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:11:19,261 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5741, 1.7635, 4.4551, 2.1566, 4.2977, 4.5272, 4.0464, 5.0170], device='cuda:3'), covar=tensor([0.0199, 0.2849, 0.0341, 0.2006, 0.0312, 0.0314, 0.0378, 0.0129], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0156, 0.0157, 0.0167, 0.0167, 0.0176, 0.0132, 0.0145], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-08 03:11:23,656 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.089e+02 2.194e+02 2.551e+02 3.061e+02 6.051e+02, threshold=5.101e+02, percent-clipped=1.0 2022-12-08 03:11:47,910 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9643, 1.3998, 3.1315, 1.6318, 3.2644, 3.1322, 2.2287, 3.3301], device='cuda:3'), covar=tensor([0.0303, 0.2959, 0.0405, 0.1987, 0.0334, 0.0458, 0.0984, 0.0220], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0157, 0.0158, 0.0167, 0.0168, 0.0177, 0.0133, 0.0145], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 03:11:54,594 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=91603.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 03:12:27,380 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91641.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:12:30,321 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8434, 3.4244, 2.7048, 3.9980, 3.8234, 3.8801, 3.4547, 2.7484], device='cuda:3'), covar=tensor([0.0876, 0.1598, 0.4063, 0.0612, 0.1005, 0.1274, 0.1277, 0.3473], device='cuda:3'), in_proj_covar=tensor([0.0269, 0.0297, 0.0270, 0.0262, 0.0315, 0.0296, 0.0258, 0.0249], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 03:12:30,941 INFO [train.py:873] (3/4) Epoch 13, batch 900, loss[loss=0.1256, simple_loss=0.154, pruned_loss=0.04853, over 14302.00 frames. ], tot_loss[loss=0.1218, simple_loss=0.1526, pruned_loss=0.04551, over 1885200.45 frames. ], batch size: 60, lr: 6.18e-03, grad_scale: 16.0 2022-12-08 03:12:52,063 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.003e+02 2.371e+02 2.875e+02 3.801e+02 8.432e+02, threshold=5.750e+02, percent-clipped=7.0 2022-12-08 03:13:58,930 INFO [train.py:873] (3/4) Epoch 13, batch 1000, loss[loss=0.1084, simple_loss=0.1487, pruned_loss=0.03408, over 11171.00 frames. ], tot_loss[loss=0.1215, simple_loss=0.1526, pruned_loss=0.04522, over 1951140.32 frames. ], batch size: 100, lr: 6.17e-03, grad_scale: 8.0 2022-12-08 03:14:19,310 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.021e+02 2.131e+02 2.597e+02 3.397e+02 6.255e+02, threshold=5.195e+02, percent-clipped=2.0 2022-12-08 03:14:47,465 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1888, 1.9206, 3.3320, 2.3863, 3.1396, 1.8682, 2.4617, 3.2017], device='cuda:3'), covar=tensor([0.0817, 0.4020, 0.0521, 0.4791, 0.0845, 0.3374, 0.1410, 0.0674], device='cuda:3'), in_proj_covar=tensor([0.0252, 0.0215, 0.0209, 0.0289, 0.0230, 0.0215, 0.0214, 0.0214], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 03:15:26,316 INFO [train.py:873] (3/4) Epoch 13, batch 1100, loss[loss=0.1228, simple_loss=0.1538, pruned_loss=0.0459, over 14171.00 frames. ], tot_loss[loss=0.1208, simple_loss=0.1524, pruned_loss=0.04465, over 1992920.92 frames. ], batch size: 99, lr: 6.17e-03, grad_scale: 8.0 2022-12-08 03:15:47,467 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.763e+01 2.120e+02 2.631e+02 3.193e+02 5.598e+02, threshold=5.262e+02, percent-clipped=1.0 2022-12-08 03:15:58,770 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2168, 2.2198, 3.1428, 3.2723, 3.1558, 2.1875, 3.2197, 2.5036], device='cuda:3'), covar=tensor([0.0405, 0.0927, 0.0671, 0.0413, 0.0462, 0.1341, 0.0348, 0.0800], device='cuda:3'), in_proj_covar=tensor([0.0288, 0.0256, 0.0374, 0.0324, 0.0265, 0.0301, 0.0301, 0.0280], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 03:16:07,052 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91891.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:16:13,604 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91898.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 03:16:44,990 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.65 vs. limit=2.0 2022-12-08 03:16:51,302 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=91941.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:16:54,658 INFO [train.py:873] (3/4) Epoch 13, batch 1200, loss[loss=0.1277, simple_loss=0.1639, pruned_loss=0.04578, over 14228.00 frames. ], tot_loss[loss=0.1221, simple_loss=0.1531, pruned_loss=0.04562, over 1897656.75 frames. ], batch size: 46, lr: 6.17e-03, grad_scale: 8.0 2022-12-08 03:17:00,825 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=91952.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:17:03,189 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91955.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:17:08,340 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5411, 4.2567, 4.0180, 4.2074, 4.3436, 4.4456, 4.5204, 4.4720], device='cuda:3'), covar=tensor([0.0722, 0.0493, 0.2218, 0.2596, 0.0696, 0.0706, 0.1048, 0.0859], device='cuda:3'), in_proj_covar=tensor([0.0378, 0.0265, 0.0444, 0.0558, 0.0333, 0.0434, 0.0394, 0.0372], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 03:17:15,484 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.061e+02 2.334e+02 2.878e+02 3.521e+02 1.024e+03, threshold=5.756e+02, percent-clipped=7.0 2022-12-08 03:17:26,282 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91981.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 03:17:33,119 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=91989.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:17:39,889 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7586, 2.8118, 2.7974, 2.9247, 2.7814, 2.7751, 1.3897, 2.5448], device='cuda:3'), covar=tensor([0.0551, 0.0485, 0.0659, 0.0411, 0.0568, 0.0760, 0.3307, 0.0513], device='cuda:3'), in_proj_covar=tensor([0.0161, 0.0168, 0.0142, 0.0139, 0.0199, 0.0134, 0.0157, 0.0186], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 03:17:57,339 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92016.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:18:16,608 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6654, 1.5564, 1.7750, 1.4470, 1.4942, 1.2492, 1.0722, 1.0099], device='cuda:3'), covar=tensor([0.0169, 0.0262, 0.0153, 0.0196, 0.0195, 0.0302, 0.0254, 0.0419], device='cuda:3'), in_proj_covar=tensor([0.0017, 0.0018, 0.0016, 0.0016, 0.0017, 0.0027, 0.0022, 0.0027], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 03:18:19,846 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92042.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 03:18:22,471 INFO [train.py:873] (3/4) Epoch 13, batch 1300, loss[loss=0.1354, simple_loss=0.1663, pruned_loss=0.05223, over 14292.00 frames. ], tot_loss[loss=0.1213, simple_loss=0.1532, pruned_loss=0.04473, over 2000196.17 frames. ], batch size: 76, lr: 6.16e-03, grad_scale: 8.0 2022-12-08 03:18:44,122 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.023e+02 2.102e+02 2.497e+02 3.213e+02 6.107e+02, threshold=4.993e+02, percent-clipped=1.0 2022-12-08 03:19:52,208 INFO [train.py:873] (3/4) Epoch 13, batch 1400, loss[loss=0.1041, simple_loss=0.1485, pruned_loss=0.02989, over 14273.00 frames. ], tot_loss[loss=0.1227, simple_loss=0.1539, pruned_loss=0.04574, over 1977789.09 frames. ], batch size: 63, lr: 6.16e-03, grad_scale: 8.0 2022-12-08 03:20:13,182 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.308e+02 2.362e+02 2.970e+02 3.546e+02 6.469e+02, threshold=5.940e+02, percent-clipped=3.0 2022-12-08 03:20:31,803 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2022-12-08 03:20:39,395 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92198.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 03:21:21,064 INFO [train.py:873] (3/4) Epoch 13, batch 1500, loss[loss=0.1011, simple_loss=0.1447, pruned_loss=0.02879, over 14511.00 frames. ], tot_loss[loss=0.1224, simple_loss=0.1534, pruned_loss=0.04569, over 1948261.66 frames. ], batch size: 49, lr: 6.16e-03, grad_scale: 8.0 2022-12-08 03:21:22,070 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=92246.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 03:21:22,849 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92247.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:21:41,369 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.124e+02 2.104e+02 2.663e+02 3.435e+02 6.153e+02, threshold=5.325e+02, percent-clipped=2.0 2022-12-08 03:22:10,006 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9860, 2.6727, 4.9670, 3.5003, 4.6536, 2.4433, 3.8708, 4.6870], device='cuda:3'), covar=tensor([0.0411, 0.3753, 0.0274, 0.5762, 0.0637, 0.3049, 0.1030, 0.0356], device='cuda:3'), in_proj_covar=tensor([0.0253, 0.0214, 0.0210, 0.0290, 0.0232, 0.0215, 0.0215, 0.0214], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 03:22:18,299 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92311.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:22:40,876 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92337.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 03:22:48,257 INFO [train.py:873] (3/4) Epoch 13, batch 1600, loss[loss=0.1126, simple_loss=0.1296, pruned_loss=0.04781, over 2615.00 frames. ], tot_loss[loss=0.1215, simple_loss=0.153, pruned_loss=0.04501, over 1970597.33 frames. ], batch size: 100, lr: 6.15e-03, grad_scale: 8.0 2022-12-08 03:23:08,502 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.316e+02 2.214e+02 2.735e+02 3.327e+02 6.728e+02, threshold=5.470e+02, percent-clipped=2.0 2022-12-08 03:23:20,962 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.61 vs. limit=2.0 2022-12-08 03:23:42,422 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2022-12-08 03:23:46,407 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.96 vs. limit=5.0 2022-12-08 03:24:12,191 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9783, 1.6042, 2.0233, 1.4107, 1.6569, 2.0730, 1.7596, 1.7238], device='cuda:3'), covar=tensor([0.0824, 0.0712, 0.0807, 0.1294, 0.1435, 0.0792, 0.0699, 0.1738], device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0179, 0.0139, 0.0127, 0.0139, 0.0148, 0.0124, 0.0141], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0005, 0.0006], device='cuda:3') 2022-12-08 03:24:14,891 INFO [train.py:873] (3/4) Epoch 13, batch 1700, loss[loss=0.1077, simple_loss=0.1466, pruned_loss=0.03436, over 14131.00 frames. ], tot_loss[loss=0.1215, simple_loss=0.1529, pruned_loss=0.04505, over 1958302.81 frames. ], batch size: 99, lr: 6.15e-03, grad_scale: 8.0 2022-12-08 03:24:19,566 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.11 vs. limit=2.0 2022-12-08 03:24:36,008 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.170e+02 2.177e+02 2.526e+02 3.126e+02 5.467e+02, threshold=5.052e+02, percent-clipped=0.0 2022-12-08 03:24:56,410 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.7978, 0.7826, 0.7217, 0.8359, 0.8165, 0.2983, 0.7099, 0.8243], device='cuda:3'), covar=tensor([0.0333, 0.0561, 0.0502, 0.0352, 0.0431, 0.0276, 0.0807, 0.0762], device='cuda:3'), in_proj_covar=tensor([0.0028, 0.0029, 0.0030, 0.0027, 0.0029, 0.0041, 0.0029, 0.0031], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 03:25:22,749 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.0030, 1.0689, 1.1678, 0.8938, 0.8990, 0.5915, 0.7343, 0.7804], device='cuda:3'), covar=tensor([0.0181, 0.0170, 0.0134, 0.0189, 0.0185, 0.0432, 0.0245, 0.0348], device='cuda:3'), in_proj_covar=tensor([0.0017, 0.0018, 0.0016, 0.0016, 0.0016, 0.0027, 0.0022, 0.0027], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 03:25:28,808 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0261, 1.9018, 4.3840, 4.0790, 3.9948, 4.4334, 3.9172, 4.4345], device='cuda:3'), covar=tensor([0.1450, 0.1358, 0.0103, 0.0185, 0.0210, 0.0109, 0.0186, 0.0114], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0159, 0.0129, 0.0167, 0.0146, 0.0141, 0.0122, 0.0121], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 03:25:42,444 INFO [train.py:873] (3/4) Epoch 13, batch 1800, loss[loss=0.1233, simple_loss=0.1608, pruned_loss=0.04293, over 14267.00 frames. ], tot_loss[loss=0.121, simple_loss=0.1527, pruned_loss=0.04467, over 1976432.35 frames. ], batch size: 76, lr: 6.15e-03, grad_scale: 8.0 2022-12-08 03:25:44,267 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92547.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:26:02,905 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.175e+02 2.080e+02 2.746e+02 3.507e+02 5.512e+02, threshold=5.493e+02, percent-clipped=3.0 2022-12-08 03:26:25,754 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=92595.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:26:27,388 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9039, 4.0975, 4.0159, 3.5637, 4.0336, 4.3000, 1.7956, 3.8382], device='cuda:3'), covar=tensor([0.0409, 0.0458, 0.0750, 0.0791, 0.0601, 0.0294, 0.3695, 0.0415], device='cuda:3'), in_proj_covar=tensor([0.0160, 0.0167, 0.0141, 0.0139, 0.0200, 0.0133, 0.0155, 0.0185], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 03:26:35,513 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7525, 1.7291, 1.8955, 1.6164, 1.6676, 1.4729, 1.1908, 1.0528], device='cuda:3'), covar=tensor([0.0199, 0.0277, 0.0234, 0.0217, 0.0233, 0.0305, 0.0281, 0.0437], device='cuda:3'), in_proj_covar=tensor([0.0017, 0.0018, 0.0016, 0.0017, 0.0017, 0.0027, 0.0022, 0.0027], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 03:26:39,661 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92611.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:26:52,232 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92625.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:26:57,425 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0907, 2.0229, 2.0508, 1.8489, 2.0024, 1.6597, 1.6011, 1.9516], device='cuda:3'), covar=tensor([0.0412, 0.0661, 0.0499, 0.0536, 0.0508, 0.0636, 0.0556, 0.0782], device='cuda:3'), in_proj_covar=tensor([0.0017, 0.0018, 0.0016, 0.0017, 0.0017, 0.0028, 0.0022, 0.0027], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 03:27:02,576 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92637.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 03:27:09,285 INFO [train.py:873] (3/4) Epoch 13, batch 1900, loss[loss=0.1234, simple_loss=0.1568, pruned_loss=0.04502, over 14282.00 frames. ], tot_loss[loss=0.1213, simple_loss=0.1527, pruned_loss=0.04497, over 2022458.94 frames. ], batch size: 35, lr: 6.14e-03, grad_scale: 4.0 2022-12-08 03:27:21,763 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=92659.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:27:31,531 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.247e+02 2.202e+02 2.634e+02 3.218e+02 6.502e+02, threshold=5.268e+02, percent-clipped=3.0 2022-12-08 03:27:44,644 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=92685.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 03:27:45,522 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92686.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:27:56,642 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3120, 2.9739, 2.8579, 1.8603, 2.8308, 2.9573, 3.3178, 2.5239], device='cuda:3'), covar=tensor([0.0599, 0.0901, 0.0977, 0.1622, 0.0704, 0.0751, 0.0481, 0.1276], device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0178, 0.0138, 0.0126, 0.0137, 0.0147, 0.0124, 0.0140], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0005, 0.0006], device='cuda:3') 2022-12-08 03:28:12,341 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8277, 1.9664, 2.6352, 2.1333, 2.6386, 2.6088, 2.3950, 2.3197], device='cuda:3'), covar=tensor([0.0817, 0.2540, 0.0767, 0.1662, 0.0495, 0.1227, 0.0920, 0.1481], device='cuda:3'), in_proj_covar=tensor([0.0345, 0.0309, 0.0389, 0.0300, 0.0371, 0.0319, 0.0357, 0.0302], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 03:28:15,709 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92720.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:28:34,698 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0927, 1.9291, 3.1680, 2.3797, 3.0107, 1.8557, 2.5022, 3.0318], device='cuda:3'), covar=tensor([0.0870, 0.4119, 0.0564, 0.4934, 0.0901, 0.3193, 0.1213, 0.0709], device='cuda:3'), in_proj_covar=tensor([0.0252, 0.0210, 0.0208, 0.0284, 0.0227, 0.0211, 0.0210, 0.0211], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 03:28:37,931 INFO [train.py:873] (3/4) Epoch 13, batch 2000, loss[loss=0.1488, simple_loss=0.1611, pruned_loss=0.06825, over 7820.00 frames. ], tot_loss[loss=0.122, simple_loss=0.1532, pruned_loss=0.04541, over 2003943.92 frames. ], batch size: 100, lr: 6.14e-03, grad_scale: 8.0 2022-12-08 03:28:56,271 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.6218, 2.3844, 2.9646, 2.1641, 1.9300, 2.5596, 1.3943, 2.6224], device='cuda:3'), covar=tensor([0.1089, 0.1416, 0.1052, 0.1967, 0.2710, 0.1191, 0.4753, 0.1362], device='cuda:3'), in_proj_covar=tensor([0.0084, 0.0098, 0.0091, 0.0099, 0.0115, 0.0086, 0.0123, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 03:28:59,688 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.406e+02 2.366e+02 2.802e+02 3.411e+02 7.982e+02, threshold=5.603e+02, percent-clipped=8.0 2022-12-08 03:29:10,482 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92781.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:29:54,479 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92831.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:30:06,191 INFO [train.py:873] (3/4) Epoch 13, batch 2100, loss[loss=0.1461, simple_loss=0.1393, pruned_loss=0.07648, over 1255.00 frames. ], tot_loss[loss=0.1218, simple_loss=0.1532, pruned_loss=0.04525, over 1989571.22 frames. ], batch size: 100, lr: 6.14e-03, grad_scale: 8.0 2022-12-08 03:30:28,667 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.062e+02 2.147e+02 2.777e+02 3.486e+02 6.981e+02, threshold=5.554e+02, percent-clipped=1.0 2022-12-08 03:30:48,250 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92892.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:30:52,507 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2022-12-08 03:30:54,524 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92899.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:31:22,068 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92931.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:31:22,506 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2022-12-08 03:31:24,301 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2022-12-08 03:31:34,899 INFO [train.py:873] (3/4) Epoch 13, batch 2200, loss[loss=0.1225, simple_loss=0.1289, pruned_loss=0.05809, over 2573.00 frames. ], tot_loss[loss=0.1219, simple_loss=0.153, pruned_loss=0.04545, over 1926685.10 frames. ], batch size: 100, lr: 6.13e-03, grad_scale: 8.0 2022-12-08 03:31:47,503 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92960.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:31:56,269 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.342e+02 2.331e+02 2.968e+02 3.670e+02 7.983e+02, threshold=5.936e+02, percent-clipped=6.0 2022-12-08 03:32:00,944 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92975.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:32:06,341 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92981.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:32:16,154 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92992.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:32:31,719 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93010.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:32:55,188 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93036.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:32:55,208 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93036.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:33:02,673 INFO [train.py:873] (3/4) Epoch 13, batch 2300, loss[loss=0.1641, simple_loss=0.1606, pruned_loss=0.0838, over 1313.00 frames. ], tot_loss[loss=0.1212, simple_loss=0.1525, pruned_loss=0.04497, over 1928248.82 frames. ], batch size: 100, lr: 6.13e-03, grad_scale: 8.0 2022-12-08 03:33:25,584 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.411e+02 2.107e+02 2.508e+02 3.191e+02 7.606e+02, threshold=5.016e+02, percent-clipped=1.0 2022-12-08 03:33:26,663 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93071.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:33:27,140 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.64 vs. limit=5.0 2022-12-08 03:33:30,985 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93076.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:33:49,194 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93097.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 03:33:58,166 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0897, 3.1990, 3.3747, 3.1767, 3.3034, 2.8678, 1.5030, 3.0760], device='cuda:3'), covar=tensor([0.0399, 0.0389, 0.0409, 0.0450, 0.0373, 0.0741, 0.3052, 0.0317], device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0170, 0.0142, 0.0139, 0.0201, 0.0135, 0.0157, 0.0188], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 03:34:04,282 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4314, 1.0323, 1.2788, 0.8177, 1.0967, 1.3910, 1.0774, 1.1746], device='cuda:3'), covar=tensor([0.0450, 0.0772, 0.0647, 0.0514, 0.0867, 0.0694, 0.0428, 0.0998], device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0175, 0.0136, 0.0125, 0.0136, 0.0145, 0.0123, 0.0137], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0005, 0.0006], device='cuda:3') 2022-12-08 03:34:05,033 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93115.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:34:09,184 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4746, 2.1670, 4.4950, 3.0448, 4.2892, 2.0763, 3.2882, 4.2462], device='cuda:3'), covar=tensor([0.0594, 0.4455, 0.0338, 0.5999, 0.0588, 0.3630, 0.1322, 0.0431], device='cuda:3'), in_proj_covar=tensor([0.0251, 0.0210, 0.0209, 0.0286, 0.0228, 0.0212, 0.0212, 0.0212], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 03:34:26,512 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4337, 4.6249, 4.9902, 4.1982, 4.6885, 4.9865, 1.8178, 4.4121], device='cuda:3'), covar=tensor([0.0262, 0.0252, 0.0284, 0.0433, 0.0331, 0.0134, 0.3070, 0.0271], device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0169, 0.0142, 0.0139, 0.0201, 0.0136, 0.0157, 0.0188], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 03:34:28,775 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2022-12-08 03:34:31,778 INFO [train.py:873] (3/4) Epoch 13, batch 2400, loss[loss=0.1054, simple_loss=0.1472, pruned_loss=0.03179, over 14064.00 frames. ], tot_loss[loss=0.1214, simple_loss=0.1528, pruned_loss=0.04497, over 1979887.74 frames. ], batch size: 29, lr: 6.13e-03, grad_scale: 8.0 2022-12-08 03:34:53,123 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.501e+02 2.312e+02 2.703e+02 3.635e+02 6.826e+02, threshold=5.406e+02, percent-clipped=5.0 2022-12-08 03:34:58,990 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93176.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:35:01,216 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.4335, 5.1695, 4.9235, 5.4271, 5.0346, 4.7333, 5.5068, 5.2677], device='cuda:3'), covar=tensor([0.0450, 0.0552, 0.0584, 0.0413, 0.0525, 0.0363, 0.0392, 0.0467], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0128, 0.0137, 0.0148, 0.0137, 0.0115, 0.0155, 0.0135], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 03:35:09,431 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93187.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:35:25,811 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7197, 3.4312, 3.1896, 2.3001, 3.1500, 3.3549, 3.6908, 2.9252], device='cuda:3'), covar=tensor([0.0573, 0.1214, 0.0891, 0.1478, 0.0841, 0.0647, 0.0630, 0.1175], device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0176, 0.0137, 0.0125, 0.0137, 0.0146, 0.0123, 0.0138], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0005, 0.0006], device='cuda:3') 2022-12-08 03:35:59,851 INFO [train.py:873] (3/4) Epoch 13, batch 2500, loss[loss=0.1767, simple_loss=0.1585, pruned_loss=0.09751, over 1266.00 frames. ], tot_loss[loss=0.1211, simple_loss=0.1526, pruned_loss=0.04476, over 2027251.42 frames. ], batch size: 100, lr: 6.12e-03, grad_scale: 8.0 2022-12-08 03:36:08,990 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93255.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:36:22,499 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.150e+02 2.321e+02 2.910e+02 3.520e+02 9.929e+02, threshold=5.819e+02, percent-clipped=5.0 2022-12-08 03:36:27,699 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3315, 1.5053, 2.5014, 1.4972, 2.4599, 2.4934, 1.9670, 2.6214], device='cuda:3'), covar=tensor([0.0263, 0.2135, 0.0349, 0.1695, 0.0435, 0.0487, 0.0985, 0.0254], device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0156, 0.0159, 0.0167, 0.0169, 0.0178, 0.0135, 0.0147], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 03:36:32,037 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93281.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:36:37,167 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93287.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:37:12,132 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2022-12-08 03:37:15,090 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93329.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:37:16,834 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93331.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:37:28,917 INFO [train.py:873] (3/4) Epoch 13, batch 2600, loss[loss=0.1164, simple_loss=0.1197, pruned_loss=0.05654, over 2629.00 frames. ], tot_loss[loss=0.1225, simple_loss=0.1536, pruned_loss=0.04575, over 2028441.77 frames. ], batch size: 100, lr: 6.12e-03, grad_scale: 8.0 2022-12-08 03:37:47,483 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93366.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:37:50,652 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.166e+02 2.180e+02 2.788e+02 3.347e+02 5.990e+02, threshold=5.575e+02, percent-clipped=1.0 2022-12-08 03:37:56,060 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93376.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:38:10,551 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93392.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 03:38:39,214 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93424.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:38:57,480 INFO [train.py:873] (3/4) Epoch 13, batch 2700, loss[loss=0.1114, simple_loss=0.1247, pruned_loss=0.04903, over 2620.00 frames. ], tot_loss[loss=0.1219, simple_loss=0.1534, pruned_loss=0.04517, over 2063088.18 frames. ], batch size: 100, lr: 6.12e-03, grad_scale: 8.0 2022-12-08 03:39:19,854 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.738e+01 2.118e+02 2.554e+02 3.202e+02 5.751e+02, threshold=5.109e+02, percent-clipped=2.0 2022-12-08 03:39:20,919 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93471.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:39:34,781 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93487.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:39:36,410 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3256, 3.2618, 3.1010, 3.4221, 2.9866, 2.9296, 3.4181, 3.3114], device='cuda:3'), covar=tensor([0.0743, 0.0956, 0.0908, 0.0626, 0.1123, 0.0750, 0.0739, 0.0765], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0129, 0.0138, 0.0148, 0.0138, 0.0115, 0.0156, 0.0136], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 03:39:52,126 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.5610, 1.6140, 4.1366, 3.9600, 3.8416, 4.3159, 3.7873, 4.2905], device='cuda:3'), covar=tensor([0.2556, 0.2300, 0.0222, 0.0317, 0.0349, 0.0228, 0.0323, 0.0203], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0160, 0.0128, 0.0167, 0.0146, 0.0142, 0.0122, 0.0121], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 03:40:09,226 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9236, 1.2818, 1.9972, 1.2663, 1.9665, 2.0809, 1.6926, 2.1403], device='cuda:3'), covar=tensor([0.0336, 0.1909, 0.0511, 0.1880, 0.0597, 0.0535, 0.1067, 0.0325], device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0155, 0.0158, 0.0167, 0.0168, 0.0177, 0.0134, 0.0146], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 03:40:16,925 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93535.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:40:25,982 INFO [train.py:873] (3/4) Epoch 13, batch 2800, loss[loss=0.1316, simple_loss=0.1596, pruned_loss=0.05178, over 14241.00 frames. ], tot_loss[loss=0.1228, simple_loss=0.1539, pruned_loss=0.04587, over 2033531.09 frames. ], batch size: 69, lr: 6.11e-03, grad_scale: 8.0 2022-12-08 03:40:33,490 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8377, 2.5220, 2.7192, 1.8220, 2.3981, 2.6798, 2.9309, 2.4172], device='cuda:3'), covar=tensor([0.0789, 0.1091, 0.1008, 0.1723, 0.1137, 0.0748, 0.0558, 0.1348], device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0174, 0.0137, 0.0125, 0.0137, 0.0146, 0.0123, 0.0138], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0005, 0.0006], device='cuda:3') 2022-12-08 03:40:35,137 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93555.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:40:47,337 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.714e+01 2.417e+02 3.010e+02 3.786e+02 1.160e+03, threshold=6.020e+02, percent-clipped=14.0 2022-12-08 03:41:02,910 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93587.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:41:16,934 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93603.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:41:41,561 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93631.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:41:43,274 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0078, 3.7913, 3.4983, 3.6636, 3.8719, 3.9421, 3.9945, 4.0000], device='cuda:3'), covar=tensor([0.0770, 0.0543, 0.2072, 0.2533, 0.0720, 0.0770, 0.0913, 0.0698], device='cuda:3'), in_proj_covar=tensor([0.0376, 0.0263, 0.0444, 0.0562, 0.0330, 0.0433, 0.0388, 0.0371], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 03:41:44,869 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93635.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:41:53,319 INFO [train.py:873] (3/4) Epoch 13, batch 2900, loss[loss=0.1126, simple_loss=0.1492, pruned_loss=0.03802, over 11281.00 frames. ], tot_loss[loss=0.1231, simple_loss=0.1537, pruned_loss=0.0462, over 1938536.81 frames. ], batch size: 14, lr: 6.11e-03, grad_scale: 8.0 2022-12-08 03:42:11,950 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93666.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:42:15,457 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.161e+02 2.101e+02 2.613e+02 3.235e+02 8.582e+02, threshold=5.226e+02, percent-clipped=1.0 2022-12-08 03:42:23,199 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93679.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:42:34,298 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93692.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:42:53,838 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93714.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:43:16,475 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93740.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:43:21,029 INFO [train.py:873] (3/4) Epoch 13, batch 3000, loss[loss=0.1217, simple_loss=0.1603, pruned_loss=0.04154, over 14144.00 frames. ], tot_loss[loss=0.1232, simple_loss=0.1539, pruned_loss=0.04624, over 1939546.04 frames. ], batch size: 29, lr: 6.11e-03, grad_scale: 8.0 2022-12-08 03:43:21,029 INFO [train.py:896] (3/4) Computing validation loss 2022-12-08 03:43:29,438 INFO [train.py:905] (3/4) Epoch 13, validation: loss=0.1304, simple_loss=0.1697, pruned_loss=0.04555, over 857387.00 frames. 2022-12-08 03:43:29,439 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-08 03:43:52,064 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.563e+02 2.240e+02 2.864e+02 3.755e+02 7.950e+02, threshold=5.728e+02, percent-clipped=3.0 2022-12-08 03:43:52,313 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93770.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:43:53,107 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93771.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:44:16,620 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.2162, 1.4978, 1.7219, 1.6059, 1.5928, 1.6696, 1.3074, 1.2821], device='cuda:3'), covar=tensor([0.1704, 0.1535, 0.0619, 0.0476, 0.1365, 0.0841, 0.1832, 0.2032], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0081, 0.0065, 0.0068, 0.0095, 0.0080, 0.0095, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005, 0.0006, 0.0006], device='cuda:3') 2022-12-08 03:44:35,612 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93819.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:44:46,123 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93831.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:44:58,706 INFO [train.py:873] (3/4) Epoch 13, batch 3100, loss[loss=0.1114, simple_loss=0.1501, pruned_loss=0.03639, over 14389.00 frames. ], tot_loss[loss=0.1218, simple_loss=0.1532, pruned_loss=0.04527, over 1917427.37 frames. ], batch size: 73, lr: 6.10e-03, grad_scale: 8.0 2022-12-08 03:45:03,874 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4808, 4.1918, 4.0726, 4.4745, 4.2149, 3.9126, 4.5225, 3.7743], device='cuda:3'), covar=tensor([0.0442, 0.0935, 0.0437, 0.0449, 0.0785, 0.1077, 0.0496, 0.0527], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0270, 0.0193, 0.0190, 0.0184, 0.0153, 0.0276, 0.0168], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 03:45:13,319 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1569, 2.5483, 3.9261, 2.9361, 3.8895, 3.8241, 3.6561, 3.2126], device='cuda:3'), covar=tensor([0.0732, 0.3184, 0.1207, 0.1988, 0.1092, 0.0925, 0.1842, 0.1849], device='cuda:3'), in_proj_covar=tensor([0.0344, 0.0310, 0.0394, 0.0298, 0.0372, 0.0315, 0.0355, 0.0304], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 03:45:19,448 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.360e+02 2.395e+02 2.887e+02 3.517e+02 6.635e+02, threshold=5.774e+02, percent-clipped=1.0 2022-12-08 03:46:12,848 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.10 vs. limit=2.0 2022-12-08 03:46:25,613 INFO [train.py:873] (3/4) Epoch 13, batch 3200, loss[loss=0.1044, simple_loss=0.1465, pruned_loss=0.03119, over 14597.00 frames. ], tot_loss[loss=0.122, simple_loss=0.1534, pruned_loss=0.04527, over 1959907.89 frames. ], batch size: 30, lr: 6.10e-03, grad_scale: 8.0 2022-12-08 03:46:37,787 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.05 vs. limit=5.0 2022-12-08 03:46:48,615 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.501e+02 2.002e+02 2.396e+02 3.046e+02 4.955e+02, threshold=4.792e+02, percent-clipped=0.0 2022-12-08 03:46:57,554 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93981.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:47:50,727 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=94042.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:47:52,401 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.78 vs. limit=2.0 2022-12-08 03:47:53,370 INFO [train.py:873] (3/4) Epoch 13, batch 3300, loss[loss=0.1109, simple_loss=0.1266, pruned_loss=0.04755, over 2578.00 frames. ], tot_loss[loss=0.1219, simple_loss=0.1532, pruned_loss=0.04527, over 1943800.61 frames. ], batch size: 100, lr: 6.10e-03, grad_scale: 4.0 2022-12-08 03:48:15,567 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.372e+02 2.207e+02 2.819e+02 3.660e+02 9.376e+02, threshold=5.637e+02, percent-clipped=6.0 2022-12-08 03:48:51,209 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9277, 2.0777, 1.9961, 1.8834, 2.0421, 1.8582, 1.9072, 2.0741], device='cuda:3'), covar=tensor([0.0610, 0.0944, 0.0682, 0.0473, 0.0501, 0.0489, 0.0423, 0.0569], device='cuda:3'), in_proj_covar=tensor([0.0018, 0.0018, 0.0016, 0.0017, 0.0017, 0.0028, 0.0023, 0.0028], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 03:49:03,571 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=94126.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:49:08,597 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.7767, 4.4511, 4.2979, 4.7753, 4.4475, 4.2281, 4.7901, 4.0013], device='cuda:3'), covar=tensor([0.0391, 0.0938, 0.0406, 0.0392, 0.0840, 0.0679, 0.0500, 0.0538], device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0274, 0.0195, 0.0191, 0.0186, 0.0154, 0.0278, 0.0169], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 03:49:19,900 INFO [train.py:873] (3/4) Epoch 13, batch 3400, loss[loss=0.1055, simple_loss=0.1396, pruned_loss=0.0357, over 14144.00 frames. ], tot_loss[loss=0.1206, simple_loss=0.1524, pruned_loss=0.04437, over 1947684.98 frames. ], batch size: 25, lr: 6.09e-03, grad_scale: 4.0 2022-12-08 03:49:42,649 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.317e+02 2.362e+02 2.991e+02 3.918e+02 1.858e+03, threshold=5.982e+02, percent-clipped=6.0 2022-12-08 03:49:58,698 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2022-12-08 03:50:28,546 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=94223.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 03:50:47,831 INFO [train.py:873] (3/4) Epoch 13, batch 3500, loss[loss=0.09382, simple_loss=0.1352, pruned_loss=0.0262, over 14303.00 frames. ], tot_loss[loss=0.1209, simple_loss=0.1526, pruned_loss=0.0446, over 1978265.16 frames. ], batch size: 39, lr: 6.09e-03, grad_scale: 4.0 2022-12-08 03:51:10,108 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.171e+02 2.271e+02 2.788e+02 3.634e+02 1.227e+03, threshold=5.575e+02, percent-clipped=1.0 2022-12-08 03:51:11,948 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0311, 3.8093, 3.7303, 4.1070, 3.8540, 3.5296, 4.1433, 3.4358], device='cuda:3'), covar=tensor([0.0618, 0.1090, 0.0481, 0.0485, 0.0909, 0.1630, 0.0528, 0.0606], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0274, 0.0194, 0.0189, 0.0185, 0.0153, 0.0276, 0.0167], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 03:51:22,020 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=94284.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 03:52:00,948 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2022-12-08 03:52:08,025 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=94337.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:52:14,717 INFO [train.py:873] (3/4) Epoch 13, batch 3600, loss[loss=0.1211, simple_loss=0.154, pruned_loss=0.04414, over 14458.00 frames. ], tot_loss[loss=0.1216, simple_loss=0.1532, pruned_loss=0.04505, over 2000200.72 frames. ], batch size: 51, lr: 6.09e-03, grad_scale: 8.0 2022-12-08 03:52:17,433 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=94348.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:52:37,864 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.263e+02 2.011e+02 2.604e+02 3.365e+02 5.868e+02, threshold=5.208e+02, percent-clipped=1.0 2022-12-08 03:53:03,777 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9251, 2.9864, 4.6558, 3.5828, 4.6078, 4.4080, 4.3654, 3.9723], device='cuda:3'), covar=tensor([0.0524, 0.2936, 0.0769, 0.1645, 0.0855, 0.0944, 0.1786, 0.1697], device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0313, 0.0395, 0.0301, 0.0377, 0.0321, 0.0363, 0.0304], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 03:53:10,917 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=94409.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 03:53:25,276 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=94426.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:53:26,644 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2022-12-08 03:53:34,000 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4530, 3.2569, 3.1438, 3.4943, 3.3095, 3.4482, 3.5573, 2.9073], device='cuda:3'), covar=tensor([0.0525, 0.1089, 0.0527, 0.0514, 0.0802, 0.0438, 0.0568, 0.0625], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0271, 0.0193, 0.0187, 0.0182, 0.0151, 0.0272, 0.0166], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 03:53:41,903 INFO [train.py:873] (3/4) Epoch 13, batch 3700, loss[loss=0.1649, simple_loss=0.1766, pruned_loss=0.07655, over 7758.00 frames. ], tot_loss[loss=0.1225, simple_loss=0.1533, pruned_loss=0.04581, over 1984830.96 frames. ], batch size: 100, lr: 6.08e-03, grad_scale: 4.0 2022-12-08 03:54:04,972 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.075e+02 2.353e+02 2.765e+02 3.627e+02 6.924e+02, threshold=5.531e+02, percent-clipped=2.0 2022-12-08 03:54:06,742 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=94474.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:55:09,008 INFO [train.py:873] (3/4) Epoch 13, batch 3800, loss[loss=0.09724, simple_loss=0.1412, pruned_loss=0.02665, over 14212.00 frames. ], tot_loss[loss=0.1213, simple_loss=0.1526, pruned_loss=0.04496, over 1968070.01 frames. ], batch size: 35, lr: 6.08e-03, grad_scale: 4.0 2022-12-08 03:55:10,999 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8599, 3.4847, 2.6659, 4.2725, 3.9144, 4.0102, 3.4618, 2.7727], device='cuda:3'), covar=tensor([0.1001, 0.1438, 0.3847, 0.0510, 0.1129, 0.1465, 0.1343, 0.3363], device='cuda:3'), in_proj_covar=tensor([0.0275, 0.0294, 0.0267, 0.0268, 0.0316, 0.0298, 0.0259, 0.0250], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 03:55:32,195 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.222e+02 2.117e+02 2.466e+02 3.115e+02 4.944e+02, threshold=4.932e+02, percent-clipped=0.0 2022-12-08 03:55:38,513 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=94579.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 03:56:29,403 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=94637.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:56:36,275 INFO [train.py:873] (3/4) Epoch 13, batch 3900, loss[loss=0.1524, simple_loss=0.1399, pruned_loss=0.08251, over 1275.00 frames. ], tot_loss[loss=0.1202, simple_loss=0.1524, pruned_loss=0.04395, over 2078474.80 frames. ], batch size: 100, lr: 6.08e-03, grad_scale: 4.0 2022-12-08 03:56:52,064 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9355, 3.7681, 3.4306, 3.6413, 3.8322, 3.8594, 3.9329, 3.8958], device='cuda:3'), covar=tensor([0.0814, 0.0549, 0.1988, 0.2221, 0.0744, 0.0863, 0.0866, 0.0849], device='cuda:3'), in_proj_covar=tensor([0.0378, 0.0265, 0.0445, 0.0559, 0.0328, 0.0434, 0.0385, 0.0372], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 03:56:59,461 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.067e+02 2.050e+02 2.441e+02 3.382e+02 7.650e+02, threshold=4.881e+02, percent-clipped=6.0 2022-12-08 03:57:11,480 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=94685.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 03:57:27,717 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=94704.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 03:57:34,566 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2022-12-08 03:57:37,087 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.0511, 1.2644, 1.4053, 0.9477, 0.8812, 1.2179, 0.8920, 1.2819], device='cuda:3'), covar=tensor([0.2304, 0.2442, 0.0868, 0.2845, 0.2827, 0.0887, 0.1651, 0.0986], device='cuda:3'), in_proj_covar=tensor([0.0082, 0.0096, 0.0089, 0.0096, 0.0113, 0.0084, 0.0121, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 03:58:02,598 INFO [train.py:873] (3/4) Epoch 13, batch 4000, loss[loss=0.1317, simple_loss=0.1463, pruned_loss=0.05853, over 4981.00 frames. ], tot_loss[loss=0.1198, simple_loss=0.1521, pruned_loss=0.04372, over 2057357.04 frames. ], batch size: 100, lr: 6.07e-03, grad_scale: 8.0 2022-12-08 03:58:26,194 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.300e+02 2.218e+02 2.831e+02 3.684e+02 6.338e+02, threshold=5.663e+02, percent-clipped=4.0 2022-12-08 03:59:09,605 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.3637, 4.9249, 4.7910, 5.3549, 4.9925, 4.6916, 5.3201, 4.3937], device='cuda:3'), covar=tensor([0.0338, 0.1100, 0.0391, 0.0350, 0.0796, 0.0450, 0.0464, 0.0477], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0272, 0.0191, 0.0188, 0.0182, 0.0151, 0.0275, 0.0165], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 03:59:27,763 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4207, 3.2286, 2.9791, 3.1376, 3.3431, 3.3238, 3.3874, 3.3726], device='cuda:3'), covar=tensor([0.0861, 0.0667, 0.2136, 0.2410, 0.0812, 0.1051, 0.1062, 0.0912], device='cuda:3'), in_proj_covar=tensor([0.0378, 0.0266, 0.0444, 0.0559, 0.0330, 0.0433, 0.0385, 0.0369], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 03:59:29,381 INFO [train.py:873] (3/4) Epoch 13, batch 4100, loss[loss=0.1697, simple_loss=0.1635, pruned_loss=0.08792, over 2611.00 frames. ], tot_loss[loss=0.1203, simple_loss=0.1523, pruned_loss=0.04419, over 1974444.65 frames. ], batch size: 100, lr: 6.07e-03, grad_scale: 8.0 2022-12-08 03:59:52,465 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.135e+02 2.091e+02 2.510e+02 3.105e+02 7.353e+02, threshold=5.020e+02, percent-clipped=2.0 2022-12-08 03:59:58,842 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=94879.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 04:00:25,744 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3168, 2.7332, 3.6483, 2.5275, 2.1767, 2.9859, 1.7688, 3.1194], device='cuda:3'), covar=tensor([0.0829, 0.1034, 0.0625, 0.2084, 0.2227, 0.0877, 0.3292, 0.1036], device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0098, 0.0090, 0.0097, 0.0115, 0.0086, 0.0122, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 04:00:40,991 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=94927.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 04:00:56,136 INFO [train.py:873] (3/4) Epoch 13, batch 4200, loss[loss=0.09794, simple_loss=0.1373, pruned_loss=0.02928, over 14015.00 frames. ], tot_loss[loss=0.1197, simple_loss=0.152, pruned_loss=0.04375, over 1952585.47 frames. ], batch size: 26, lr: 6.07e-03, grad_scale: 8.0 2022-12-08 04:01:15,889 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9760, 1.6627, 1.9900, 1.3830, 1.6502, 2.0582, 1.8099, 1.7358], device='cuda:3'), covar=tensor([0.0861, 0.0779, 0.0772, 0.1181, 0.1618, 0.0781, 0.0721, 0.1699], device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0177, 0.0139, 0.0125, 0.0138, 0.0149, 0.0125, 0.0138], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0005, 0.0006], device='cuda:3') 2022-12-08 04:01:19,978 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.421e+02 2.402e+02 2.842e+02 3.518e+02 1.114e+03, threshold=5.683e+02, percent-clipped=6.0 2022-12-08 04:01:30,207 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0907, 1.9826, 2.0374, 2.1466, 2.0591, 2.0438, 2.1647, 1.8191], device='cuda:3'), covar=tensor([0.0875, 0.1271, 0.0795, 0.0845, 0.1009, 0.0739, 0.0819, 0.0807], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0271, 0.0191, 0.0188, 0.0182, 0.0152, 0.0276, 0.0165], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 04:01:37,805 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5729, 2.2216, 4.4128, 4.6277, 4.3655, 2.6465, 4.5967, 3.5999], device='cuda:3'), covar=tensor([0.0296, 0.1160, 0.0842, 0.0303, 0.0369, 0.1517, 0.0271, 0.0799], device='cuda:3'), in_proj_covar=tensor([0.0290, 0.0254, 0.0368, 0.0323, 0.0266, 0.0301, 0.0303, 0.0279], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 04:01:38,520 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0423, 3.8250, 3.7125, 4.1101, 3.8460, 3.6061, 4.1178, 3.3635], device='cuda:3'), covar=tensor([0.0517, 0.0939, 0.0449, 0.0478, 0.0794, 0.1328, 0.0602, 0.0609], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0271, 0.0191, 0.0187, 0.0182, 0.0152, 0.0275, 0.0165], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 04:01:38,613 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6322, 1.6468, 1.6394, 1.4639, 1.3752, 1.1728, 0.9437, 1.0511], device='cuda:3'), covar=tensor([0.0167, 0.0209, 0.0166, 0.0203, 0.0225, 0.0349, 0.0266, 0.0367], device='cuda:3'), in_proj_covar=tensor([0.0017, 0.0018, 0.0016, 0.0017, 0.0017, 0.0028, 0.0022, 0.0027], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 04:01:51,687 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=95004.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:02:00,015 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.64 vs. limit=2.0 2022-12-08 04:02:21,535 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3703, 2.1028, 2.3212, 1.5709, 2.0407, 2.3394, 2.3455, 2.0310], device='cuda:3'), covar=tensor([0.0751, 0.0663, 0.0807, 0.1414, 0.1195, 0.0687, 0.0486, 0.1322], device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0178, 0.0139, 0.0126, 0.0139, 0.0150, 0.0126, 0.0139], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0005, 0.0006], device='cuda:3') 2022-12-08 04:02:26,456 INFO [train.py:873] (3/4) Epoch 13, batch 4300, loss[loss=0.1476, simple_loss=0.1707, pruned_loss=0.06223, over 12751.00 frames. ], tot_loss[loss=0.1211, simple_loss=0.1528, pruned_loss=0.0447, over 1951472.47 frames. ], batch size: 100, lr: 6.06e-03, grad_scale: 8.0 2022-12-08 04:02:32,645 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=95052.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:02:49,665 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.049e+02 2.351e+02 2.797e+02 3.272e+02 6.145e+02, threshold=5.594e+02, percent-clipped=1.0 2022-12-08 04:02:56,831 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3744, 4.5256, 4.8601, 4.0434, 4.6624, 4.8198, 1.7143, 4.3872], device='cuda:3'), covar=tensor([0.0295, 0.0253, 0.0285, 0.0405, 0.0248, 0.0264, 0.3081, 0.0238], device='cuda:3'), in_proj_covar=tensor([0.0161, 0.0169, 0.0141, 0.0139, 0.0199, 0.0133, 0.0157, 0.0188], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 04:03:18,356 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95105.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 04:03:53,024 INFO [train.py:873] (3/4) Epoch 13, batch 4400, loss[loss=0.1319, simple_loss=0.1618, pruned_loss=0.05097, over 14156.00 frames. ], tot_loss[loss=0.1204, simple_loss=0.1525, pruned_loss=0.04417, over 1937373.42 frames. ], batch size: 84, lr: 6.06e-03, grad_scale: 8.0 2022-12-08 04:03:54,968 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.2412, 3.7063, 2.9842, 4.5394, 4.1915, 4.3344, 3.7779, 3.0288], device='cuda:3'), covar=tensor([0.0749, 0.1227, 0.3323, 0.0411, 0.0925, 0.1084, 0.1079, 0.3112], device='cuda:3'), in_proj_covar=tensor([0.0274, 0.0292, 0.0267, 0.0265, 0.0314, 0.0297, 0.0256, 0.0246], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 04:04:11,066 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95166.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 04:04:16,201 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.201e+02 2.204e+02 2.628e+02 3.192e+02 7.807e+02, threshold=5.256e+02, percent-clipped=2.0 2022-12-08 04:04:41,238 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.76 vs. limit=5.0 2022-12-08 04:05:20,306 INFO [train.py:873] (3/4) Epoch 13, batch 4500, loss[loss=0.08399, simple_loss=0.1325, pruned_loss=0.01774, over 13888.00 frames. ], tot_loss[loss=0.1204, simple_loss=0.1526, pruned_loss=0.04404, over 1932336.03 frames. ], batch size: 20, lr: 6.06e-03, grad_scale: 8.0 2022-12-08 04:05:33,308 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95260.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:05:38,523 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.4192, 5.1419, 4.6840, 4.9451, 4.8422, 5.2467, 5.3077, 5.3066], device='cuda:3'), covar=tensor([0.0617, 0.0392, 0.1979, 0.2381, 0.0801, 0.0635, 0.0830, 0.0746], device='cuda:3'), in_proj_covar=tensor([0.0378, 0.0270, 0.0443, 0.0559, 0.0331, 0.0433, 0.0388, 0.0371], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 04:05:40,331 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95268.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:05:43,635 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.015e+02 2.327e+02 3.103e+02 3.682e+02 7.295e+02, threshold=6.206e+02, percent-clipped=3.0 2022-12-08 04:06:26,378 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95321.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:06:33,400 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95329.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:06:41,725 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95339.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:06:42,517 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8766, 1.8389, 1.5823, 2.0032, 1.8112, 1.8683, 1.7533, 1.6760], device='cuda:3'), covar=tensor([0.1009, 0.0912, 0.2097, 0.0536, 0.1135, 0.0558, 0.1544, 0.0752], device='cuda:3'), in_proj_covar=tensor([0.0275, 0.0294, 0.0268, 0.0267, 0.0315, 0.0298, 0.0258, 0.0247], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 04:06:46,994 INFO [train.py:873] (3/4) Epoch 13, batch 4600, loss[loss=0.1383, simple_loss=0.1623, pruned_loss=0.05718, over 11182.00 frames. ], tot_loss[loss=0.1208, simple_loss=0.1528, pruned_loss=0.04437, over 1911753.04 frames. ], batch size: 100, lr: 6.05e-03, grad_scale: 8.0 2022-12-08 04:07:10,470 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.352e+02 2.077e+02 2.550e+02 3.185e+02 6.474e+02, threshold=5.099e+02, percent-clipped=1.0 2022-12-08 04:07:29,640 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8868, 1.6457, 1.7235, 1.7229, 1.9270, 1.0557, 1.6422, 1.7469], device='cuda:3'), covar=tensor([0.1148, 0.1049, 0.0796, 0.1006, 0.0987, 0.0910, 0.0773, 0.0654], device='cuda:3'), in_proj_covar=tensor([0.0029, 0.0029, 0.0031, 0.0027, 0.0029, 0.0041, 0.0029, 0.0031], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 04:07:35,106 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95400.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:07:50,121 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8431, 1.9492, 2.1019, 1.4187, 1.4929, 1.9893, 1.0781, 1.8609], device='cuda:3'), covar=tensor([0.1107, 0.1645, 0.0910, 0.2351, 0.2629, 0.0962, 0.3654, 0.1041], device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0097, 0.0091, 0.0097, 0.0114, 0.0085, 0.0121, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 04:08:14,531 INFO [train.py:873] (3/4) Epoch 13, batch 4700, loss[loss=0.1506, simple_loss=0.1631, pruned_loss=0.069, over 5961.00 frames. ], tot_loss[loss=0.1217, simple_loss=0.1535, pruned_loss=0.04492, over 1935187.78 frames. ], batch size: 100, lr: 6.05e-03, grad_scale: 4.0 2022-12-08 04:08:25,507 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.0649, 1.1845, 1.0284, 1.1454, 1.2683, 0.6295, 0.9496, 1.0714], device='cuda:3'), covar=tensor([0.0607, 0.0794, 0.0733, 0.0598, 0.0503, 0.0678, 0.0980, 0.0692], device='cuda:3'), in_proj_covar=tensor([0.0029, 0.0029, 0.0031, 0.0028, 0.0029, 0.0041, 0.0029, 0.0031], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 04:08:27,935 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9239, 1.7615, 3.1159, 2.2439, 2.9200, 1.7597, 2.3059, 2.8620], device='cuda:3'), covar=tensor([0.1051, 0.4449, 0.0644, 0.4590, 0.0883, 0.3347, 0.1601, 0.0864], device='cuda:3'), in_proj_covar=tensor([0.0250, 0.0210, 0.0207, 0.0283, 0.0225, 0.0209, 0.0210, 0.0209], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 04:08:28,709 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95461.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 04:08:29,634 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8926, 2.0159, 4.9757, 4.5263, 4.2888, 5.1768, 4.8870, 5.1129], device='cuda:3'), covar=tensor([0.1678, 0.1494, 0.0109, 0.0200, 0.0219, 0.0099, 0.0128, 0.0105], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0159, 0.0127, 0.0167, 0.0145, 0.0140, 0.0123, 0.0120], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 04:08:33,790 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2022-12-08 04:08:35,956 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0977, 3.2591, 3.1402, 3.2014, 3.1892, 2.8957, 1.3501, 3.0122], device='cuda:3'), covar=tensor([0.0565, 0.0538, 0.0776, 0.0557, 0.0617, 0.0804, 0.4141, 0.0569], device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0174, 0.0145, 0.0143, 0.0203, 0.0139, 0.0162, 0.0194], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 04:08:39,145 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.429e+02 2.071e+02 2.759e+02 3.394e+02 8.943e+02, threshold=5.519e+02, percent-clipped=4.0 2022-12-08 04:09:10,419 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.2341, 2.0902, 4.2636, 2.9697, 4.0574, 2.2138, 3.2074, 4.0552], device='cuda:3'), covar=tensor([0.0576, 0.4318, 0.0436, 0.5685, 0.0596, 0.3172, 0.1300, 0.0381], device='cuda:3'), in_proj_covar=tensor([0.0251, 0.0209, 0.0207, 0.0282, 0.0226, 0.0210, 0.0211, 0.0209], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 04:09:41,415 INFO [train.py:873] (3/4) Epoch 13, batch 4800, loss[loss=0.1116, simple_loss=0.1482, pruned_loss=0.03747, over 14643.00 frames. ], tot_loss[loss=0.1216, simple_loss=0.1531, pruned_loss=0.04506, over 1918471.23 frames. ], batch size: 23, lr: 6.05e-03, grad_scale: 8.0 2022-12-08 04:10:05,559 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2022-12-08 04:10:05,764 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.497e+02 2.302e+02 2.972e+02 3.662e+02 7.055e+02, threshold=5.944e+02, percent-clipped=3.0 2022-12-08 04:10:15,458 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.52 vs. limit=2.0 2022-12-08 04:10:43,513 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95616.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:10:44,434 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95617.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:10:50,198 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95624.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:11:08,088 INFO [train.py:873] (3/4) Epoch 13, batch 4900, loss[loss=0.1201, simple_loss=0.1579, pruned_loss=0.04112, over 14412.00 frames. ], tot_loss[loss=0.1216, simple_loss=0.1532, pruned_loss=0.04504, over 1930872.99 frames. ], batch size: 41, lr: 6.05e-03, grad_scale: 8.0 2022-12-08 04:11:32,372 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.329e+02 2.158e+02 2.610e+02 3.322e+02 1.003e+03, threshold=5.220e+02, percent-clipped=4.0 2022-12-08 04:11:36,603 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95678.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:11:43,269 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9678, 4.0565, 4.2869, 3.7958, 4.1441, 4.2080, 1.7389, 3.8970], device='cuda:3'), covar=tensor([0.0295, 0.0305, 0.0318, 0.0448, 0.0271, 0.0297, 0.3200, 0.0271], device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0173, 0.0144, 0.0143, 0.0203, 0.0138, 0.0161, 0.0192], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 04:11:44,573 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.43 vs. limit=5.0 2022-12-08 04:11:51,489 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95695.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:12:35,019 INFO [train.py:873] (3/4) Epoch 13, batch 5000, loss[loss=0.1433, simple_loss=0.1582, pruned_loss=0.06418, over 4971.00 frames. ], tot_loss[loss=0.1214, simple_loss=0.153, pruned_loss=0.04496, over 1925981.96 frames. ], batch size: 100, lr: 6.04e-03, grad_scale: 8.0 2022-12-08 04:12:49,273 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=95761.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 04:12:59,674 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.354e+02 2.284e+02 2.864e+02 3.553e+02 7.569e+02, threshold=5.729e+02, percent-clipped=2.0 2022-12-08 04:13:01,030 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4976, 3.7540, 3.7460, 3.3714, 2.7085, 3.6484, 3.4735, 1.8865], device='cuda:3'), covar=tensor([0.1928, 0.1032, 0.1235, 0.1026, 0.0988, 0.0778, 0.1055, 0.2270], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0084, 0.0066, 0.0069, 0.0097, 0.0082, 0.0097, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0006, 0.0006], device='cuda:3') 2022-12-08 04:13:31,718 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=95809.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 04:13:34,709 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95812.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 04:13:56,604 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.12 vs. limit=2.0 2022-12-08 04:14:03,054 INFO [train.py:873] (3/4) Epoch 13, batch 5100, loss[loss=0.1393, simple_loss=0.1688, pruned_loss=0.05491, over 14482.00 frames. ], tot_loss[loss=0.1221, simple_loss=0.1532, pruned_loss=0.0455, over 1888016.94 frames. ], batch size: 51, lr: 6.04e-03, grad_scale: 8.0 2022-12-08 04:14:05,092 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95847.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:14:08,906 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.66 vs. limit=2.0 2022-12-08 04:14:27,856 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.147e+02 2.050e+02 2.508e+02 3.173e+02 5.976e+02, threshold=5.017e+02, percent-clipped=1.0 2022-12-08 04:14:28,133 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95873.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 04:14:45,808 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0240, 2.2010, 2.2851, 2.2629, 2.0247, 2.3474, 2.0730, 1.4676], device='cuda:3'), covar=tensor([0.0889, 0.1108, 0.0624, 0.1080, 0.1060, 0.0579, 0.1086, 0.2033], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0083, 0.0066, 0.0069, 0.0097, 0.0083, 0.0098, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0006, 0.0006], device='cuda:3') 2022-12-08 04:14:52,142 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.2230, 3.8495, 3.8241, 4.2120, 3.8218, 3.4400, 4.2766, 4.1231], device='cuda:3'), covar=tensor([0.0595, 0.0834, 0.0843, 0.0556, 0.0831, 0.0785, 0.0559, 0.0625], device='cuda:3'), in_proj_covar=tensor([0.0140, 0.0135, 0.0145, 0.0154, 0.0142, 0.0119, 0.0163, 0.0143], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 04:14:58,776 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 2022-12-08 04:14:59,200 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95908.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 04:15:06,254 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=95916.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:15:13,316 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=95924.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:15:26,827 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.6060, 4.2451, 4.2027, 4.6601, 4.3240, 4.0835, 4.6077, 3.9518], device='cuda:3'), covar=tensor([0.0456, 0.1232, 0.0406, 0.0437, 0.0796, 0.0786, 0.0613, 0.0530], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0270, 0.0190, 0.0187, 0.0184, 0.0151, 0.0278, 0.0164], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 04:15:31,806 INFO [train.py:873] (3/4) Epoch 13, batch 5200, loss[loss=0.1539, simple_loss=0.1773, pruned_loss=0.06523, over 10372.00 frames. ], tot_loss[loss=0.1222, simple_loss=0.1535, pruned_loss=0.0455, over 1961522.36 frames. ], batch size: 100, lr: 6.04e-03, grad_scale: 8.0 2022-12-08 04:15:43,082 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2022-12-08 04:15:48,364 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=95964.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:15:51,293 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5785, 4.6604, 5.0502, 4.2674, 4.8555, 5.0895, 1.9268, 4.4921], device='cuda:3'), covar=tensor([0.0292, 0.0266, 0.0338, 0.0394, 0.0249, 0.0135, 0.3123, 0.0286], device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0170, 0.0140, 0.0140, 0.0199, 0.0136, 0.0157, 0.0188], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 04:15:55,707 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=95972.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:15:56,531 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.143e+02 2.184e+02 2.660e+02 3.283e+02 4.676e+02, threshold=5.321e+02, percent-clipped=0.0 2022-12-08 04:15:56,634 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95973.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:16:03,038 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95980.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:16:15,823 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=95995.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:16:18,549 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0372, 2.1371, 2.1975, 1.7798, 2.1751, 1.3151, 2.0162, 2.0490], device='cuda:3'), covar=tensor([0.1026, 0.0989, 0.0784, 0.2700, 0.1164, 0.0809, 0.0708, 0.0685], device='cuda:3'), in_proj_covar=tensor([0.0029, 0.0029, 0.0031, 0.0028, 0.0029, 0.0042, 0.0030, 0.0032], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 04:16:56,507 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96041.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:16:58,092 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96043.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:16:59,770 INFO [train.py:873] (3/4) Epoch 13, batch 5300, loss[loss=0.1269, simple_loss=0.1633, pruned_loss=0.0452, over 14292.00 frames. ], tot_loss[loss=0.1212, simple_loss=0.1531, pruned_loss=0.04468, over 1990960.04 frames. ], batch size: 37, lr: 6.03e-03, grad_scale: 8.0 2022-12-08 04:17:23,907 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.342e+02 2.223e+02 2.915e+02 3.617e+02 9.982e+02, threshold=5.829e+02, percent-clipped=6.0 2022-12-08 04:17:26,580 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4419, 2.4570, 2.5752, 2.5073, 2.5038, 2.2206, 1.4376, 2.2634], device='cuda:3'), covar=tensor([0.0594, 0.0481, 0.0422, 0.0385, 0.0415, 0.1105, 0.2619, 0.0451], device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0170, 0.0141, 0.0141, 0.0201, 0.0136, 0.0158, 0.0190], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 04:17:29,226 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5376, 1.9839, 2.4975, 2.6025, 2.4526, 2.0509, 2.6109, 2.1718], device='cuda:3'), covar=tensor([0.0358, 0.0821, 0.0468, 0.0392, 0.0492, 0.0980, 0.0395, 0.0626], device='cuda:3'), in_proj_covar=tensor([0.0287, 0.0253, 0.0366, 0.0322, 0.0264, 0.0299, 0.0302, 0.0278], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 04:17:29,941 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96080.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:17:53,192 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.8769, 5.2815, 5.3423, 5.8008, 5.3954, 4.7023, 5.7387, 4.6705], device='cuda:3'), covar=tensor([0.0290, 0.0950, 0.0310, 0.0404, 0.0783, 0.0376, 0.0468, 0.0551], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0272, 0.0190, 0.0188, 0.0184, 0.0151, 0.0278, 0.0165], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 04:18:23,824 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96141.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:18:27,060 INFO [train.py:873] (3/4) Epoch 13, batch 5400, loss[loss=0.12, simple_loss=0.1506, pruned_loss=0.04472, over 6899.00 frames. ], tot_loss[loss=0.1198, simple_loss=0.1521, pruned_loss=0.04372, over 2002013.05 frames. ], batch size: 100, lr: 6.03e-03, grad_scale: 8.0 2022-12-08 04:18:38,370 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3530, 4.1616, 3.8466, 3.9693, 4.2159, 4.2330, 4.3430, 4.3060], device='cuda:3'), covar=tensor([0.0838, 0.0563, 0.2178, 0.2712, 0.0744, 0.0842, 0.0970, 0.0865], device='cuda:3'), in_proj_covar=tensor([0.0379, 0.0269, 0.0443, 0.0557, 0.0332, 0.0429, 0.0385, 0.0372], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 04:18:44,252 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96165.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:18:46,939 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96168.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 04:18:51,339 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.196e+02 2.106e+02 2.759e+02 3.696e+02 7.492e+02, threshold=5.517e+02, percent-clipped=3.0 2022-12-08 04:19:00,193 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.46 vs. limit=2.0 2022-12-08 04:19:17,389 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96203.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 04:19:37,240 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96226.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:19:53,397 INFO [train.py:873] (3/4) Epoch 13, batch 5500, loss[loss=0.114, simple_loss=0.1583, pruned_loss=0.03487, over 14262.00 frames. ], tot_loss[loss=0.1198, simple_loss=0.1519, pruned_loss=0.04384, over 1966288.91 frames. ], batch size: 39, lr: 6.03e-03, grad_scale: 8.0 2022-12-08 04:20:17,921 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.945e+01 2.434e+02 2.937e+02 3.811e+02 8.980e+02, threshold=5.874e+02, percent-clipped=5.0 2022-12-08 04:20:18,452 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96273.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:20:42,571 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.3814, 5.1566, 4.7427, 5.0081, 4.9649, 5.2469, 5.3725, 5.3352], device='cuda:3'), covar=tensor([0.0589, 0.0378, 0.2077, 0.2271, 0.0685, 0.0758, 0.0682, 0.0723], device='cuda:3'), in_proj_covar=tensor([0.0377, 0.0270, 0.0441, 0.0554, 0.0330, 0.0427, 0.0382, 0.0371], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 04:20:46,295 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6000, 1.5837, 1.7734, 1.5371, 1.5614, 1.4444, 1.4219, 1.1284], device='cuda:3'), covar=tensor([0.0233, 0.0411, 0.0192, 0.0203, 0.0246, 0.0317, 0.0251, 0.0498], device='cuda:3'), in_proj_covar=tensor([0.0018, 0.0019, 0.0016, 0.0017, 0.0017, 0.0029, 0.0023, 0.0028], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 04:21:00,273 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96321.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:21:02,935 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96324.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:21:13,589 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96336.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:21:21,090 INFO [train.py:873] (3/4) Epoch 13, batch 5600, loss[loss=0.1275, simple_loss=0.1556, pruned_loss=0.04968, over 14255.00 frames. ], tot_loss[loss=0.1202, simple_loss=0.1522, pruned_loss=0.04406, over 1940221.41 frames. ], batch size: 80, lr: 6.02e-03, grad_scale: 8.0 2022-12-08 04:21:45,613 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.402e+02 2.301e+02 2.704e+02 3.458e+02 6.924e+02, threshold=5.409e+02, percent-clipped=1.0 2022-12-08 04:21:56,172 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96385.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:22:00,540 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96390.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:22:05,755 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7617, 1.2811, 1.7413, 1.2560, 1.4055, 1.7891, 1.5247, 1.5275], device='cuda:3'), covar=tensor([0.0687, 0.0847, 0.0627, 0.0899, 0.1317, 0.0739, 0.0533, 0.1452], device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0177, 0.0139, 0.0125, 0.0137, 0.0150, 0.0124, 0.0138], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0005, 0.0006], device='cuda:3') 2022-12-08 04:22:09,408 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1363, 2.0184, 1.8838, 1.9139, 2.0355, 2.0988, 2.1028, 2.1026], device='cuda:3'), covar=tensor([0.1321, 0.0975, 0.2981, 0.2701, 0.1620, 0.1157, 0.1691, 0.1168], device='cuda:3'), in_proj_covar=tensor([0.0373, 0.0266, 0.0435, 0.0549, 0.0329, 0.0422, 0.0381, 0.0367], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 04:22:09,918 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 2022-12-08 04:22:11,272 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0259, 2.6288, 3.4988, 2.5225, 2.1075, 2.8993, 1.6080, 2.9246], device='cuda:3'), covar=tensor([0.1213, 0.1586, 0.0776, 0.1792, 0.2522, 0.1164, 0.3563, 0.1401], device='cuda:3'), in_proj_covar=tensor([0.0084, 0.0098, 0.0091, 0.0097, 0.0115, 0.0087, 0.0122, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 04:22:12,655 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2022-12-08 04:22:20,453 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4685, 4.5318, 4.8358, 4.1879, 4.6602, 4.9211, 1.7764, 4.3636], device='cuda:3'), covar=tensor([0.0269, 0.0294, 0.0386, 0.0479, 0.0334, 0.0189, 0.3294, 0.0278], device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0172, 0.0144, 0.0142, 0.0203, 0.0138, 0.0160, 0.0191], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 04:22:41,419 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96436.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:22:42,290 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7390, 3.5292, 3.2961, 3.3919, 3.6272, 3.6419, 3.7097, 3.6817], device='cuda:3'), covar=tensor([0.0801, 0.0595, 0.2063, 0.2562, 0.0782, 0.0892, 0.1042, 0.0920], device='cuda:3'), in_proj_covar=tensor([0.0372, 0.0267, 0.0436, 0.0549, 0.0329, 0.0422, 0.0381, 0.0367], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 04:22:48,796 INFO [train.py:873] (3/4) Epoch 13, batch 5700, loss[loss=0.1577, simple_loss=0.1547, pruned_loss=0.08033, over 2665.00 frames. ], tot_loss[loss=0.1215, simple_loss=0.1532, pruned_loss=0.04491, over 1965110.22 frames. ], batch size: 100, lr: 6.02e-03, grad_scale: 8.0 2022-12-08 04:22:54,741 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96451.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:23:09,366 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96468.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 04:23:11,383 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2022-12-08 04:23:13,579 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.334e+02 2.382e+02 2.762e+02 3.323e+02 6.415e+02, threshold=5.524e+02, percent-clipped=3.0 2022-12-08 04:23:33,927 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3746, 2.4216, 4.9028, 4.4702, 4.3880, 4.9899, 4.7398, 5.0141], device='cuda:3'), covar=tensor([0.1214, 0.1105, 0.0085, 0.0169, 0.0166, 0.0088, 0.0096, 0.0089], device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0156, 0.0126, 0.0165, 0.0144, 0.0138, 0.0121, 0.0118], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 04:23:40,524 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96503.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:23:51,374 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96516.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 04:23:55,896 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96521.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:24:17,061 INFO [train.py:873] (3/4) Epoch 13, batch 5800, loss[loss=0.111, simple_loss=0.1465, pruned_loss=0.03779, over 14215.00 frames. ], tot_loss[loss=0.1217, simple_loss=0.1533, pruned_loss=0.04502, over 1978229.55 frames. ], batch size: 37, lr: 6.02e-03, grad_scale: 4.0 2022-12-08 04:24:22,201 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96551.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:24:34,696 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.71 vs. limit=5.0 2022-12-08 04:24:42,624 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.152e+02 2.229e+02 2.647e+02 3.440e+02 6.528e+02, threshold=5.293e+02, percent-clipped=3.0 2022-12-08 04:25:30,062 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9348, 1.9356, 1.9138, 2.0683, 1.9107, 1.7120, 1.4247, 1.6887], device='cuda:3'), covar=tensor([0.0804, 0.0745, 0.0769, 0.0448, 0.0761, 0.1606, 0.2481, 0.0811], device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0173, 0.0144, 0.0142, 0.0203, 0.0138, 0.0160, 0.0190], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 04:25:37,117 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96636.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:25:37,176 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7323, 1.8945, 1.8921, 1.8624, 2.0566, 1.5848, 1.6568, 1.3107], device='cuda:3'), covar=tensor([0.0303, 0.0258, 0.0332, 0.0266, 0.0289, 0.0245, 0.0196, 0.0315], device='cuda:3'), in_proj_covar=tensor([0.0018, 0.0018, 0.0016, 0.0017, 0.0017, 0.0028, 0.0022, 0.0027], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 04:25:44,390 INFO [train.py:873] (3/4) Epoch 13, batch 5900, loss[loss=0.1513, simple_loss=0.1448, pruned_loss=0.07887, over 1221.00 frames. ], tot_loss[loss=0.1208, simple_loss=0.1524, pruned_loss=0.04461, over 1906271.67 frames. ], batch size: 100, lr: 6.01e-03, grad_scale: 4.0 2022-12-08 04:26:10,182 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.137e+02 2.158e+02 2.566e+02 3.278e+02 1.082e+03, threshold=5.131e+02, percent-clipped=5.0 2022-12-08 04:26:15,616 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96680.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:26:18,913 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96684.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:26:58,909 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0311, 3.5791, 2.8492, 4.3526, 4.0193, 4.1105, 3.5773, 2.9820], device='cuda:3'), covar=tensor([0.0793, 0.1445, 0.3630, 0.0489, 0.0892, 0.1459, 0.1288, 0.3151], device='cuda:3'), in_proj_covar=tensor([0.0272, 0.0291, 0.0266, 0.0267, 0.0313, 0.0297, 0.0256, 0.0244], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 04:27:03,981 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96736.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:27:11,549 INFO [train.py:873] (3/4) Epoch 13, batch 6000, loss[loss=0.1414, simple_loss=0.1567, pruned_loss=0.06306, over 7781.00 frames. ], tot_loss[loss=0.1197, simple_loss=0.1517, pruned_loss=0.04382, over 1986272.29 frames. ], batch size: 100, lr: 6.01e-03, grad_scale: 8.0 2022-12-08 04:27:11,550 INFO [train.py:896] (3/4) Computing validation loss 2022-12-08 04:27:19,805 INFO [train.py:905] (3/4) Epoch 13, validation: loss=0.132, simple_loss=0.1717, pruned_loss=0.04613, over 857387.00 frames. 2022-12-08 04:27:19,806 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-08 04:27:20,805 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96746.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:27:45,010 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.225e+02 2.016e+02 2.616e+02 3.028e+02 5.890e+02, threshold=5.232e+02, percent-clipped=1.0 2022-12-08 04:27:46,770 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96776.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:27:53,292 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96784.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:28:25,797 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96821.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:28:26,832 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2022-12-08 04:28:40,254 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96837.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:28:47,195 INFO [train.py:873] (3/4) Epoch 13, batch 6100, loss[loss=0.1305, simple_loss=0.1436, pruned_loss=0.05874, over 3867.00 frames. ], tot_loss[loss=0.1208, simple_loss=0.1528, pruned_loss=0.04447, over 2051094.46 frames. ], batch size: 100, lr: 6.01e-03, grad_scale: 8.0 2022-12-08 04:28:55,684 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.6768, 4.4605, 4.0784, 4.2986, 4.4628, 4.5746, 4.6812, 4.6589], device='cuda:3'), covar=tensor([0.0795, 0.0473, 0.2214, 0.2897, 0.0785, 0.0821, 0.0831, 0.0786], device='cuda:3'), in_proj_covar=tensor([0.0376, 0.0266, 0.0439, 0.0553, 0.0330, 0.0430, 0.0381, 0.0371], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 04:29:07,686 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96869.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:29:12,500 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.404e+02 2.137e+02 2.568e+02 3.326e+02 6.550e+02, threshold=5.137e+02, percent-clipped=4.0 2022-12-08 04:29:17,874 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.68 vs. limit=2.0 2022-12-08 04:30:04,733 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4983, 1.4824, 3.5561, 1.5708, 3.4095, 3.6076, 2.4081, 3.8485], device='cuda:3'), covar=tensor([0.0248, 0.3153, 0.0384, 0.2277, 0.0676, 0.0404, 0.0973, 0.0170], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0157, 0.0158, 0.0168, 0.0167, 0.0175, 0.0133, 0.0148], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 04:30:12,562 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.2573, 2.8908, 4.0355, 3.1368, 4.0426, 4.0677, 3.8411, 3.4460], device='cuda:3'), covar=tensor([0.0911, 0.2847, 0.1256, 0.2035, 0.0973, 0.0960, 0.1888, 0.2084], device='cuda:3'), in_proj_covar=tensor([0.0355, 0.0316, 0.0400, 0.0304, 0.0379, 0.0323, 0.0367, 0.0308], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 04:30:13,976 INFO [train.py:873] (3/4) Epoch 13, batch 6200, loss[loss=0.16, simple_loss=0.1605, pruned_loss=0.0797, over 3843.00 frames. ], tot_loss[loss=0.1192, simple_loss=0.1519, pruned_loss=0.04331, over 2064565.80 frames. ], batch size: 100, lr: 6.00e-03, grad_scale: 8.0 2022-12-08 04:30:39,952 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.066e+02 2.243e+02 2.655e+02 3.164e+02 5.951e+02, threshold=5.310e+02, percent-clipped=2.0 2022-12-08 04:30:45,301 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96980.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:30:59,151 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5591, 2.1481, 3.4720, 3.6542, 3.4941, 2.2933, 3.5328, 2.6683], device='cuda:3'), covar=tensor([0.0473, 0.1055, 0.0809, 0.0480, 0.0445, 0.1466, 0.0385, 0.1009], device='cuda:3'), in_proj_covar=tensor([0.0289, 0.0254, 0.0369, 0.0323, 0.0264, 0.0299, 0.0303, 0.0279], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 04:31:03,616 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97000.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:31:08,001 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.95 vs. limit=2.0 2022-12-08 04:31:28,040 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=97028.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:31:43,227 INFO [train.py:873] (3/4) Epoch 13, batch 6300, loss[loss=0.1664, simple_loss=0.1473, pruned_loss=0.09278, over 1264.00 frames. ], tot_loss[loss=0.1195, simple_loss=0.1519, pruned_loss=0.04359, over 1989089.98 frames. ], batch size: 100, lr: 6.00e-03, grad_scale: 4.0 2022-12-08 04:31:44,303 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97046.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:31:56,994 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97061.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:32:09,641 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.237e+02 2.227e+02 2.717e+02 3.666e+02 6.842e+02, threshold=5.435e+02, percent-clipped=6.0 2022-12-08 04:32:25,910 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=97094.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:32:49,429 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8525, 1.4926, 2.9614, 2.6965, 2.8990, 3.0070, 2.1657, 2.9528], device='cuda:3'), covar=tensor([0.1187, 0.1333, 0.0151, 0.0344, 0.0313, 0.0161, 0.0495, 0.0182], device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0156, 0.0127, 0.0165, 0.0142, 0.0138, 0.0121, 0.0119], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 04:32:58,879 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97132.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:33:06,694 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.3584, 1.0782, 1.2646, 0.8062, 1.0653, 1.4179, 1.0061, 1.1383], device='cuda:3'), covar=tensor([0.0358, 0.0772, 0.0626, 0.0455, 0.1097, 0.0796, 0.0456, 0.1097], device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0174, 0.0137, 0.0125, 0.0136, 0.0147, 0.0123, 0.0137], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0005, 0.0006], device='cuda:3') 2022-12-08 04:33:09,961 INFO [train.py:873] (3/4) Epoch 13, batch 6400, loss[loss=0.1328, simple_loss=0.1351, pruned_loss=0.06527, over 2655.00 frames. ], tot_loss[loss=0.1199, simple_loss=0.1522, pruned_loss=0.04376, over 1985751.33 frames. ], batch size: 100, lr: 6.00e-03, grad_scale: 8.0 2022-12-08 04:33:36,266 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.400e+02 2.249e+02 2.635e+02 3.271e+02 6.214e+02, threshold=5.270e+02, percent-clipped=1.0 2022-12-08 04:34:37,201 INFO [train.py:873] (3/4) Epoch 13, batch 6500, loss[loss=0.1611, simple_loss=0.1452, pruned_loss=0.08853, over 1262.00 frames. ], tot_loss[loss=0.1207, simple_loss=0.1527, pruned_loss=0.04434, over 1978803.26 frames. ], batch size: 100, lr: 6.00e-03, grad_scale: 8.0 2022-12-08 04:34:50,071 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97260.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:35:03,585 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.464e+02 2.327e+02 2.807e+02 3.571e+02 7.532e+02, threshold=5.613e+02, percent-clipped=5.0 2022-12-08 04:35:44,117 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97321.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:35:46,906 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4875, 4.5649, 4.8404, 4.1022, 4.6611, 4.9019, 1.9859, 4.3818], device='cuda:3'), covar=tensor([0.0248, 0.0300, 0.0385, 0.0464, 0.0264, 0.0183, 0.3167, 0.0266], device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0170, 0.0142, 0.0140, 0.0199, 0.0135, 0.0157, 0.0189], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 04:36:02,190 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97342.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:36:04,579 INFO [train.py:873] (3/4) Epoch 13, batch 6600, loss[loss=0.1272, simple_loss=0.1589, pruned_loss=0.04775, over 13542.00 frames. ], tot_loss[loss=0.1205, simple_loss=0.1522, pruned_loss=0.04441, over 1926038.98 frames. ], batch size: 100, lr: 5.99e-03, grad_scale: 8.0 2022-12-08 04:36:05,604 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1506, 1.9474, 2.0437, 1.8624, 2.0315, 1.3077, 1.7308, 1.9017], device='cuda:3'), covar=tensor([0.0846, 0.0831, 0.0600, 0.0828, 0.1325, 0.0798, 0.0776, 0.0671], device='cuda:3'), in_proj_covar=tensor([0.0030, 0.0030, 0.0032, 0.0028, 0.0030, 0.0043, 0.0031, 0.0032], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 04:36:06,374 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97347.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:36:14,719 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97356.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:36:30,359 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97374.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:36:31,022 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.234e+02 2.194e+02 2.777e+02 3.424e+02 5.452e+02, threshold=5.554e+02, percent-clipped=0.0 2022-12-08 04:36:49,123 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.46 vs. limit=2.0 2022-12-08 04:36:49,656 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9800, 2.0145, 2.0386, 2.0586, 1.9887, 1.6576, 1.3988, 1.7689], device='cuda:3'), covar=tensor([0.0636, 0.0527, 0.0528, 0.0362, 0.0427, 0.1251, 0.2177, 0.0505], device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0170, 0.0141, 0.0139, 0.0198, 0.0134, 0.0156, 0.0188], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 04:36:55,848 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97403.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 04:37:00,394 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97408.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:37:02,859 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5813, 4.6792, 5.0289, 4.2371, 4.7728, 5.0336, 1.8799, 4.4168], device='cuda:3'), covar=tensor([0.0297, 0.0255, 0.0309, 0.0350, 0.0260, 0.0136, 0.3198, 0.0261], device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0170, 0.0141, 0.0139, 0.0198, 0.0134, 0.0156, 0.0188], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 04:37:18,371 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.63 vs. limit=5.0 2022-12-08 04:37:21,174 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97432.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:37:23,896 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97435.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:37:32,758 INFO [train.py:873] (3/4) Epoch 13, batch 6700, loss[loss=0.1196, simple_loss=0.1477, pruned_loss=0.04576, over 6913.00 frames. ], tot_loss[loss=0.1203, simple_loss=0.1522, pruned_loss=0.04423, over 1923218.47 frames. ], batch size: 100, lr: 5.99e-03, grad_scale: 8.0 2022-12-08 04:37:59,104 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.315e+02 2.128e+02 2.579e+02 3.037e+02 5.604e+02, threshold=5.159e+02, percent-clipped=1.0 2022-12-08 04:38:03,381 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=97480.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:38:21,695 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.17 vs. limit=5.0 2022-12-08 04:38:59,023 INFO [train.py:873] (3/4) Epoch 13, batch 6800, loss[loss=0.1158, simple_loss=0.154, pruned_loss=0.03882, over 14593.00 frames. ], tot_loss[loss=0.1211, simple_loss=0.1524, pruned_loss=0.04491, over 1913599.13 frames. ], batch size: 23, lr: 5.99e-03, grad_scale: 8.0 2022-12-08 04:39:08,520 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97555.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:39:25,407 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.218e+02 2.339e+02 3.043e+02 3.694e+02 1.051e+03, threshold=6.086e+02, percent-clipped=5.0 2022-12-08 04:39:27,315 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6126, 1.7826, 1.8973, 1.2643, 1.2242, 1.7146, 1.1662, 1.6926], device='cuda:3'), covar=tensor([0.1512, 0.2085, 0.0825, 0.2303, 0.3096, 0.1221, 0.2804, 0.1074], device='cuda:3'), in_proj_covar=tensor([0.0084, 0.0100, 0.0091, 0.0098, 0.0115, 0.0087, 0.0122, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 04:39:29,112 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0031, 1.8054, 4.2773, 3.9536, 3.8990, 4.3646, 3.7782, 4.3321], device='cuda:3'), covar=tensor([0.1444, 0.1454, 0.0110, 0.0214, 0.0231, 0.0118, 0.0206, 0.0121], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0160, 0.0129, 0.0168, 0.0146, 0.0141, 0.0123, 0.0121], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 04:39:54,609 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.63 vs. limit=2.0 2022-12-08 04:40:01,895 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97616.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:40:02,006 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97616.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:40:27,573 INFO [train.py:873] (3/4) Epoch 13, batch 6900, loss[loss=0.09966, simple_loss=0.1453, pruned_loss=0.02702, over 14571.00 frames. ], tot_loss[loss=0.1211, simple_loss=0.1522, pruned_loss=0.04504, over 1897294.07 frames. ], batch size: 34, lr: 5.98e-03, grad_scale: 8.0 2022-12-08 04:40:36,925 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97656.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:40:52,680 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97674.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 04:40:53,268 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.208e+02 2.182e+02 2.726e+02 3.211e+02 6.785e+02, threshold=5.451e+02, percent-clipped=1.0 2022-12-08 04:41:13,564 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97698.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 04:41:17,865 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97703.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:41:18,592 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=97704.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:41:19,117 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.64 vs. limit=2.0 2022-12-08 04:41:41,318 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97730.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:41:45,646 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97735.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 04:41:53,812 INFO [train.py:873] (3/4) Epoch 13, batch 7000, loss[loss=0.1261, simple_loss=0.1578, pruned_loss=0.0472, over 14256.00 frames. ], tot_loss[loss=0.1203, simple_loss=0.1517, pruned_loss=0.0444, over 1942961.42 frames. ], batch size: 60, lr: 5.98e-03, grad_scale: 8.0 2022-12-08 04:41:57,604 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9406, 1.7361, 4.1175, 3.7739, 3.8579, 4.2124, 3.6569, 4.1656], device='cuda:3'), covar=tensor([0.1539, 0.1499, 0.0114, 0.0245, 0.0227, 0.0124, 0.0246, 0.0128], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0159, 0.0129, 0.0169, 0.0146, 0.0141, 0.0123, 0.0120], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 04:42:20,482 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.837e+01 2.234e+02 2.676e+02 3.367e+02 8.383e+02, threshold=5.352e+02, percent-clipped=3.0 2022-12-08 04:43:01,885 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.98 vs. limit=5.0 2022-12-08 04:43:22,739 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9371, 3.4242, 2.6828, 4.1889, 3.8899, 3.9608, 3.3845, 2.8654], device='cuda:3'), covar=tensor([0.0860, 0.1463, 0.3516, 0.0531, 0.0976, 0.1154, 0.1280, 0.3375], device='cuda:3'), in_proj_covar=tensor([0.0273, 0.0292, 0.0266, 0.0271, 0.0313, 0.0296, 0.0255, 0.0246], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 04:43:23,310 INFO [train.py:873] (3/4) Epoch 13, batch 7100, loss[loss=0.1335, simple_loss=0.1669, pruned_loss=0.05003, over 14386.00 frames. ], tot_loss[loss=0.1198, simple_loss=0.1516, pruned_loss=0.044, over 1943161.57 frames. ], batch size: 55, lr: 5.98e-03, grad_scale: 8.0 2022-12-08 04:43:37,102 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0417, 2.0431, 2.0941, 2.0773, 2.0279, 1.6673, 1.3690, 1.8528], device='cuda:3'), covar=tensor([0.0547, 0.0498, 0.0443, 0.0383, 0.0420, 0.1359, 0.2065, 0.0433], device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0171, 0.0143, 0.0141, 0.0200, 0.0135, 0.0158, 0.0189], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 04:43:49,282 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.222e+02 2.322e+02 2.890e+02 3.385e+02 9.839e+02, threshold=5.780e+02, percent-clipped=7.0 2022-12-08 04:44:01,130 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97888.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:44:20,320 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8238, 1.2773, 2.8174, 2.5276, 2.7086, 2.8298, 2.0327, 2.8167], device='cuda:3'), covar=tensor([0.1027, 0.1396, 0.0166, 0.0363, 0.0325, 0.0178, 0.0475, 0.0178], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0159, 0.0129, 0.0169, 0.0146, 0.0140, 0.0122, 0.0120], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 04:44:21,102 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97911.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:44:26,074 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97916.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:44:51,431 INFO [train.py:873] (3/4) Epoch 13, batch 7200, loss[loss=0.1205, simple_loss=0.1502, pruned_loss=0.04546, over 14163.00 frames. ], tot_loss[loss=0.1203, simple_loss=0.152, pruned_loss=0.04427, over 1957368.41 frames. ], batch size: 99, lr: 5.97e-03, grad_scale: 8.0 2022-12-08 04:44:54,831 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97949.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:45:08,592 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=97964.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:45:18,186 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.225e+02 2.164e+02 2.816e+02 3.534e+02 1.223e+03, threshold=5.633e+02, percent-clipped=3.0 2022-12-08 04:45:38,812 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97998.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:45:43,504 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98003.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:45:46,824 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2022-12-08 04:45:48,167 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1795, 2.0965, 2.2634, 2.0594, 2.1106, 1.4535, 1.8178, 2.3008], device='cuda:3'), covar=tensor([0.0778, 0.0670, 0.0482, 0.1638, 0.1614, 0.0676, 0.0808, 0.0661], device='cuda:3'), in_proj_covar=tensor([0.0030, 0.0029, 0.0031, 0.0028, 0.0030, 0.0042, 0.0031, 0.0032], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 04:46:05,753 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.12 vs. limit=2.0 2022-12-08 04:46:07,190 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98030.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 04:46:07,214 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98030.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:46:11,499 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4635, 3.7499, 3.3596, 3.9251, 2.6170, 3.6487, 3.5477, 2.0966], device='cuda:3'), covar=tensor([0.1493, 0.0855, 0.1471, 0.0286, 0.1017, 0.0555, 0.0850, 0.2162], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0084, 0.0066, 0.0069, 0.0097, 0.0082, 0.0097, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0006, 0.0006], device='cuda:3') 2022-12-08 04:46:20,636 INFO [train.py:873] (3/4) Epoch 13, batch 7300, loss[loss=0.1254, simple_loss=0.1537, pruned_loss=0.04854, over 5991.00 frames. ], tot_loss[loss=0.1181, simple_loss=0.1508, pruned_loss=0.04276, over 1912333.54 frames. ], batch size: 100, lr: 5.97e-03, grad_scale: 8.0 2022-12-08 04:46:21,868 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=98046.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:46:26,104 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=98051.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:46:46,648 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.423e+02 2.194e+02 2.668e+02 3.260e+02 4.997e+02, threshold=5.336e+02, percent-clipped=0.0 2022-12-08 04:46:50,053 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=98078.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:46:52,656 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6321, 1.7114, 1.8883, 1.3356, 1.2774, 1.7236, 1.0674, 1.6477], device='cuda:3'), covar=tensor([0.1353, 0.2147, 0.0866, 0.2190, 0.2647, 0.1052, 0.2785, 0.1068], device='cuda:3'), in_proj_covar=tensor([0.0085, 0.0101, 0.0092, 0.0100, 0.0116, 0.0087, 0.0123, 0.0091], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 04:47:03,604 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.00 vs. limit=5.0 2022-12-08 04:47:14,628 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98106.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:47:15,412 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2519, 2.1124, 2.4721, 1.5435, 1.7199, 2.3016, 1.2734, 2.1830], device='cuda:3'), covar=tensor([0.0904, 0.1459, 0.0720, 0.2380, 0.2672, 0.1045, 0.3769, 0.0933], device='cuda:3'), in_proj_covar=tensor([0.0085, 0.0101, 0.0092, 0.0100, 0.0116, 0.0087, 0.0123, 0.0091], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 04:47:44,676 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.0853, 2.6923, 5.1165, 3.5165, 4.8456, 2.3379, 3.9780, 4.9198], device='cuda:3'), covar=tensor([0.0410, 0.3533, 0.0264, 0.5630, 0.0486, 0.3132, 0.0941, 0.0301], device='cuda:3'), in_proj_covar=tensor([0.0252, 0.0209, 0.0210, 0.0283, 0.0227, 0.0210, 0.0207, 0.0212], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 04:47:48,798 INFO [train.py:873] (3/4) Epoch 13, batch 7400, loss[loss=0.1056, simple_loss=0.1471, pruned_loss=0.03206, over 14209.00 frames. ], tot_loss[loss=0.1196, simple_loss=0.1519, pruned_loss=0.04367, over 1961296.62 frames. ], batch size: 84, lr: 5.97e-03, grad_scale: 8.0 2022-12-08 04:48:08,643 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98167.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:48:15,367 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.067e+02 1.931e+02 2.491e+02 3.118e+02 6.123e+02, threshold=4.982e+02, percent-clipped=1.0 2022-12-08 04:48:48,154 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98211.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:49:16,927 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98244.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:49:17,647 INFO [train.py:873] (3/4) Epoch 13, batch 7500, loss[loss=0.1055, simple_loss=0.1476, pruned_loss=0.0317, over 14248.00 frames. ], tot_loss[loss=0.1181, simple_loss=0.1509, pruned_loss=0.04262, over 2009987.10 frames. ], batch size: 57, lr: 5.96e-03, grad_scale: 4.0 2022-12-08 04:49:29,616 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=98259.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:49:44,017 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.582e+02 2.297e+02 2.932e+02 3.500e+02 6.688e+02, threshold=5.864e+02, percent-clipped=4.0 2022-12-08 04:50:44,948 INFO [train.py:873] (3/4) Epoch 14, batch 0, loss[loss=0.1133, simple_loss=0.1631, pruned_loss=0.03175, over 13995.00 frames. ], tot_loss[loss=0.1133, simple_loss=0.1631, pruned_loss=0.03175, over 13995.00 frames. ], batch size: 22, lr: 5.75e-03, grad_scale: 8.0 2022-12-08 04:50:44,948 INFO [train.py:896] (3/4) Computing validation loss 2022-12-08 04:50:52,247 INFO [train.py:905] (3/4) Epoch 14, validation: loss=0.1389, simple_loss=0.1805, pruned_loss=0.04866, over 857387.00 frames. 2022-12-08 04:50:52,248 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-08 04:51:13,018 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98330.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 04:51:17,075 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.58 vs. limit=2.0 2022-12-08 04:51:53,529 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 5.725e+01 2.112e+02 3.078e+02 4.479e+02 1.243e+03, threshold=6.157e+02, percent-clipped=15.0 2022-12-08 04:51:55,461 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=98378.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 04:51:58,971 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6033, 3.4139, 3.2524, 3.6563, 3.4113, 3.6439, 3.7040, 3.0374], device='cuda:3'), covar=tensor([0.0523, 0.0997, 0.0528, 0.0509, 0.0938, 0.0359, 0.0558, 0.0655], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0267, 0.0190, 0.0186, 0.0183, 0.0151, 0.0272, 0.0163], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 04:51:59,069 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.6422, 1.6949, 2.9320, 2.0846, 2.7993, 1.6947, 2.2168, 2.7328], device='cuda:3'), covar=tensor([0.1158, 0.4442, 0.0738, 0.4175, 0.0940, 0.3409, 0.1408, 0.0747], device='cuda:3'), in_proj_covar=tensor([0.0251, 0.0208, 0.0211, 0.0282, 0.0226, 0.0208, 0.0206, 0.0211], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 04:52:20,897 INFO [train.py:873] (3/4) Epoch 14, batch 100, loss[loss=0.1178, simple_loss=0.1419, pruned_loss=0.04685, over 6005.00 frames. ], tot_loss[loss=0.1191, simple_loss=0.1521, pruned_loss=0.04302, over 826430.64 frames. ], batch size: 100, lr: 5.74e-03, grad_scale: 8.0 2022-12-08 04:52:24,334 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98410.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:52:39,055 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7049, 2.1952, 3.7500, 3.9439, 3.9304, 2.1481, 3.9541, 2.7583], device='cuda:3'), covar=tensor([0.0468, 0.1114, 0.0943, 0.0460, 0.0413, 0.1744, 0.0358, 0.1140], device='cuda:3'), in_proj_covar=tensor([0.0290, 0.0255, 0.0369, 0.0325, 0.0267, 0.0298, 0.0304, 0.0278], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 04:52:46,565 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6442, 1.2319, 1.6240, 1.2039, 1.3859, 1.7070, 1.3601, 1.3853], device='cuda:3'), covar=tensor([0.0767, 0.0979, 0.0712, 0.0994, 0.1245, 0.0776, 0.0715, 0.1724], device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0176, 0.0137, 0.0125, 0.0138, 0.0148, 0.0124, 0.0139], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0005, 0.0006], device='cuda:3') 2022-12-08 04:53:10,385 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98462.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:53:18,362 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98471.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:53:22,339 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.180e+02 2.038e+02 2.657e+02 3.395e+02 7.955e+02, threshold=5.314e+02, percent-clipped=1.0 2022-12-08 04:53:49,438 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.06 vs. limit=5.0 2022-12-08 04:53:49,905 INFO [train.py:873] (3/4) Epoch 14, batch 200, loss[loss=0.1751, simple_loss=0.1896, pruned_loss=0.08032, over 8626.00 frames. ], tot_loss[loss=0.1191, simple_loss=0.1515, pruned_loss=0.04339, over 1245620.50 frames. ], batch size: 100, lr: 5.74e-03, grad_scale: 8.0 2022-12-08 04:53:50,087 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4318, 3.1067, 3.0092, 2.2101, 2.8701, 3.1168, 3.3585, 2.6828], device='cuda:3'), covar=tensor([0.0696, 0.1268, 0.0991, 0.1418, 0.1006, 0.0827, 0.0766, 0.1291], device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0176, 0.0137, 0.0125, 0.0138, 0.0148, 0.0125, 0.0139], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0005, 0.0006], device='cuda:3') 2022-12-08 04:53:56,845 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1220, 2.4181, 4.1051, 4.2185, 4.1608, 2.4893, 4.2407, 3.3115], device='cuda:3'), covar=tensor([0.0353, 0.1032, 0.0795, 0.0450, 0.0396, 0.1573, 0.0387, 0.0842], device='cuda:3'), in_proj_covar=tensor([0.0292, 0.0256, 0.0371, 0.0328, 0.0268, 0.0300, 0.0306, 0.0279], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 04:54:22,234 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98543.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:54:22,939 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98544.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:54:50,566 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.342e+02 2.243e+02 2.774e+02 3.161e+02 6.652e+02, threshold=5.548e+02, percent-clipped=3.0 2022-12-08 04:55:03,806 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=98592.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:55:12,132 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3931, 2.7487, 4.2759, 3.1622, 4.2087, 4.0342, 3.9307, 3.5827], device='cuda:3'), covar=tensor([0.0986, 0.3049, 0.0801, 0.1859, 0.0817, 0.0873, 0.1498, 0.1860], device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0314, 0.0392, 0.0303, 0.0372, 0.0321, 0.0363, 0.0304], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 04:55:14,590 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98604.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:55:16,906 INFO [train.py:873] (3/4) Epoch 14, batch 300, loss[loss=0.0969, simple_loss=0.1268, pruned_loss=0.03349, over 4951.00 frames. ], tot_loss[loss=0.1198, simple_loss=0.1517, pruned_loss=0.04391, over 1526481.67 frames. ], batch size: 100, lr: 5.74e-03, grad_scale: 8.0 2022-12-08 04:55:36,596 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.2194, 2.6980, 4.0666, 3.0323, 4.0320, 3.8829, 3.7595, 3.3938], device='cuda:3'), covar=tensor([0.0829, 0.2874, 0.0786, 0.1777, 0.0821, 0.0914, 0.1736, 0.1953], device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0314, 0.0391, 0.0302, 0.0371, 0.0319, 0.0362, 0.0303], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 04:56:18,354 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.000e+02 2.234e+02 2.680e+02 3.385e+02 6.957e+02, threshold=5.360e+02, percent-clipped=3.0 2022-12-08 04:56:39,764 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.0176, 2.8010, 4.9774, 3.6238, 4.7020, 2.4085, 3.7908, 4.7728], device='cuda:3'), covar=tensor([0.0348, 0.3214, 0.0443, 0.5375, 0.0427, 0.2997, 0.1085, 0.0384], device='cuda:3'), in_proj_covar=tensor([0.0251, 0.0209, 0.0211, 0.0281, 0.0227, 0.0210, 0.0207, 0.0210], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 04:56:45,719 INFO [train.py:873] (3/4) Epoch 14, batch 400, loss[loss=0.151, simple_loss=0.1744, pruned_loss=0.06376, over 7792.00 frames. ], tot_loss[loss=0.1196, simple_loss=0.1514, pruned_loss=0.04389, over 1690551.52 frames. ], batch size: 100, lr: 5.73e-03, grad_scale: 8.0 2022-12-08 04:56:53,845 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.7366, 2.4543, 4.6791, 3.2818, 4.3883, 2.1251, 3.5021, 4.4521], device='cuda:3'), covar=tensor([0.0477, 0.4172, 0.0410, 0.6814, 0.0664, 0.3790, 0.1267, 0.0425], device='cuda:3'), in_proj_covar=tensor([0.0252, 0.0210, 0.0212, 0.0283, 0.0228, 0.0211, 0.0208, 0.0211], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 04:57:01,295 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.1644, 4.8743, 4.4425, 4.6957, 4.7270, 5.0666, 5.1377, 5.0964], device='cuda:3'), covar=tensor([0.0685, 0.0438, 0.1996, 0.2823, 0.0719, 0.0692, 0.0756, 0.0801], device='cuda:3'), in_proj_covar=tensor([0.0385, 0.0272, 0.0446, 0.0565, 0.0340, 0.0440, 0.0392, 0.0382], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 04:57:33,729 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98762.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:57:36,934 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98766.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:57:46,468 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.225e+02 2.221e+02 2.679e+02 3.803e+02 8.740e+02, threshold=5.359e+02, percent-clipped=4.0 2022-12-08 04:58:13,696 INFO [train.py:873] (3/4) Epoch 14, batch 500, loss[loss=0.1136, simple_loss=0.151, pruned_loss=0.03815, over 14023.00 frames. ], tot_loss[loss=0.1202, simple_loss=0.1521, pruned_loss=0.04418, over 1753439.27 frames. ], batch size: 22, lr: 5.73e-03, grad_scale: 8.0 2022-12-08 04:58:16,990 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=98810.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:59:14,755 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.381e+02 1.935e+02 2.381e+02 2.894e+02 6.356e+02, threshold=4.763e+02, percent-clipped=3.0 2022-12-08 04:59:27,000 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98890.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:59:31,802 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98895.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:59:35,195 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98899.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 04:59:42,009 INFO [train.py:873] (3/4) Epoch 14, batch 600, loss[loss=0.1149, simple_loss=0.1459, pruned_loss=0.04192, over 11179.00 frames. ], tot_loss[loss=0.1188, simple_loss=0.1511, pruned_loss=0.04327, over 1859636.90 frames. ], batch size: 100, lr: 5.73e-03, grad_scale: 8.0 2022-12-08 04:59:59,932 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.2955, 5.0676, 4.7635, 5.2233, 4.8221, 4.6303, 5.2896, 5.1059], device='cuda:3'), covar=tensor([0.0545, 0.0546, 0.0792, 0.0591, 0.0753, 0.0551, 0.0522, 0.0712], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0136, 0.0144, 0.0155, 0.0143, 0.0119, 0.0163, 0.0143], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 05:00:20,864 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98951.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:00:25,446 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98956.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:00:38,391 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98971.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:00:43,054 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.204e+01 1.954e+02 2.310e+02 3.088e+02 7.406e+02, threshold=4.620e+02, percent-clipped=2.0 2022-12-08 05:01:09,652 INFO [train.py:873] (3/4) Epoch 14, batch 700, loss[loss=0.1277, simple_loss=0.1507, pruned_loss=0.05237, over 5946.00 frames. ], tot_loss[loss=0.117, simple_loss=0.1497, pruned_loss=0.04218, over 1899600.70 frames. ], batch size: 100, lr: 5.73e-03, grad_scale: 8.0 2022-12-08 05:01:22,730 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1128, 2.0272, 1.7786, 1.8624, 2.0389, 2.0855, 2.0739, 2.0609], device='cuda:3'), covar=tensor([0.1339, 0.0921, 0.3295, 0.3376, 0.1505, 0.1313, 0.1833, 0.1216], device='cuda:3'), in_proj_covar=tensor([0.0382, 0.0269, 0.0443, 0.0562, 0.0338, 0.0442, 0.0391, 0.0380], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 05:01:32,141 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99032.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:01:38,156 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9232, 1.9502, 2.1291, 2.0506, 1.9100, 1.8127, 1.5917, 1.3744], device='cuda:3'), covar=tensor([0.0332, 0.0526, 0.0387, 0.0359, 0.0387, 0.0445, 0.0505, 0.0805], device='cuda:3'), in_proj_covar=tensor([0.0018, 0.0019, 0.0017, 0.0018, 0.0018, 0.0030, 0.0024, 0.0029], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 05:01:48,619 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99051.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:02:01,353 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99066.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:02:09,758 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.403e+02 2.198e+02 2.677e+02 3.490e+02 1.171e+03, threshold=5.353e+02, percent-clipped=9.0 2022-12-08 05:02:37,295 INFO [train.py:873] (3/4) Epoch 14, batch 800, loss[loss=0.1196, simple_loss=0.1529, pruned_loss=0.04319, over 14422.00 frames. ], tot_loss[loss=0.1189, simple_loss=0.1508, pruned_loss=0.04348, over 1908064.90 frames. ], batch size: 51, lr: 5.72e-03, grad_scale: 8.0 2022-12-08 05:02:41,528 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99112.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:02:43,048 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99114.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:03:38,207 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.475e+02 2.181e+02 2.713e+02 3.365e+02 5.447e+02, threshold=5.425e+02, percent-clipped=1.0 2022-12-08 05:03:58,120 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99199.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:04:05,016 INFO [train.py:873] (3/4) Epoch 14, batch 900, loss[loss=0.1163, simple_loss=0.1566, pruned_loss=0.03796, over 14295.00 frames. ], tot_loss[loss=0.1189, simple_loss=0.1513, pruned_loss=0.04327, over 1960661.00 frames. ], batch size: 28, lr: 5.72e-03, grad_scale: 8.0 2022-12-08 05:04:24,493 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99229.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:04:39,141 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99246.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:04:39,940 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99247.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:04:43,468 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99251.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:04:44,701 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.9054, 1.1807, 1.0046, 0.9731, 1.1644, 0.6823, 0.8746, 1.0338], device='cuda:3'), covar=tensor([0.0647, 0.0630, 0.0647, 0.0606, 0.0415, 0.0685, 0.1203, 0.0908], device='cuda:3'), in_proj_covar=tensor([0.0030, 0.0030, 0.0033, 0.0028, 0.0030, 0.0043, 0.0032, 0.0034], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 05:04:58,906 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99268.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:05:05,666 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.191e+02 2.212e+02 2.795e+02 3.543e+02 9.051e+02, threshold=5.590e+02, percent-clipped=4.0 2022-12-08 05:05:17,556 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99290.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:05:24,130 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5417, 3.4047, 3.2796, 3.6669, 3.1985, 3.1560, 3.6138, 3.4805], device='cuda:3'), covar=tensor([0.0658, 0.1002, 0.0867, 0.0595, 0.1006, 0.0762, 0.0650, 0.0784], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0137, 0.0145, 0.0155, 0.0143, 0.0119, 0.0163, 0.0143], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 05:05:32,839 INFO [train.py:873] (3/4) Epoch 14, batch 1000, loss[loss=0.1171, simple_loss=0.1525, pruned_loss=0.04082, over 14191.00 frames. ], tot_loss[loss=0.1182, simple_loss=0.1507, pruned_loss=0.0428, over 1995530.92 frames. ], batch size: 94, lr: 5.72e-03, grad_scale: 8.0 2022-12-08 05:05:47,432 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99324.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:05:49,891 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99327.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:05:51,547 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99329.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:06:09,425 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.8074, 5.2482, 5.1642, 5.7578, 5.3557, 4.6895, 5.6794, 4.7628], device='cuda:3'), covar=tensor([0.0319, 0.1076, 0.0319, 0.0374, 0.0705, 0.0450, 0.0476, 0.0465], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0270, 0.0190, 0.0188, 0.0184, 0.0153, 0.0276, 0.0165], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 05:06:22,775 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.2977, 4.1123, 3.9856, 4.3510, 4.0010, 3.6476, 4.3593, 4.1415], device='cuda:3'), covar=tensor([0.0656, 0.0832, 0.0783, 0.0571, 0.0759, 0.0715, 0.0583, 0.0807], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0136, 0.0145, 0.0155, 0.0143, 0.0119, 0.0162, 0.0143], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 05:06:32,477 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.188e+02 2.112e+02 2.804e+02 3.540e+02 1.187e+03, threshold=5.607e+02, percent-clipped=4.0 2022-12-08 05:06:41,121 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99385.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:06:59,882 INFO [train.py:873] (3/4) Epoch 14, batch 1100, loss[loss=0.1553, simple_loss=0.1743, pruned_loss=0.06809, over 7765.00 frames. ], tot_loss[loss=0.1184, simple_loss=0.1508, pruned_loss=0.04298, over 1973385.08 frames. ], batch size: 100, lr: 5.71e-03, grad_scale: 8.0 2022-12-08 05:06:59,957 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99407.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:07:01,156 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8798, 1.4773, 4.0500, 3.8362, 3.8082, 4.1325, 3.6318, 4.1047], device='cuda:3'), covar=tensor([0.1570, 0.1754, 0.0132, 0.0210, 0.0234, 0.0133, 0.0203, 0.0144], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0158, 0.0128, 0.0168, 0.0146, 0.0140, 0.0122, 0.0121], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 05:07:12,279 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99420.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:07:14,946 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2713, 2.0697, 1.9956, 1.6911, 1.8953, 1.9359, 1.7183, 1.8890], device='cuda:3'), covar=tensor([0.0491, 0.0741, 0.0605, 0.0473, 0.0540, 0.0470, 0.0497, 0.0644], device='cuda:3'), in_proj_covar=tensor([0.0019, 0.0019, 0.0017, 0.0018, 0.0018, 0.0029, 0.0024, 0.0029], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 05:07:53,758 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99468.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 05:08:00,247 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.366e+02 2.195e+02 2.512e+02 3.214e+02 5.194e+02, threshold=5.024e+02, percent-clipped=0.0 2022-12-08 05:08:04,834 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99481.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:08:27,692 INFO [train.py:873] (3/4) Epoch 14, batch 1200, loss[loss=0.1758, simple_loss=0.1801, pruned_loss=0.08571, over 7794.00 frames. ], tot_loss[loss=0.1188, simple_loss=0.1513, pruned_loss=0.04315, over 1972731.61 frames. ], batch size: 100, lr: 5.71e-03, grad_scale: 8.0 2022-12-08 05:08:47,074 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99529.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 05:08:47,762 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.3562, 1.3429, 1.5799, 1.2267, 1.2516, 1.2250, 1.1849, 1.0268], device='cuda:3'), covar=tensor([0.0217, 0.0269, 0.0151, 0.0237, 0.0221, 0.0423, 0.0293, 0.0515], device='cuda:3'), in_proj_covar=tensor([0.0019, 0.0019, 0.0017, 0.0018, 0.0018, 0.0029, 0.0024, 0.0029], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 05:09:01,525 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99546.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:09:05,794 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99551.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:09:07,816 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2022-12-08 05:09:20,158 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99568.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:09:28,000 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.218e+02 2.156e+02 2.823e+02 3.405e+02 6.610e+02, threshold=5.645e+02, percent-clipped=3.0 2022-12-08 05:09:35,415 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99585.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:09:38,988 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99589.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:09:43,004 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99594.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:09:47,148 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99599.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:09:54,365 INFO [train.py:873] (3/4) Epoch 14, batch 1300, loss[loss=0.1552, simple_loss=0.1479, pruned_loss=0.08128, over 1168.00 frames. ], tot_loss[loss=0.1192, simple_loss=0.1514, pruned_loss=0.04345, over 1910214.96 frames. ], batch size: 100, lr: 5.71e-03, grad_scale: 4.0 2022-12-08 05:09:54,864 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.43 vs. limit=5.0 2022-12-08 05:10:10,434 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99624.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:10:12,991 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99627.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:10:14,831 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99629.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:10:32,849 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99650.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:10:35,353 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3682, 2.2431, 2.7691, 1.7619, 1.7611, 2.4593, 1.4170, 2.4370], device='cuda:3'), covar=tensor([0.0943, 0.1343, 0.0690, 0.2322, 0.2586, 0.1000, 0.3528, 0.1078], device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0099, 0.0091, 0.0098, 0.0116, 0.0088, 0.0122, 0.0091], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 05:10:40,862 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.63 vs. limit=2.0 2022-12-08 05:10:54,749 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99675.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:10:56,553 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.319e+02 2.185e+02 2.869e+02 3.573e+02 7.828e+02, threshold=5.739e+02, percent-clipped=4.0 2022-12-08 05:10:59,266 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99680.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:10:59,340 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99680.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:11:17,025 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8332, 0.7668, 0.8009, 0.7550, 0.7134, 0.4345, 0.5464, 0.6325], device='cuda:3'), covar=tensor([0.0126, 0.0153, 0.0125, 0.0147, 0.0177, 0.0298, 0.0208, 0.0251], device='cuda:3'), in_proj_covar=tensor([0.0018, 0.0019, 0.0017, 0.0018, 0.0018, 0.0029, 0.0024, 0.0029], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 05:11:17,354 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 2022-12-08 05:11:23,269 INFO [train.py:873] (3/4) Epoch 14, batch 1400, loss[loss=0.1329, simple_loss=0.1467, pruned_loss=0.05949, over 3828.00 frames. ], tot_loss[loss=0.1192, simple_loss=0.1514, pruned_loss=0.04348, over 1888890.63 frames. ], batch size: 100, lr: 5.71e-03, grad_scale: 4.0 2022-12-08 05:11:23,380 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99707.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:11:51,279 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7382, 3.2620, 2.5200, 4.0807, 3.8383, 3.7953, 3.2693, 2.6139], device='cuda:3'), covar=tensor([0.1150, 0.1815, 0.4065, 0.0669, 0.1280, 0.1789, 0.1435, 0.3672], device='cuda:3'), in_proj_covar=tensor([0.0275, 0.0294, 0.0267, 0.0272, 0.0315, 0.0298, 0.0255, 0.0247], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 05:11:52,928 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99741.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:12:05,061 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99755.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:12:23,714 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99776.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:12:24,429 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.204e+02 2.049e+02 2.625e+02 3.154e+02 6.097e+02, threshold=5.251e+02, percent-clipped=2.0 2022-12-08 05:12:50,316 INFO [train.py:873] (3/4) Epoch 14, batch 1500, loss[loss=0.1613, simple_loss=0.148, pruned_loss=0.08728, over 1208.00 frames. ], tot_loss[loss=0.119, simple_loss=0.1512, pruned_loss=0.04338, over 1941007.38 frames. ], batch size: 100, lr: 5.70e-03, grad_scale: 4.0 2022-12-08 05:13:06,041 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99824.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 05:13:50,607 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2022-12-08 05:13:51,975 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.709e+01 2.163e+02 2.561e+02 3.336e+02 7.134e+02, threshold=5.122e+02, percent-clipped=2.0 2022-12-08 05:13:57,168 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0293, 3.5780, 2.7305, 4.2037, 3.9739, 4.0234, 3.4923, 2.7530], device='cuda:3'), covar=tensor([0.0662, 0.1281, 0.3512, 0.0604, 0.0968, 0.1236, 0.1224, 0.3334], device='cuda:3'), in_proj_covar=tensor([0.0270, 0.0287, 0.0261, 0.0267, 0.0310, 0.0292, 0.0251, 0.0241], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:3') 2022-12-08 05:13:58,844 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99885.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:14:18,360 INFO [train.py:873] (3/4) Epoch 14, batch 1600, loss[loss=0.1297, simple_loss=0.1393, pruned_loss=0.06009, over 2615.00 frames. ], tot_loss[loss=0.1184, simple_loss=0.1504, pruned_loss=0.04315, over 1908024.48 frames. ], batch size: 100, lr: 5.70e-03, grad_scale: 8.0 2022-12-08 05:14:21,856 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7694, 3.9250, 4.0887, 3.5531, 3.9269, 3.8956, 1.6202, 3.7496], device='cuda:3'), covar=tensor([0.0339, 0.0348, 0.0351, 0.0424, 0.0328, 0.0370, 0.3014, 0.0284], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0173, 0.0144, 0.0141, 0.0203, 0.0137, 0.0159, 0.0191], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 05:14:32,781 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99924.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:14:32,828 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99924.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:14:40,526 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99933.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:14:46,157 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.2248, 4.7909, 4.7146, 5.2257, 4.8213, 4.4745, 5.1152, 4.3922], device='cuda:3'), covar=tensor([0.0301, 0.0918, 0.0366, 0.0351, 0.0700, 0.0507, 0.0511, 0.0481], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0272, 0.0192, 0.0192, 0.0184, 0.0155, 0.0281, 0.0168], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 05:14:52,030 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99945.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:15:04,196 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.0013, 1.0727, 1.1458, 0.9329, 0.9516, 0.7342, 0.8987, 0.8941], device='cuda:3'), covar=tensor([0.0227, 0.0201, 0.0179, 0.0261, 0.0224, 0.0417, 0.0303, 0.0425], device='cuda:3'), in_proj_covar=tensor([0.0019, 0.0019, 0.0017, 0.0018, 0.0018, 0.0029, 0.0024, 0.0029], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 05:15:06,021 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0732, 2.6728, 2.8530, 1.8561, 2.5070, 2.6981, 2.9568, 2.4590], device='cuda:3'), covar=tensor([0.0684, 0.0949, 0.0851, 0.1637, 0.1027, 0.0880, 0.0740, 0.1292], device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0174, 0.0136, 0.0125, 0.0138, 0.0149, 0.0127, 0.0141], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0005, 0.0006], device='cuda:3') 2022-12-08 05:15:15,694 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99972.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:15:20,185 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.257e+02 2.180e+02 2.741e+02 3.315e+02 6.557e+02, threshold=5.482e+02, percent-clipped=1.0 2022-12-08 05:15:23,150 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99980.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:15:50,678 INFO [train.py:873] (3/4) Epoch 14, batch 1700, loss[loss=0.1234, simple_loss=0.1534, pruned_loss=0.04667, over 9487.00 frames. ], tot_loss[loss=0.1182, simple_loss=0.1503, pruned_loss=0.043, over 1898235.53 frames. ], batch size: 100, lr: 5.70e-03, grad_scale: 8.0 2022-12-08 05:15:50,861 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100007.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 05:16:09,960 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=100028.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:16:16,911 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100036.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:16:23,003 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7374, 1.5502, 1.9508, 1.7064, 1.8930, 1.0077, 1.4887, 1.6501], device='cuda:3'), covar=tensor([0.0908, 0.0903, 0.0478, 0.0953, 0.0759, 0.1121, 0.1002, 0.0774], device='cuda:3'), in_proj_covar=tensor([0.0030, 0.0030, 0.0033, 0.0028, 0.0030, 0.0043, 0.0031, 0.0033], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 05:16:41,774 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.72 vs. limit=2.0 2022-12-08 05:16:45,292 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100068.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 05:16:51,795 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100076.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:16:52,530 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.360e+02 2.197e+02 2.632e+02 3.389e+02 8.054e+02, threshold=5.264e+02, percent-clipped=5.0 2022-12-08 05:17:04,377 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8208, 1.8655, 1.6196, 1.9391, 1.8639, 1.8465, 1.8391, 1.6827], device='cuda:3'), covar=tensor([0.1183, 0.0870, 0.1976, 0.0592, 0.0809, 0.0620, 0.1332, 0.0911], device='cuda:3'), in_proj_covar=tensor([0.0269, 0.0288, 0.0260, 0.0265, 0.0311, 0.0292, 0.0251, 0.0240], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:3') 2022-12-08 05:17:19,680 INFO [train.py:873] (3/4) Epoch 14, batch 1800, loss[loss=0.117, simple_loss=0.1497, pruned_loss=0.04211, over 14221.00 frames. ], tot_loss[loss=0.1185, simple_loss=0.1501, pruned_loss=0.04345, over 1856053.37 frames. ], batch size: 35, lr: 5.69e-03, grad_scale: 8.0 2022-12-08 05:17:32,079 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9791, 3.3185, 3.0463, 3.2017, 2.5374, 3.3005, 3.0614, 1.6991], device='cuda:3'), covar=tensor([0.1384, 0.0604, 0.0778, 0.0669, 0.0987, 0.0504, 0.0906, 0.2103], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0084, 0.0067, 0.0071, 0.0098, 0.0084, 0.0098, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0006, 0.0006], device='cuda:3') 2022-12-08 05:17:34,616 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=100124.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:17:34,654 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0221, 2.0585, 2.0771, 2.0923, 2.0012, 1.6847, 1.2925, 1.8108], device='cuda:3'), covar=tensor([0.0587, 0.0497, 0.0553, 0.0348, 0.0430, 0.1318, 0.2332, 0.0451], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0172, 0.0144, 0.0141, 0.0203, 0.0137, 0.0159, 0.0190], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 05:17:34,673 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100124.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 05:18:17,538 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=100172.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 05:18:21,977 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.282e+02 2.047e+02 2.482e+02 3.140e+02 5.159e+02, threshold=4.965e+02, percent-clipped=0.0 2022-12-08 05:18:48,336 INFO [train.py:873] (3/4) Epoch 14, batch 1900, loss[loss=0.1226, simple_loss=0.1574, pruned_loss=0.04393, over 14566.00 frames. ], tot_loss[loss=0.1196, simple_loss=0.1509, pruned_loss=0.04416, over 1878518.75 frames. ], batch size: 23, lr: 5.69e-03, grad_scale: 8.0 2022-12-08 05:18:52,832 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4384, 2.4164, 3.4202, 3.5958, 3.4615, 2.2823, 3.4887, 2.7461], device='cuda:3'), covar=tensor([0.0422, 0.0910, 0.0766, 0.0520, 0.0426, 0.1427, 0.0390, 0.0888], device='cuda:3'), in_proj_covar=tensor([0.0290, 0.0256, 0.0374, 0.0328, 0.0265, 0.0302, 0.0307, 0.0282], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 05:19:03,344 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100224.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:19:13,775 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8295, 1.4439, 1.8040, 1.3323, 1.5447, 1.9157, 1.6034, 1.6605], device='cuda:3'), covar=tensor([0.0934, 0.0837, 0.0873, 0.1475, 0.1523, 0.0932, 0.0740, 0.1564], device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0174, 0.0136, 0.0125, 0.0138, 0.0149, 0.0126, 0.0140], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0005, 0.0006], device='cuda:3') 2022-12-08 05:19:15,362 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100238.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:19:21,284 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100245.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:19:23,855 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8646, 1.6626, 1.9187, 1.7774, 1.8738, 1.2341, 1.7031, 1.7164], device='cuda:3'), covar=tensor([0.0797, 0.0739, 0.0823, 0.0907, 0.0827, 0.0912, 0.0711, 0.0920], device='cuda:3'), in_proj_covar=tensor([0.0030, 0.0030, 0.0033, 0.0029, 0.0030, 0.0044, 0.0031, 0.0034], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 05:19:43,705 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.5057, 1.1271, 2.0296, 1.8526, 1.8731, 2.0766, 1.5030, 2.0247], device='cuda:3'), covar=tensor([0.0854, 0.1431, 0.0269, 0.0512, 0.0657, 0.0283, 0.0690, 0.0332], device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0160, 0.0128, 0.0168, 0.0147, 0.0139, 0.0123, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 05:19:44,562 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2467, 2.9943, 2.9599, 3.2538, 3.1154, 3.2335, 3.2936, 2.7069], device='cuda:3'), covar=tensor([0.0670, 0.1122, 0.0654, 0.0603, 0.0877, 0.0475, 0.0629, 0.0728], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0270, 0.0191, 0.0190, 0.0182, 0.0152, 0.0279, 0.0165], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 05:19:45,381 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=100272.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:19:49,479 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.349e+02 2.314e+02 2.818e+02 3.736e+02 1.298e+03, threshold=5.635e+02, percent-clipped=11.0 2022-12-08 05:20:03,309 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=100293.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:20:08,748 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100299.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:20:15,748 INFO [train.py:873] (3/4) Epoch 14, batch 2000, loss[loss=0.09677, simple_loss=0.1282, pruned_loss=0.03267, over 5991.00 frames. ], tot_loss[loss=0.1195, simple_loss=0.151, pruned_loss=0.04404, over 1875165.14 frames. ], batch size: 100, lr: 5.69e-03, grad_scale: 8.0 2022-12-08 05:20:24,934 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5134, 3.2429, 3.2303, 3.4919, 3.3601, 3.4784, 3.5567, 2.9707], device='cuda:3'), covar=tensor([0.0454, 0.1122, 0.0487, 0.0507, 0.0718, 0.0360, 0.0535, 0.0577], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0269, 0.0189, 0.0188, 0.0181, 0.0152, 0.0276, 0.0164], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 05:20:34,858 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 2022-12-08 05:20:41,583 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100336.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:21:05,113 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100363.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 05:21:17,436 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.095e+02 1.945e+02 2.468e+02 3.146e+02 6.937e+02, threshold=4.935e+02, percent-clipped=2.0 2022-12-08 05:21:23,776 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=100384.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:21:27,338 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5019, 4.6471, 4.9542, 4.0536, 4.6561, 5.0091, 1.8408, 4.4478], device='cuda:3'), covar=tensor([0.0296, 0.0268, 0.0314, 0.0454, 0.0326, 0.0118, 0.3078, 0.0247], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0172, 0.0144, 0.0141, 0.0203, 0.0137, 0.0158, 0.0190], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 05:21:34,363 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7496, 2.0764, 3.8231, 2.6488, 3.6851, 1.9175, 2.8812, 3.6734], device='cuda:3'), covar=tensor([0.0698, 0.4233, 0.0575, 0.5556, 0.0680, 0.3657, 0.1437, 0.0462], device='cuda:3'), in_proj_covar=tensor([0.0254, 0.0206, 0.0211, 0.0279, 0.0229, 0.0208, 0.0207, 0.0212], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 05:21:38,626 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2022-12-08 05:21:39,964 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100402.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:21:44,077 INFO [train.py:873] (3/4) Epoch 14, batch 2100, loss[loss=0.1232, simple_loss=0.1565, pruned_loss=0.045, over 14528.00 frames. ], tot_loss[loss=0.1193, simple_loss=0.1511, pruned_loss=0.04377, over 1924098.41 frames. ], batch size: 49, lr: 5.69e-03, grad_scale: 8.0 2022-12-08 05:22:19,502 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1271, 2.0366, 1.7878, 1.8994, 2.1009, 2.1155, 2.1047, 2.0533], device='cuda:3'), covar=tensor([0.1130, 0.0811, 0.2397, 0.2362, 0.1130, 0.1029, 0.1279, 0.1200], device='cuda:3'), in_proj_covar=tensor([0.0382, 0.0269, 0.0441, 0.0557, 0.0339, 0.0437, 0.0381, 0.0380], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 05:22:34,519 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100463.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 05:22:46,710 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.244e+02 2.182e+02 2.587e+02 3.120e+02 6.565e+02, threshold=5.173e+02, percent-clipped=5.0 2022-12-08 05:22:50,843 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.63 vs. limit=2.0 2022-12-08 05:23:10,001 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.1097, 4.8927, 4.7102, 5.1205, 4.6648, 4.6019, 5.1736, 4.9854], device='cuda:3'), covar=tensor([0.0622, 0.0821, 0.0673, 0.0575, 0.0751, 0.0404, 0.0530, 0.0664], device='cuda:3'), in_proj_covar=tensor([0.0137, 0.0135, 0.0140, 0.0152, 0.0141, 0.0117, 0.0160, 0.0141], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 05:23:13,308 INFO [train.py:873] (3/4) Epoch 14, batch 2200, loss[loss=0.1109, simple_loss=0.1428, pruned_loss=0.03952, over 13919.00 frames. ], tot_loss[loss=0.1192, simple_loss=0.1512, pruned_loss=0.0436, over 1955278.93 frames. ], batch size: 23, lr: 5.68e-03, grad_scale: 8.0 2022-12-08 05:23:47,540 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9981, 3.7434, 3.4052, 2.7963, 3.3932, 3.7216, 3.9477, 3.3961], device='cuda:3'), covar=tensor([0.0454, 0.1199, 0.0812, 0.1253, 0.0726, 0.0541, 0.0771, 0.0880], device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0174, 0.0136, 0.0125, 0.0138, 0.0149, 0.0126, 0.0139], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0005, 0.0006], device='cuda:3') 2022-12-08 05:24:15,306 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.135e+02 2.000e+02 2.520e+02 3.271e+02 5.030e+02, threshold=5.040e+02, percent-clipped=0.0 2022-12-08 05:24:19,775 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2022-12-08 05:24:30,033 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100594.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:24:37,699 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.21 vs. limit=5.0 2022-12-08 05:24:41,408 INFO [train.py:873] (3/4) Epoch 14, batch 2300, loss[loss=0.1137, simple_loss=0.1508, pruned_loss=0.03827, over 14287.00 frames. ], tot_loss[loss=0.1174, simple_loss=0.1502, pruned_loss=0.04231, over 1999617.51 frames. ], batch size: 25, lr: 5.68e-03, grad_scale: 4.0 2022-12-08 05:25:30,962 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100663.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 05:25:44,076 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.238e+02 1.999e+02 2.526e+02 3.011e+02 4.450e+02, threshold=5.052e+02, percent-clipped=0.0 2022-12-08 05:25:58,262 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1356, 2.0492, 1.7928, 1.8652, 2.0887, 2.1092, 2.0362, 2.0749], device='cuda:3'), covar=tensor([0.1126, 0.0873, 0.3004, 0.2885, 0.1314, 0.1198, 0.1891, 0.1239], device='cuda:3'), in_proj_covar=tensor([0.0381, 0.0271, 0.0442, 0.0564, 0.0341, 0.0442, 0.0386, 0.0381], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 05:26:02,960 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2532, 1.4554, 3.2876, 1.5025, 3.1387, 3.3838, 2.3538, 3.5844], device='cuda:3'), covar=tensor([0.0260, 0.2913, 0.0451, 0.2298, 0.0972, 0.0399, 0.1047, 0.0220], device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0156, 0.0161, 0.0169, 0.0170, 0.0178, 0.0133, 0.0149], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 05:26:05,603 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8967, 1.8245, 2.0539, 2.0076, 2.0867, 1.1616, 1.7012, 1.8469], device='cuda:3'), covar=tensor([0.0708, 0.0684, 0.0526, 0.1122, 0.0693, 0.0904, 0.0847, 0.0673], device='cuda:3'), in_proj_covar=tensor([0.0030, 0.0030, 0.0033, 0.0028, 0.0030, 0.0044, 0.0031, 0.0033], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 05:26:09,987 INFO [train.py:873] (3/4) Epoch 14, batch 2400, loss[loss=0.1128, simple_loss=0.1527, pruned_loss=0.03646, over 14064.00 frames. ], tot_loss[loss=0.1179, simple_loss=0.1507, pruned_loss=0.04259, over 2000128.69 frames. ], batch size: 29, lr: 5.68e-03, grad_scale: 8.0 2022-12-08 05:26:13,582 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=100711.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 05:26:30,731 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100730.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 05:26:55,110 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100758.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 05:27:13,001 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.419e+02 2.043e+02 2.633e+02 3.365e+02 7.380e+02, threshold=5.267e+02, percent-clipped=6.0 2022-12-08 05:27:24,748 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100791.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 05:27:38,942 INFO [train.py:873] (3/4) Epoch 14, batch 2500, loss[loss=0.1607, simple_loss=0.1769, pruned_loss=0.07229, over 7814.00 frames. ], tot_loss[loss=0.1172, simple_loss=0.1503, pruned_loss=0.04202, over 2020301.83 frames. ], batch size: 100, lr: 5.67e-03, grad_scale: 8.0 2022-12-08 05:28:15,946 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3766, 2.3635, 1.9282, 2.4095, 2.2452, 2.2743, 2.1728, 2.0392], device='cuda:3'), covar=tensor([0.1046, 0.0932, 0.2563, 0.0814, 0.1218, 0.0764, 0.1341, 0.1319], device='cuda:3'), in_proj_covar=tensor([0.0276, 0.0297, 0.0267, 0.0273, 0.0321, 0.0299, 0.0258, 0.0248], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 05:28:41,717 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.248e+02 2.248e+02 2.659e+02 3.410e+02 6.665e+02, threshold=5.318e+02, percent-clipped=2.0 2022-12-08 05:28:55,567 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100894.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:29:06,707 INFO [train.py:873] (3/4) Epoch 14, batch 2600, loss[loss=0.1359, simple_loss=0.1663, pruned_loss=0.05274, over 11181.00 frames. ], tot_loss[loss=0.1178, simple_loss=0.1507, pruned_loss=0.04245, over 2014538.57 frames. ], batch size: 100, lr: 5.67e-03, grad_scale: 8.0 2022-12-08 05:29:30,560 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.2456, 2.1078, 4.2820, 2.9394, 4.0826, 2.0253, 3.1430, 4.0593], device='cuda:3'), covar=tensor([0.0580, 0.4470, 0.0405, 0.6533, 0.0555, 0.3541, 0.1405, 0.0390], device='cuda:3'), in_proj_covar=tensor([0.0255, 0.0210, 0.0213, 0.0283, 0.0232, 0.0212, 0.0210, 0.0215], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 05:29:37,200 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=100942.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:30:08,943 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.168e+02 2.110e+02 2.658e+02 3.162e+02 7.088e+02, threshold=5.316e+02, percent-clipped=3.0 2022-12-08 05:30:34,935 INFO [train.py:873] (3/4) Epoch 14, batch 2700, loss[loss=0.1627, simple_loss=0.1556, pruned_loss=0.08487, over 1267.00 frames. ], tot_loss[loss=0.1176, simple_loss=0.1507, pruned_loss=0.04228, over 2038377.24 frames. ], batch size: 100, lr: 5.67e-03, grad_scale: 8.0 2022-12-08 05:31:16,735 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.62 vs. limit=2.0 2022-12-08 05:31:20,847 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=101058.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:31:38,598 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.153e+02 2.205e+02 2.633e+02 3.607e+02 6.784e+02, threshold=5.265e+02, percent-clipped=5.0 2022-12-08 05:31:45,723 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=101086.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 05:32:03,773 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=101106.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:32:04,578 INFO [train.py:873] (3/4) Epoch 14, batch 2800, loss[loss=0.1358, simple_loss=0.1253, pruned_loss=0.07321, over 1288.00 frames. ], tot_loss[loss=0.1172, simple_loss=0.1504, pruned_loss=0.04202, over 2032062.29 frames. ], batch size: 100, lr: 5.67e-03, grad_scale: 8.0 2022-12-08 05:32:04,781 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=101107.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:32:26,299 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.80 vs. limit=2.0 2022-12-08 05:32:58,750 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=101168.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:33:07,500 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.225e+02 2.080e+02 2.537e+02 3.233e+02 5.412e+02, threshold=5.073e+02, percent-clipped=2.0 2022-12-08 05:33:33,024 INFO [train.py:873] (3/4) Epoch 14, batch 2900, loss[loss=0.1047, simple_loss=0.1458, pruned_loss=0.03183, over 14486.00 frames. ], tot_loss[loss=0.1171, simple_loss=0.1504, pruned_loss=0.04187, over 2109692.44 frames. ], batch size: 34, lr: 5.66e-03, grad_scale: 4.0 2022-12-08 05:33:42,662 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2022-12-08 05:34:36,775 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.297e+02 2.145e+02 2.615e+02 3.247e+02 7.058e+02, threshold=5.230e+02, percent-clipped=3.0 2022-12-08 05:34:39,183 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3563, 3.8767, 2.9859, 4.6166, 4.3107, 4.4301, 3.6323, 3.1540], device='cuda:3'), covar=tensor([0.0722, 0.1101, 0.3426, 0.0641, 0.0856, 0.1227, 0.1318, 0.2931], device='cuda:3'), in_proj_covar=tensor([0.0271, 0.0293, 0.0262, 0.0269, 0.0315, 0.0293, 0.0256, 0.0245], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 05:35:01,746 INFO [train.py:873] (3/4) Epoch 14, batch 3000, loss[loss=0.1251, simple_loss=0.1607, pruned_loss=0.04474, over 14269.00 frames. ], tot_loss[loss=0.1174, simple_loss=0.1505, pruned_loss=0.04213, over 2077287.86 frames. ], batch size: 60, lr: 5.66e-03, grad_scale: 4.0 2022-12-08 05:35:01,747 INFO [train.py:896] (3/4) Computing validation loss 2022-12-08 05:35:10,214 INFO [train.py:905] (3/4) Epoch 14, validation: loss=0.134, simple_loss=0.1722, pruned_loss=0.04793, over 857387.00 frames. 2022-12-08 05:35:10,215 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-08 05:36:12,836 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.532e+02 2.305e+02 3.036e+02 3.734e+02 6.961e+02, threshold=6.071e+02, percent-clipped=4.0 2022-12-08 05:36:18,845 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=101386.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 05:36:37,209 INFO [train.py:873] (3/4) Epoch 14, batch 3100, loss[loss=0.114, simple_loss=0.1374, pruned_loss=0.04527, over 5988.00 frames. ], tot_loss[loss=0.117, simple_loss=0.1504, pruned_loss=0.0418, over 2107684.34 frames. ], batch size: 100, lr: 5.66e-03, grad_scale: 4.0 2022-12-08 05:37:00,774 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=101434.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 05:37:12,220 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1457, 3.7635, 2.8450, 4.3658, 4.0313, 4.2269, 3.7563, 3.1546], device='cuda:3'), covar=tensor([0.0846, 0.1192, 0.3599, 0.0539, 0.0998, 0.1202, 0.1109, 0.2990], device='cuda:3'), in_proj_covar=tensor([0.0274, 0.0295, 0.0265, 0.0274, 0.0318, 0.0298, 0.0258, 0.0248], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 05:37:20,466 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.04 vs. limit=5.0 2022-12-08 05:37:26,093 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=101463.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:37:34,112 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.5103, 1.5942, 1.7672, 1.4835, 1.4064, 1.3139, 1.2341, 1.0966], device='cuda:3'), covar=tensor([0.0190, 0.0170, 0.0186, 0.0186, 0.0209, 0.0353, 0.0225, 0.0404], device='cuda:3'), in_proj_covar=tensor([0.0019, 0.0019, 0.0017, 0.0018, 0.0018, 0.0030, 0.0024, 0.0029], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 05:37:39,841 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.043e+01 2.198e+02 2.600e+02 3.209e+02 7.473e+02, threshold=5.200e+02, percent-clipped=3.0 2022-12-08 05:37:42,066 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6310, 1.6124, 1.6925, 1.5223, 1.4314, 1.4572, 1.1345, 1.1409], device='cuda:3'), covar=tensor([0.0151, 0.0182, 0.0208, 0.0142, 0.0195, 0.0292, 0.0263, 0.0363], device='cuda:3'), in_proj_covar=tensor([0.0019, 0.0019, 0.0017, 0.0018, 0.0018, 0.0030, 0.0024, 0.0029], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 05:38:04,115 INFO [train.py:873] (3/4) Epoch 14, batch 3200, loss[loss=0.1381, simple_loss=0.1567, pruned_loss=0.05969, over 3874.00 frames. ], tot_loss[loss=0.1171, simple_loss=0.1503, pruned_loss=0.04199, over 2036485.58 frames. ], batch size: 100, lr: 5.65e-03, grad_scale: 8.0 2022-12-08 05:38:24,965 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.6530, 5.1446, 5.1362, 5.5848, 5.2780, 4.6573, 5.5257, 4.6707], device='cuda:3'), covar=tensor([0.0293, 0.0849, 0.0305, 0.0353, 0.0704, 0.0442, 0.0469, 0.0455], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0273, 0.0192, 0.0190, 0.0181, 0.0153, 0.0281, 0.0166], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 05:38:28,511 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9735, 3.0329, 3.0141, 3.0182, 3.0735, 2.9584, 1.3725, 2.8257], device='cuda:3'), covar=tensor([0.0462, 0.0498, 0.0553, 0.0487, 0.0427, 0.0705, 0.3259, 0.0461], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0174, 0.0143, 0.0141, 0.0203, 0.0139, 0.0159, 0.0193], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 05:39:06,763 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8050, 1.8823, 4.2819, 3.9689, 3.9040, 4.4092, 4.0209, 4.3673], device='cuda:3'), covar=tensor([0.2141, 0.2022, 0.0188, 0.0321, 0.0343, 0.0265, 0.0206, 0.0198], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0158, 0.0129, 0.0168, 0.0145, 0.0140, 0.0121, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 05:39:08,338 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.211e+02 2.131e+02 2.492e+02 3.275e+02 6.158e+02, threshold=4.985e+02, percent-clipped=2.0 2022-12-08 05:39:28,728 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.4303, 5.3426, 4.9330, 5.6535, 5.1661, 4.8222, 5.5943, 5.4470], device='cuda:3'), covar=tensor([0.0670, 0.0579, 0.0860, 0.0371, 0.0798, 0.0435, 0.0527, 0.0496], device='cuda:3'), in_proj_covar=tensor([0.0140, 0.0137, 0.0142, 0.0156, 0.0144, 0.0120, 0.0163, 0.0142], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 05:39:32,251 INFO [train.py:873] (3/4) Epoch 14, batch 3300, loss[loss=0.1174, simple_loss=0.1451, pruned_loss=0.04483, over 4957.00 frames. ], tot_loss[loss=0.1162, simple_loss=0.1494, pruned_loss=0.04151, over 2018732.57 frames. ], batch size: 100, lr: 5.65e-03, grad_scale: 4.0 2022-12-08 05:39:35,330 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0100, 1.7369, 1.9617, 2.0478, 2.1701, 1.3309, 1.9785, 2.2849], device='cuda:3'), covar=tensor([0.1393, 0.1316, 0.1494, 0.0739, 0.0892, 0.0906, 0.1035, 0.0478], device='cuda:3'), in_proj_covar=tensor([0.0030, 0.0029, 0.0032, 0.0028, 0.0029, 0.0043, 0.0031, 0.0033], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 05:40:02,702 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.27 vs. limit=5.0 2022-12-08 05:40:36,072 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.167e+02 2.269e+02 2.791e+02 3.629e+02 1.218e+03, threshold=5.581e+02, percent-clipped=7.0 2022-12-08 05:40:39,680 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=101684.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:40:59,542 INFO [train.py:873] (3/4) Epoch 14, batch 3400, loss[loss=0.1187, simple_loss=0.1351, pruned_loss=0.0512, over 2658.00 frames. ], tot_loss[loss=0.1163, simple_loss=0.1494, pruned_loss=0.04162, over 1951173.01 frames. ], batch size: 100, lr: 5.65e-03, grad_scale: 4.0 2022-12-08 05:41:32,709 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=101745.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:41:48,183 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=101763.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:42:02,723 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.178e+02 2.274e+02 2.665e+02 3.242e+02 5.348e+02, threshold=5.329e+02, percent-clipped=0.0 2022-12-08 05:42:26,268 INFO [train.py:873] (3/4) Epoch 14, batch 3500, loss[loss=0.119, simple_loss=0.1517, pruned_loss=0.04312, over 14364.00 frames. ], tot_loss[loss=0.1171, simple_loss=0.1495, pruned_loss=0.04238, over 1883649.34 frames. ], batch size: 73, lr: 5.65e-03, grad_scale: 4.0 2022-12-08 05:42:29,435 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=101811.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:43:09,595 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.0271, 4.9656, 5.4300, 4.6295, 5.2566, 5.5393, 2.2671, 4.8537], device='cuda:3'), covar=tensor([0.0258, 0.0319, 0.0320, 0.0444, 0.0258, 0.0122, 0.2826, 0.0250], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0174, 0.0143, 0.0142, 0.0204, 0.0140, 0.0160, 0.0193], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 05:43:29,694 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.267e+02 2.210e+02 2.811e+02 3.369e+02 8.008e+02, threshold=5.622e+02, percent-clipped=4.0 2022-12-08 05:43:29,861 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.2218, 4.3061, 4.6084, 3.8678, 4.4739, 4.5834, 1.8247, 4.1839], device='cuda:3'), covar=tensor([0.0313, 0.0358, 0.0338, 0.0604, 0.0281, 0.0228, 0.3085, 0.0268], device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0175, 0.0144, 0.0143, 0.0205, 0.0140, 0.0161, 0.0194], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 05:43:33,069 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.0111, 1.3047, 1.3705, 0.9911, 0.8480, 1.1010, 0.8928, 1.2050], device='cuda:3'), covar=tensor([0.1658, 0.2337, 0.0902, 0.2130, 0.2848, 0.1158, 0.1743, 0.1134], device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0099, 0.0091, 0.0099, 0.0116, 0.0087, 0.0121, 0.0092], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 05:43:53,152 INFO [train.py:873] (3/4) Epoch 14, batch 3600, loss[loss=0.1199, simple_loss=0.1468, pruned_loss=0.04655, over 4942.00 frames. ], tot_loss[loss=0.1166, simple_loss=0.1494, pruned_loss=0.04189, over 1987008.46 frames. ], batch size: 100, lr: 5.64e-03, grad_scale: 8.0 2022-12-08 05:44:05,136 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=101920.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 05:44:42,745 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6914, 1.7189, 1.8211, 1.8121, 1.9430, 1.2110, 1.5879, 1.7710], device='cuda:3'), covar=tensor([0.1249, 0.0968, 0.0789, 0.1043, 0.0697, 0.0931, 0.0960, 0.0928], device='cuda:3'), in_proj_covar=tensor([0.0030, 0.0030, 0.0033, 0.0029, 0.0030, 0.0043, 0.0031, 0.0033], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 05:44:55,588 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.03 vs. limit=2.0 2022-12-08 05:44:57,576 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.281e+02 2.121e+02 2.665e+02 3.565e+02 6.931e+02, threshold=5.329e+02, percent-clipped=3.0 2022-12-08 05:44:58,646 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=101981.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 05:45:03,620 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=101987.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:45:21,975 INFO [train.py:873] (3/4) Epoch 14, batch 3700, loss[loss=0.09823, simple_loss=0.1118, pruned_loss=0.04235, over 2614.00 frames. ], tot_loss[loss=0.1177, simple_loss=0.1502, pruned_loss=0.04256, over 1968531.33 frames. ], batch size: 100, lr: 5.64e-03, grad_scale: 8.0 2022-12-08 05:45:51,418 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=102040.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:45:58,301 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=102048.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:46:26,400 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.122e+02 2.149e+02 2.707e+02 3.533e+02 7.252e+02, threshold=5.415e+02, percent-clipped=4.0 2022-12-08 05:46:29,166 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7266, 2.6903, 2.0240, 2.7231, 2.5695, 2.6563, 2.3958, 2.1960], device='cuda:3'), covar=tensor([0.1007, 0.1608, 0.3102, 0.1136, 0.1195, 0.1516, 0.1838, 0.3059], device='cuda:3'), in_proj_covar=tensor([0.0275, 0.0295, 0.0265, 0.0275, 0.0321, 0.0298, 0.0261, 0.0247], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 05:46:49,972 INFO [train.py:873] (3/4) Epoch 14, batch 3800, loss[loss=0.1742, simple_loss=0.1616, pruned_loss=0.09342, over 1225.00 frames. ], tot_loss[loss=0.1173, simple_loss=0.1502, pruned_loss=0.04217, over 1961728.59 frames. ], batch size: 100, lr: 5.64e-03, grad_scale: 8.0 2022-12-08 05:47:56,010 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.049e+02 2.208e+02 2.673e+02 3.606e+02 1.148e+03, threshold=5.346e+02, percent-clipped=3.0 2022-12-08 05:48:19,967 INFO [train.py:873] (3/4) Epoch 14, batch 3900, loss[loss=0.1164, simple_loss=0.146, pruned_loss=0.04344, over 14456.00 frames. ], tot_loss[loss=0.1173, simple_loss=0.1501, pruned_loss=0.04226, over 1942474.68 frames. ], batch size: 51, lr: 5.64e-03, grad_scale: 4.0 2022-12-08 05:48:34,363 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4029, 4.5102, 4.8291, 4.1025, 4.5681, 4.7349, 1.7099, 4.3499], device='cuda:3'), covar=tensor([0.0318, 0.0308, 0.0331, 0.0415, 0.0373, 0.0164, 0.3179, 0.0266], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0175, 0.0144, 0.0143, 0.0205, 0.0140, 0.0161, 0.0193], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 05:49:22,108 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=102276.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 05:49:26,437 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.176e+02 2.029e+02 2.446e+02 2.972e+02 7.280e+02, threshold=4.892e+02, percent-clipped=2.0 2022-12-08 05:49:28,275 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6641, 3.4123, 3.1728, 3.3429, 3.5792, 3.5762, 3.6132, 3.6372], device='cuda:3'), covar=tensor([0.1012, 0.0692, 0.2097, 0.2792, 0.0857, 0.0960, 0.1138, 0.0876], device='cuda:3'), in_proj_covar=tensor([0.0388, 0.0271, 0.0449, 0.0574, 0.0348, 0.0444, 0.0395, 0.0382], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 05:49:49,245 INFO [train.py:873] (3/4) Epoch 14, batch 4000, loss[loss=0.116, simple_loss=0.1406, pruned_loss=0.04569, over 4981.00 frames. ], tot_loss[loss=0.1166, simple_loss=0.149, pruned_loss=0.04205, over 1910098.07 frames. ], batch size: 100, lr: 5.63e-03, grad_scale: 8.0 2022-12-08 05:50:18,896 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=102340.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:50:21,383 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=102343.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:50:55,302 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.337e+02 2.024e+02 2.365e+02 2.996e+02 7.946e+02, threshold=4.730e+02, percent-clipped=3.0 2022-12-08 05:51:01,411 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=102388.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:51:18,480 INFO [train.py:873] (3/4) Epoch 14, batch 4100, loss[loss=0.1382, simple_loss=0.163, pruned_loss=0.05667, over 14377.00 frames. ], tot_loss[loss=0.1166, simple_loss=0.1492, pruned_loss=0.04193, over 1957850.51 frames. ], batch size: 31, lr: 5.63e-03, grad_scale: 8.0 2022-12-08 05:51:29,391 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8053, 1.6354, 1.9444, 1.7215, 1.9974, 1.7683, 1.7474, 1.8975], device='cuda:3'), covar=tensor([0.0666, 0.1425, 0.0414, 0.0407, 0.0403, 0.0739, 0.0295, 0.0363], device='cuda:3'), in_proj_covar=tensor([0.0355, 0.0317, 0.0398, 0.0303, 0.0375, 0.0327, 0.0368, 0.0303], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 05:52:14,196 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1850, 2.1836, 2.3216, 2.3539, 2.4232, 1.3072, 2.4758, 2.4318], device='cuda:3'), covar=tensor([0.0973, 0.1390, 0.0857, 0.1307, 0.1113, 0.0733, 0.0561, 0.0588], device='cuda:3'), in_proj_covar=tensor([0.0030, 0.0029, 0.0033, 0.0028, 0.0030, 0.0042, 0.0031, 0.0033], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 05:52:23,527 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.525e+01 2.118e+02 2.762e+02 3.343e+02 8.705e+02, threshold=5.523e+02, percent-clipped=3.0 2022-12-08 05:52:45,282 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([6.0824, 5.4607, 5.5273, 6.0083, 5.6270, 4.9225, 5.9555, 4.8614], device='cuda:3'), covar=tensor([0.0309, 0.1018, 0.0333, 0.0367, 0.0696, 0.0347, 0.0446, 0.0464], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0271, 0.0191, 0.0189, 0.0181, 0.0152, 0.0280, 0.0166], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 05:52:46,044 INFO [train.py:873] (3/4) Epoch 14, batch 4200, loss[loss=0.1027, simple_loss=0.1385, pruned_loss=0.03339, over 14019.00 frames. ], tot_loss[loss=0.1171, simple_loss=0.1498, pruned_loss=0.04215, over 1955412.04 frames. ], batch size: 19, lr: 5.63e-03, grad_scale: 8.0 2022-12-08 05:53:07,647 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2022-12-08 05:53:45,736 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=102576.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 05:53:49,866 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.062e+02 2.234e+02 2.939e+02 3.624e+02 9.310e+02, threshold=5.879e+02, percent-clipped=3.0 2022-12-08 05:54:13,400 INFO [train.py:873] (3/4) Epoch 14, batch 4300, loss[loss=0.1505, simple_loss=0.1507, pruned_loss=0.07518, over 2661.00 frames. ], tot_loss[loss=0.1181, simple_loss=0.1506, pruned_loss=0.04275, over 1945958.07 frames. ], batch size: 100, lr: 5.62e-03, grad_scale: 8.0 2022-12-08 05:54:18,247 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.69 vs. limit=5.0 2022-12-08 05:54:27,882 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=102624.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 05:54:37,512 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.86 vs. limit=2.0 2022-12-08 05:54:38,714 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.7648, 4.4009, 4.2521, 4.7601, 4.4863, 4.1878, 4.7411, 3.9519], device='cuda:3'), covar=tensor([0.0331, 0.0959, 0.0344, 0.0410, 0.0724, 0.0736, 0.0503, 0.0500], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0272, 0.0191, 0.0189, 0.0182, 0.0153, 0.0281, 0.0166], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 05:54:44,459 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=102643.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:54:48,376 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.2110, 1.6038, 1.7373, 1.8011, 1.6256, 1.6958, 1.3692, 1.1899], device='cuda:3'), covar=tensor([0.1525, 0.1222, 0.0694, 0.0742, 0.1479, 0.1380, 0.2077, 0.2686], device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0086, 0.0068, 0.0072, 0.0097, 0.0084, 0.0099, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0006, 0.0006], device='cuda:3') 2022-12-08 05:55:04,089 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9730, 3.8690, 3.6846, 4.0983, 3.7058, 3.4345, 4.0889, 3.9122], device='cuda:3'), covar=tensor([0.0673, 0.0902, 0.0929, 0.0674, 0.0796, 0.0721, 0.0596, 0.0739], device='cuda:3'), in_proj_covar=tensor([0.0138, 0.0134, 0.0140, 0.0153, 0.0141, 0.0119, 0.0161, 0.0142], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 05:55:16,065 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.389e+01 1.915e+02 2.366e+02 2.869e+02 5.820e+02, threshold=4.732e+02, percent-clipped=0.0 2022-12-08 05:55:17,731 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2022-12-08 05:55:25,243 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=102691.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:55:38,797 INFO [train.py:873] (3/4) Epoch 14, batch 4400, loss[loss=0.1039, simple_loss=0.1468, pruned_loss=0.03051, over 14281.00 frames. ], tot_loss[loss=0.118, simple_loss=0.1504, pruned_loss=0.04279, over 1937139.84 frames. ], batch size: 44, lr: 5.62e-03, grad_scale: 8.0 2022-12-08 05:55:39,434 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2022-12-08 05:56:44,117 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.277e+02 2.029e+02 2.612e+02 3.153e+02 6.081e+02, threshold=5.224e+02, percent-clipped=1.0 2022-12-08 05:56:52,230 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6665, 2.3272, 3.4666, 2.6588, 3.5236, 3.3103, 3.2706, 2.8005], device='cuda:3'), covar=tensor([0.1089, 0.3382, 0.1297, 0.2029, 0.0959, 0.1254, 0.1557, 0.2088], device='cuda:3'), in_proj_covar=tensor([0.0356, 0.0317, 0.0399, 0.0302, 0.0377, 0.0325, 0.0366, 0.0304], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 05:57:06,623 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.76 vs. limit=5.0 2022-12-08 05:57:07,855 INFO [train.py:873] (3/4) Epoch 14, batch 4500, loss[loss=0.1272, simple_loss=0.1299, pruned_loss=0.06228, over 2600.00 frames. ], tot_loss[loss=0.1168, simple_loss=0.1497, pruned_loss=0.04195, over 1929003.64 frames. ], batch size: 100, lr: 5.62e-03, grad_scale: 8.0 2022-12-08 05:57:36,999 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7472, 3.8720, 4.0506, 3.5891, 3.9119, 3.9477, 1.5415, 3.7049], device='cuda:3'), covar=tensor([0.0350, 0.0392, 0.0375, 0.0663, 0.0358, 0.0338, 0.3240, 0.0329], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0173, 0.0143, 0.0141, 0.0203, 0.0139, 0.0159, 0.0191], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 05:57:46,942 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=102851.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:58:12,667 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.348e+02 2.344e+02 2.809e+02 3.466e+02 7.090e+02, threshold=5.617e+02, percent-clipped=4.0 2022-12-08 05:58:19,005 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=102887.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:58:31,154 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.59 vs. limit=2.0 2022-12-08 05:58:35,780 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9681, 4.8513, 4.1536, 4.5144, 4.5953, 4.9555, 5.0987, 4.9337], device='cuda:3'), covar=tensor([0.1174, 0.0648, 0.2837, 0.3646, 0.1111, 0.1176, 0.1194, 0.1178], device='cuda:3'), in_proj_covar=tensor([0.0384, 0.0268, 0.0438, 0.0561, 0.0343, 0.0437, 0.0390, 0.0380], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 05:58:36,491 INFO [train.py:873] (3/4) Epoch 14, batch 4600, loss[loss=0.1126, simple_loss=0.1518, pruned_loss=0.03669, over 14028.00 frames. ], tot_loss[loss=0.1166, simple_loss=0.1496, pruned_loss=0.0418, over 2006679.41 frames. ], batch size: 26, lr: 5.62e-03, grad_scale: 8.0 2022-12-08 05:58:40,772 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=102912.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:59:10,840 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8373, 1.3411, 1.8135, 1.3005, 1.5687, 1.8753, 1.6249, 1.6510], device='cuda:3'), covar=tensor([0.1063, 0.1059, 0.0869, 0.1498, 0.1500, 0.0809, 0.0765, 0.1664], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0176, 0.0137, 0.0126, 0.0139, 0.0150, 0.0128, 0.0140], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0005, 0.0006], device='cuda:3') 2022-12-08 05:59:12,652 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=102948.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 05:59:41,554 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.429e+02 1.979e+02 2.486e+02 3.145e+02 5.919e+02, threshold=4.973e+02, percent-clipped=1.0 2022-12-08 05:59:56,353 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8920, 1.9575, 3.9128, 2.5802, 3.7872, 1.8988, 2.8138, 3.8110], device='cuda:3'), covar=tensor([0.0668, 0.4217, 0.0469, 0.6798, 0.0638, 0.3446, 0.1512, 0.0464], device='cuda:3'), in_proj_covar=tensor([0.0252, 0.0207, 0.0212, 0.0278, 0.0230, 0.0211, 0.0205, 0.0213], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 06:00:04,661 INFO [train.py:873] (3/4) Epoch 14, batch 4700, loss[loss=0.1458, simple_loss=0.1372, pruned_loss=0.07723, over 1277.00 frames. ], tot_loss[loss=0.1159, simple_loss=0.1492, pruned_loss=0.04131, over 1973602.14 frames. ], batch size: 100, lr: 5.61e-03, grad_scale: 8.0 2022-12-08 06:01:10,016 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.112e+02 2.100e+02 2.622e+02 3.309e+02 6.848e+02, threshold=5.244e+02, percent-clipped=4.0 2022-12-08 06:01:25,790 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103098.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:01:33,436 INFO [train.py:873] (3/4) Epoch 14, batch 4800, loss[loss=0.09572, simple_loss=0.1424, pruned_loss=0.02454, over 14643.00 frames. ], tot_loss[loss=0.1157, simple_loss=0.149, pruned_loss=0.04118, over 1929565.15 frames. ], batch size: 23, lr: 5.61e-03, grad_scale: 8.0 2022-12-08 06:01:40,822 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0955, 3.0371, 2.4664, 2.6313, 3.0221, 3.0951, 3.1788, 3.0886], device='cuda:3'), covar=tensor([0.1442, 0.0781, 0.3611, 0.3782, 0.1412, 0.1484, 0.1454, 0.1494], device='cuda:3'), in_proj_covar=tensor([0.0384, 0.0267, 0.0440, 0.0559, 0.0341, 0.0436, 0.0389, 0.0379], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 06:01:47,482 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 2022-12-08 06:01:55,909 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103132.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:02:17,070 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.42 vs. limit=5.0 2022-12-08 06:02:19,720 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103159.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:02:39,048 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.284e+02 2.240e+02 2.787e+02 3.398e+02 6.860e+02, threshold=5.574e+02, percent-clipped=4.0 2022-12-08 06:02:49,665 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103193.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:03:02,608 INFO [train.py:873] (3/4) Epoch 14, batch 4900, loss[loss=0.1421, simple_loss=0.164, pruned_loss=0.06005, over 14533.00 frames. ], tot_loss[loss=0.1158, simple_loss=0.1494, pruned_loss=0.04113, over 1962115.84 frames. ], batch size: 49, lr: 5.61e-03, grad_scale: 8.0 2022-12-08 06:03:02,685 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103207.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:03:33,987 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103243.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:04:06,509 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.337e+02 2.061e+02 2.525e+02 3.223e+02 5.765e+02, threshold=5.049e+02, percent-clipped=2.0 2022-12-08 06:04:07,988 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2022-12-08 06:04:26,553 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.23 vs. limit=2.0 2022-12-08 06:04:29,069 INFO [train.py:873] (3/4) Epoch 14, batch 5000, loss[loss=0.1394, simple_loss=0.1644, pruned_loss=0.05721, over 9502.00 frames. ], tot_loss[loss=0.1159, simple_loss=0.1494, pruned_loss=0.04119, over 1965785.22 frames. ], batch size: 100, lr: 5.61e-03, grad_scale: 8.0 2022-12-08 06:05:07,457 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7361, 2.6817, 2.6572, 2.8838, 2.4866, 2.5209, 2.8175, 2.7646], device='cuda:3'), covar=tensor([0.0852, 0.1268, 0.0965, 0.0827, 0.1243, 0.1081, 0.0859, 0.1036], device='cuda:3'), in_proj_covar=tensor([0.0140, 0.0136, 0.0141, 0.0155, 0.0143, 0.0120, 0.0163, 0.0144], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 06:05:15,316 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 2022-12-08 06:05:20,810 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103366.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 06:05:33,575 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.214e+02 2.256e+02 2.699e+02 3.304e+02 5.196e+02, threshold=5.398e+02, percent-clipped=1.0 2022-12-08 06:05:42,125 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7975, 1.5709, 2.0459, 1.7152, 1.9094, 1.4832, 1.6100, 1.9257], device='cuda:3'), covar=tensor([0.2634, 0.2971, 0.0582, 0.1856, 0.1628, 0.1368, 0.1395, 0.0751], device='cuda:3'), in_proj_covar=tensor([0.0251, 0.0205, 0.0211, 0.0278, 0.0230, 0.0210, 0.0204, 0.0213], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 06:05:56,322 INFO [train.py:873] (3/4) Epoch 14, batch 5100, loss[loss=0.1172, simple_loss=0.149, pruned_loss=0.04275, over 14241.00 frames. ], tot_loss[loss=0.1159, simple_loss=0.149, pruned_loss=0.04136, over 1969560.03 frames. ], batch size: 46, lr: 5.60e-03, grad_scale: 4.0 2022-12-08 06:06:13,885 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103427.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 06:06:37,877 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103454.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:07:01,406 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.337e+02 2.364e+02 2.994e+02 3.604e+02 8.056e+02, threshold=5.989e+02, percent-clipped=5.0 2022-12-08 06:07:06,897 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103488.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:07:23,313 INFO [train.py:873] (3/4) Epoch 14, batch 5200, loss[loss=0.1326, simple_loss=0.1558, pruned_loss=0.05473, over 14029.00 frames. ], tot_loss[loss=0.1165, simple_loss=0.1496, pruned_loss=0.04169, over 2007268.57 frames. ], batch size: 29, lr: 5.60e-03, grad_scale: 8.0 2022-12-08 06:07:23,479 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103507.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:07:55,078 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103543.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:08:04,930 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=103555.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:08:28,606 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.299e+02 2.069e+02 2.548e+02 2.976e+02 7.652e+02, threshold=5.095e+02, percent-clipped=1.0 2022-12-08 06:08:36,181 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=103591.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:08:45,571 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2022-12-08 06:08:50,465 INFO [train.py:873] (3/4) Epoch 14, batch 5300, loss[loss=0.1173, simple_loss=0.1547, pruned_loss=0.03992, over 14261.00 frames. ], tot_loss[loss=0.116, simple_loss=0.1494, pruned_loss=0.04128, over 1996846.93 frames. ], batch size: 35, lr: 5.60e-03, grad_scale: 8.0 2022-12-08 06:09:13,851 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7630, 2.7199, 2.0669, 2.8251, 2.6753, 2.7098, 2.4235, 2.2150], device='cuda:3'), covar=tensor([0.0925, 0.1139, 0.2759, 0.0838, 0.1006, 0.1069, 0.1462, 0.2417], device='cuda:3'), in_proj_covar=tensor([0.0273, 0.0292, 0.0261, 0.0274, 0.0318, 0.0297, 0.0255, 0.0244], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 06:09:33,843 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.96 vs. limit=5.0 2022-12-08 06:09:55,897 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.249e+02 2.008e+02 2.559e+02 3.052e+02 6.405e+02, threshold=5.118e+02, percent-clipped=1.0 2022-12-08 06:10:17,826 INFO [train.py:873] (3/4) Epoch 14, batch 5400, loss[loss=0.1522, simple_loss=0.1422, pruned_loss=0.08107, over 1311.00 frames. ], tot_loss[loss=0.116, simple_loss=0.1492, pruned_loss=0.04144, over 1947068.19 frames. ], batch size: 100, lr: 5.59e-03, grad_scale: 4.0 2022-12-08 06:10:30,705 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103722.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 06:10:31,160 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.19 vs. limit=2.0 2022-12-08 06:10:47,361 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103740.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:10:57,544 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.1495, 4.9048, 4.4910, 4.7699, 4.7267, 5.0278, 5.1321, 5.0801], device='cuda:3'), covar=tensor([0.0579, 0.0360, 0.2053, 0.2413, 0.0707, 0.0716, 0.0714, 0.0679], device='cuda:3'), in_proj_covar=tensor([0.0382, 0.0266, 0.0445, 0.0562, 0.0345, 0.0436, 0.0387, 0.0380], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 06:10:59,282 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103754.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:11:08,651 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.46 vs. limit=5.0 2022-12-08 06:11:09,100 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103765.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:11:25,001 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.416e+02 2.268e+02 2.622e+02 3.465e+02 5.931e+02, threshold=5.245e+02, percent-clipped=4.0 2022-12-08 06:11:29,561 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103788.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:11:41,215 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103801.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:11:41,966 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=103802.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:11:46,667 INFO [train.py:873] (3/4) Epoch 14, batch 5500, loss[loss=0.1215, simple_loss=0.1542, pruned_loss=0.04441, over 14249.00 frames. ], tot_loss[loss=0.1165, simple_loss=0.1493, pruned_loss=0.04178, over 1945664.51 frames. ], batch size: 69, lr: 5.59e-03, grad_scale: 4.0 2022-12-08 06:12:03,666 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103826.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:12:12,182 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=103836.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:12:47,348 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103876.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:12:51,557 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1736, 2.0741, 5.0259, 4.5551, 4.3352, 5.0956, 4.9720, 5.1515], device='cuda:3'), covar=tensor([0.1390, 0.1336, 0.0077, 0.0212, 0.0212, 0.0103, 0.0078, 0.0090], device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0158, 0.0128, 0.0169, 0.0144, 0.0140, 0.0122, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 06:12:53,008 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.081e+02 2.148e+02 2.635e+02 3.186e+02 7.478e+02, threshold=5.269e+02, percent-clipped=1.0 2022-12-08 06:13:06,139 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8101, 2.5009, 4.8135, 3.4492, 4.5518, 2.2942, 3.4653, 4.6275], device='cuda:3'), covar=tensor([0.0407, 0.3691, 0.0400, 0.5121, 0.0517, 0.3082, 0.1237, 0.0428], device='cuda:3'), in_proj_covar=tensor([0.0251, 0.0207, 0.0211, 0.0278, 0.0229, 0.0210, 0.0205, 0.0215], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 06:13:14,657 INFO [train.py:873] (3/4) Epoch 14, batch 5600, loss[loss=0.1439, simple_loss=0.1661, pruned_loss=0.06086, over 14516.00 frames. ], tot_loss[loss=0.1176, simple_loss=0.1502, pruned_loss=0.04253, over 1951852.00 frames. ], batch size: 49, lr: 5.59e-03, grad_scale: 8.0 2022-12-08 06:13:41,243 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103937.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 06:13:48,190 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.2758, 2.2613, 4.4094, 2.9348, 4.2148, 2.2239, 3.1281, 4.1804], device='cuda:3'), covar=tensor([0.0635, 0.4005, 0.0381, 0.6838, 0.0576, 0.3403, 0.1325, 0.0391], device='cuda:3'), in_proj_covar=tensor([0.0252, 0.0209, 0.0212, 0.0279, 0.0230, 0.0212, 0.0207, 0.0215], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 06:13:52,626 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 2022-12-08 06:13:56,519 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1615, 1.8688, 2.1968, 2.3024, 2.0387, 1.9614, 2.1714, 2.0598], device='cuda:3'), covar=tensor([0.0345, 0.0666, 0.0316, 0.0288, 0.0466, 0.0706, 0.0409, 0.0390], device='cuda:3'), in_proj_covar=tensor([0.0292, 0.0257, 0.0373, 0.0328, 0.0269, 0.0303, 0.0307, 0.0283], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 06:13:59,819 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8287, 2.8870, 2.9676, 2.9092, 2.8939, 2.7130, 1.5526, 2.6876], device='cuda:3'), covar=tensor([0.0416, 0.0368, 0.0388, 0.0382, 0.0357, 0.0880, 0.2663, 0.0333], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0173, 0.0144, 0.0142, 0.0203, 0.0140, 0.0159, 0.0192], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 06:14:21,890 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.081e+02 2.071e+02 2.747e+02 3.245e+02 6.809e+02, threshold=5.494e+02, percent-clipped=2.0 2022-12-08 06:14:27,090 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103990.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:14:32,982 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103997.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:14:42,178 INFO [train.py:873] (3/4) Epoch 14, batch 5700, loss[loss=0.1554, simple_loss=0.1509, pruned_loss=0.07997, over 1244.00 frames. ], tot_loss[loss=0.1169, simple_loss=0.1498, pruned_loss=0.04196, over 2004954.67 frames. ], batch size: 100, lr: 5.59e-03, grad_scale: 4.0 2022-12-08 06:14:55,472 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104022.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 06:15:18,590 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104049.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:15:20,382 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104051.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:15:26,927 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104058.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:15:36,675 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104070.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 06:15:49,005 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.572e+02 2.112e+02 2.664e+02 3.363e+02 7.605e+02, threshold=5.327e+02, percent-clipped=4.0 2022-12-08 06:15:59,763 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104096.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:16:08,847 INFO [train.py:873] (3/4) Epoch 14, batch 5800, loss[loss=0.1226, simple_loss=0.1601, pruned_loss=0.04256, over 14285.00 frames. ], tot_loss[loss=0.1172, simple_loss=0.15, pruned_loss=0.04221, over 1972122.68 frames. ], batch size: 31, lr: 5.58e-03, grad_scale: 4.0 2022-12-08 06:16:11,477 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104110.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:16:20,665 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104121.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:17:16,434 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.093e+02 2.209e+02 2.706e+02 3.376e+02 1.574e+03, threshold=5.411e+02, percent-clipped=5.0 2022-12-08 06:17:36,929 INFO [train.py:873] (3/4) Epoch 14, batch 5900, loss[loss=0.1186, simple_loss=0.1385, pruned_loss=0.04937, over 5008.00 frames. ], tot_loss[loss=0.1162, simple_loss=0.1495, pruned_loss=0.04145, over 2008460.37 frames. ], batch size: 100, lr: 5.58e-03, grad_scale: 4.0 2022-12-08 06:17:58,488 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104232.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 06:18:26,253 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7442, 3.1009, 2.9228, 3.1465, 2.4482, 3.1405, 2.9728, 1.6092], device='cuda:3'), covar=tensor([0.1432, 0.0904, 0.0819, 0.0455, 0.0857, 0.0535, 0.0821, 0.2065], device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0085, 0.0067, 0.0072, 0.0097, 0.0084, 0.0098, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0006, 0.0006], device='cuda:3') 2022-12-08 06:18:43,106 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.381e+02 2.217e+02 2.661e+02 3.155e+02 4.728e+02, threshold=5.321e+02, percent-clipped=0.0 2022-12-08 06:18:49,456 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0199, 2.0969, 1.9329, 2.1809, 1.7749, 1.9543, 2.0791, 2.1054], device='cuda:3'), covar=tensor([0.1024, 0.1139, 0.1102, 0.0849, 0.1524, 0.0964, 0.1090, 0.0913], device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0140, 0.0144, 0.0159, 0.0146, 0.0122, 0.0168, 0.0146], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 06:19:04,080 INFO [train.py:873] (3/4) Epoch 14, batch 6000, loss[loss=0.1433, simple_loss=0.1767, pruned_loss=0.05491, over 14298.00 frames. ], tot_loss[loss=0.1175, simple_loss=0.1502, pruned_loss=0.04238, over 1953664.33 frames. ], batch size: 39, lr: 5.58e-03, grad_scale: 8.0 2022-12-08 06:19:04,081 INFO [train.py:896] (3/4) Computing validation loss 2022-12-08 06:19:12,429 INFO [train.py:905] (3/4) Epoch 14, validation: loss=0.1346, simple_loss=0.1729, pruned_loss=0.04809, over 857387.00 frames. 2022-12-08 06:19:12,429 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-08 06:19:46,006 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104346.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:19:52,268 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104353.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:20:14,137 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.28 vs. limit=5.0 2022-12-08 06:20:18,638 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.302e+02 2.150e+02 2.718e+02 3.385e+02 8.730e+02, threshold=5.435e+02, percent-clipped=4.0 2022-12-08 06:20:29,302 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104396.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:20:37,576 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104405.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:20:39,256 INFO [train.py:873] (3/4) Epoch 14, batch 6100, loss[loss=0.1289, simple_loss=0.1568, pruned_loss=0.0505, over 13530.00 frames. ], tot_loss[loss=0.1168, simple_loss=0.1497, pruned_loss=0.04189, over 1983728.20 frames. ], batch size: 100, lr: 5.58e-03, grad_scale: 8.0 2022-12-08 06:20:51,277 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104421.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:20:57,060 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9875, 2.5036, 3.7661, 2.7237, 3.8080, 3.6360, 3.6345, 3.1193], device='cuda:3'), covar=tensor([0.0807, 0.3197, 0.0998, 0.2182, 0.0732, 0.0978, 0.1459, 0.2663], device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0314, 0.0393, 0.0302, 0.0374, 0.0321, 0.0360, 0.0300], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 06:21:11,707 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104444.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:21:16,395 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2022-12-08 06:21:30,413 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5016, 3.2129, 2.5073, 3.6148, 3.4720, 3.4598, 3.0860, 2.4816], device='cuda:3'), covar=tensor([0.0930, 0.1477, 0.3602, 0.0633, 0.0962, 0.1592, 0.1386, 0.3364], device='cuda:3'), in_proj_covar=tensor([0.0275, 0.0292, 0.0263, 0.0275, 0.0320, 0.0300, 0.0255, 0.0245], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 06:21:32,841 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104469.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:21:46,358 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.494e+02 2.353e+02 3.023e+02 3.584e+02 7.529e+02, threshold=6.047e+02, percent-clipped=6.0 2022-12-08 06:21:50,867 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3314, 2.1441, 1.9828, 2.2285, 2.2543, 1.2427, 1.9334, 2.2813], device='cuda:3'), covar=tensor([0.0589, 0.0701, 0.0577, 0.0723, 0.0871, 0.0782, 0.0851, 0.0719], device='cuda:3'), in_proj_covar=tensor([0.0031, 0.0030, 0.0033, 0.0029, 0.0030, 0.0043, 0.0031, 0.0034], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 06:22:05,693 INFO [train.py:873] (3/4) Epoch 14, batch 6200, loss[loss=0.1204, simple_loss=0.1563, pruned_loss=0.0422, over 14413.00 frames. ], tot_loss[loss=0.1163, simple_loss=0.1495, pruned_loss=0.04153, over 2032082.18 frames. ], batch size: 41, lr: 5.57e-03, grad_scale: 8.0 2022-12-08 06:22:07,080 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2022-12-08 06:22:21,334 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9920, 1.6238, 2.0513, 1.4416, 1.7592, 2.0552, 1.9024, 1.8124], device='cuda:3'), covar=tensor([0.0796, 0.0737, 0.0829, 0.1353, 0.1236, 0.0881, 0.0663, 0.1573], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0175, 0.0137, 0.0127, 0.0140, 0.0151, 0.0128, 0.0142], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 06:22:27,811 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104532.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 06:23:09,242 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104580.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:23:12,642 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.150e+02 2.145e+02 2.527e+02 3.084e+02 9.249e+02, threshold=5.054e+02, percent-clipped=2.0 2022-12-08 06:23:33,266 INFO [train.py:873] (3/4) Epoch 14, batch 6300, loss[loss=0.1773, simple_loss=0.1558, pruned_loss=0.09938, over 1275.00 frames. ], tot_loss[loss=0.116, simple_loss=0.1493, pruned_loss=0.04128, over 2002647.91 frames. ], batch size: 100, lr: 5.57e-03, grad_scale: 8.0 2022-12-08 06:23:44,588 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2022-12-08 06:23:58,389 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9051, 3.8183, 3.3743, 2.7271, 3.3464, 3.6872, 4.1378, 3.2704], device='cuda:3'), covar=tensor([0.0504, 0.0896, 0.0810, 0.1184, 0.0865, 0.0619, 0.0611, 0.1020], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0175, 0.0136, 0.0126, 0.0139, 0.0151, 0.0127, 0.0140], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 06:24:07,382 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104646.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:24:13,546 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104653.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:24:21,252 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5116, 1.6657, 4.4111, 2.0805, 4.3212, 4.5553, 4.0665, 4.9000], device='cuda:3'), covar=tensor([0.0198, 0.3189, 0.0356, 0.2174, 0.0325, 0.0316, 0.0333, 0.0153], device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0157, 0.0160, 0.0169, 0.0168, 0.0179, 0.0134, 0.0149], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 06:24:40,912 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.535e+01 2.052e+02 2.556e+02 3.229e+02 6.200e+02, threshold=5.113e+02, percent-clipped=1.0 2022-12-08 06:24:47,917 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1996, 1.8140, 2.2239, 2.2633, 2.1767, 1.2469, 2.0189, 2.2725], device='cuda:3'), covar=tensor([0.0756, 0.0837, 0.0406, 0.1137, 0.1423, 0.0897, 0.1002, 0.0546], device='cuda:3'), in_proj_covar=tensor([0.0031, 0.0031, 0.0034, 0.0029, 0.0031, 0.0044, 0.0032, 0.0035], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 06:24:49,226 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104694.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:24:55,242 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104701.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:24:58,694 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104705.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:25:00,315 INFO [train.py:873] (3/4) Epoch 14, batch 6400, loss[loss=0.1412, simple_loss=0.1635, pruned_loss=0.05945, over 8614.00 frames. ], tot_loss[loss=0.1151, simple_loss=0.149, pruned_loss=0.04061, over 1976978.34 frames. ], batch size: 100, lr: 5.57e-03, grad_scale: 8.0 2022-12-08 06:25:21,229 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.6275, 2.6917, 2.8074, 2.7092, 2.7549, 2.4724, 1.5519, 2.4838], device='cuda:3'), covar=tensor([0.0537, 0.0465, 0.0431, 0.0397, 0.0423, 0.1154, 0.2495, 0.0381], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0174, 0.0145, 0.0145, 0.0205, 0.0140, 0.0160, 0.0194], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 06:25:40,657 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104753.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:26:04,446 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2022-12-08 06:26:07,308 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.338e+02 2.032e+02 2.617e+02 3.308e+02 7.760e+02, threshold=5.234e+02, percent-clipped=3.0 2022-12-08 06:26:27,856 INFO [train.py:873] (3/4) Epoch 14, batch 6500, loss[loss=0.09399, simple_loss=0.1056, pruned_loss=0.04119, over 2641.00 frames. ], tot_loss[loss=0.1155, simple_loss=0.1491, pruned_loss=0.04096, over 1921298.36 frames. ], batch size: 100, lr: 5.57e-03, grad_scale: 8.0 2022-12-08 06:26:43,091 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8021, 3.2726, 2.9878, 3.1341, 2.3250, 3.2271, 2.9715, 1.5162], device='cuda:3'), covar=tensor([0.1555, 0.0642, 0.1323, 0.0771, 0.1028, 0.0543, 0.1206, 0.2351], device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0086, 0.0068, 0.0071, 0.0098, 0.0084, 0.0099, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0006, 0.0006], device='cuda:3') 2022-12-08 06:26:53,372 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.26 vs. limit=5.0 2022-12-08 06:26:57,170 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2022-12-08 06:27:21,798 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3637, 1.3835, 3.4182, 1.4876, 3.2885, 3.4589, 2.4248, 3.6949], device='cuda:3'), covar=tensor([0.0270, 0.3428, 0.0446, 0.2530, 0.0860, 0.0435, 0.1062, 0.0232], device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0157, 0.0160, 0.0169, 0.0169, 0.0179, 0.0135, 0.0150], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 06:27:35,547 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.169e+02 2.183e+02 2.623e+02 3.627e+02 5.274e+02, threshold=5.247e+02, percent-clipped=1.0 2022-12-08 06:27:54,699 INFO [train.py:873] (3/4) Epoch 14, batch 6600, loss[loss=0.1277, simple_loss=0.1572, pruned_loss=0.04915, over 13542.00 frames. ], tot_loss[loss=0.1157, simple_loss=0.1489, pruned_loss=0.0412, over 1954761.56 frames. ], batch size: 100, lr: 5.56e-03, grad_scale: 8.0 2022-12-08 06:28:55,802 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.53 vs. limit=5.0 2022-12-08 06:29:02,029 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.045e+02 2.257e+02 2.787e+02 3.469e+02 5.148e+02, threshold=5.574e+02, percent-clipped=0.0 2022-12-08 06:29:02,318 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8919, 3.4525, 2.7321, 3.9849, 3.9124, 3.8703, 3.2987, 2.6888], device='cuda:3'), covar=tensor([0.0767, 0.1392, 0.3370, 0.0663, 0.0863, 0.1508, 0.1437, 0.3249], device='cuda:3'), in_proj_covar=tensor([0.0274, 0.0291, 0.0261, 0.0274, 0.0319, 0.0297, 0.0253, 0.0244], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 06:29:20,926 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.0580, 4.6784, 4.5097, 5.0483, 4.6641, 4.3849, 5.0749, 4.2386], device='cuda:3'), covar=tensor([0.0351, 0.0820, 0.0410, 0.0365, 0.0734, 0.0489, 0.0408, 0.0465], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0268, 0.0192, 0.0190, 0.0182, 0.0153, 0.0280, 0.0165], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 06:29:27,171 INFO [train.py:873] (3/4) Epoch 14, batch 6700, loss[loss=0.0956, simple_loss=0.1359, pruned_loss=0.02763, over 14234.00 frames. ], tot_loss[loss=0.1148, simple_loss=0.1486, pruned_loss=0.04051, over 1967915.38 frames. ], batch size: 69, lr: 5.56e-03, grad_scale: 8.0 2022-12-08 06:29:57,869 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.2609, 2.2598, 4.2586, 2.9831, 4.0670, 2.0321, 3.3207, 4.1026], device='cuda:3'), covar=tensor([0.0604, 0.3627, 0.0479, 0.5171, 0.0667, 0.3344, 0.1076, 0.0420], device='cuda:3'), in_proj_covar=tensor([0.0249, 0.0204, 0.0211, 0.0275, 0.0228, 0.0209, 0.0204, 0.0211], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 06:30:27,123 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3253, 4.0431, 3.9272, 4.3794, 4.1332, 3.8362, 4.3904, 3.6632], device='cuda:3'), covar=tensor([0.0494, 0.0927, 0.0459, 0.0409, 0.0814, 0.1000, 0.0531, 0.0554], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0267, 0.0192, 0.0189, 0.0181, 0.0153, 0.0280, 0.0164], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 06:30:27,933 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1892, 3.3082, 3.4832, 3.1452, 3.3608, 3.0029, 1.4179, 3.1653], device='cuda:3'), covar=tensor([0.0429, 0.0459, 0.0395, 0.0570, 0.0379, 0.0791, 0.3192, 0.0342], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0174, 0.0145, 0.0145, 0.0205, 0.0140, 0.0159, 0.0193], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 06:30:35,608 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.155e+02 2.089e+02 2.481e+02 3.177e+02 6.321e+02, threshold=4.962e+02, percent-clipped=1.0 2022-12-08 06:30:54,207 INFO [train.py:873] (3/4) Epoch 14, batch 6800, loss[loss=0.08732, simple_loss=0.129, pruned_loss=0.0228, over 14001.00 frames. ], tot_loss[loss=0.1147, simple_loss=0.1487, pruned_loss=0.04034, over 1961675.77 frames. ], batch size: 19, lr: 5.56e-03, grad_scale: 8.0 2022-12-08 06:31:47,302 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4037, 2.8835, 2.8621, 1.8013, 2.8975, 3.1127, 3.3088, 2.5667], device='cuda:3'), covar=tensor([0.0658, 0.1475, 0.0998, 0.1768, 0.1020, 0.0738, 0.0948, 0.1407], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0174, 0.0136, 0.0125, 0.0138, 0.0151, 0.0127, 0.0138], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0005, 0.0006], device='cuda:3') 2022-12-08 06:32:01,691 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.287e+02 2.161e+02 2.579e+02 3.505e+02 6.694e+02, threshold=5.159e+02, percent-clipped=6.0 2022-12-08 06:32:21,683 INFO [train.py:873] (3/4) Epoch 14, batch 6900, loss[loss=0.1437, simple_loss=0.1643, pruned_loss=0.06148, over 10358.00 frames. ], tot_loss[loss=0.1155, simple_loss=0.149, pruned_loss=0.04098, over 1963821.40 frames. ], batch size: 100, lr: 5.55e-03, grad_scale: 8.0 2022-12-08 06:32:22,977 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2022-12-08 06:32:36,761 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2589, 3.0516, 2.9997, 3.2719, 3.0715, 3.2036, 3.3250, 2.7960], device='cuda:3'), covar=tensor([0.0576, 0.1026, 0.0575, 0.0582, 0.0883, 0.0501, 0.0670, 0.0616], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0268, 0.0192, 0.0191, 0.0182, 0.0154, 0.0282, 0.0166], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 06:32:38,679 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8050, 1.9383, 2.0200, 1.8212, 1.8478, 1.7071, 1.5903, 1.2226], device='cuda:3'), covar=tensor([0.0189, 0.0274, 0.0213, 0.0276, 0.0214, 0.0307, 0.0281, 0.0418], device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0020, 0.0018, 0.0019, 0.0019, 0.0031, 0.0025, 0.0030], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 06:33:23,582 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2022-12-08 06:33:25,909 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.0468, 2.9385, 4.7475, 3.4728, 4.7885, 4.6976, 4.4358, 4.0811], device='cuda:3'), covar=tensor([0.0486, 0.3439, 0.0714, 0.1799, 0.0626, 0.0662, 0.1451, 0.1887], device='cuda:3'), in_proj_covar=tensor([0.0354, 0.0315, 0.0395, 0.0302, 0.0374, 0.0324, 0.0361, 0.0301], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 06:33:29,080 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.242e+02 2.377e+02 2.807e+02 3.459e+02 9.593e+02, threshold=5.615e+02, percent-clipped=4.0 2022-12-08 06:33:48,171 INFO [train.py:873] (3/4) Epoch 14, batch 7000, loss[loss=0.1242, simple_loss=0.1571, pruned_loss=0.04569, over 9497.00 frames. ], tot_loss[loss=0.1161, simple_loss=0.1491, pruned_loss=0.04153, over 1953393.99 frames. ], batch size: 100, lr: 5.55e-03, grad_scale: 8.0 2022-12-08 06:33:59,163 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8367, 1.7537, 2.1257, 1.8641, 1.6918, 1.5716, 0.8550, 1.2553], device='cuda:3'), covar=tensor([0.0269, 0.0360, 0.0189, 0.0197, 0.0281, 0.0322, 0.0328, 0.0494], device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0020, 0.0018, 0.0019, 0.0019, 0.0031, 0.0025, 0.0030], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 06:34:08,446 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9635, 1.9668, 4.7533, 2.6299, 4.4254, 4.9457, 4.5256, 5.3548], device='cuda:3'), covar=tensor([0.0178, 0.2742, 0.0290, 0.1724, 0.0274, 0.0289, 0.0269, 0.0115], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0155, 0.0158, 0.0168, 0.0166, 0.0176, 0.0133, 0.0147], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 06:34:56,769 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.288e+01 2.265e+02 2.868e+02 3.572e+02 1.076e+03, threshold=5.736e+02, percent-clipped=4.0 2022-12-08 06:34:59,588 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=105388.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:35:16,756 INFO [train.py:873] (3/4) Epoch 14, batch 7100, loss[loss=0.134, simple_loss=0.1617, pruned_loss=0.0532, over 14469.00 frames. ], tot_loss[loss=0.1155, simple_loss=0.1487, pruned_loss=0.04112, over 1954428.84 frames. ], batch size: 51, lr: 5.55e-03, grad_scale: 8.0 2022-12-08 06:35:53,781 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=105449.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 06:36:25,252 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.66 vs. limit=5.0 2022-12-08 06:36:25,709 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.061e+02 2.137e+02 2.673e+02 3.373e+02 4.675e+02, threshold=5.347e+02, percent-clipped=1.0 2022-12-08 06:36:25,947 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=105486.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:36:39,545 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=105502.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:36:43,630 INFO [train.py:873] (3/4) Epoch 14, batch 7200, loss[loss=0.08824, simple_loss=0.1332, pruned_loss=0.02163, over 14274.00 frames. ], tot_loss[loss=0.1154, simple_loss=0.1489, pruned_loss=0.04098, over 2003698.11 frames. ], batch size: 18, lr: 5.55e-03, grad_scale: 8.0 2022-12-08 06:36:55,625 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3939, 4.2102, 3.8831, 4.0762, 4.2079, 4.3325, 4.4077, 4.3491], device='cuda:3'), covar=tensor([0.0785, 0.0473, 0.1928, 0.2656, 0.0747, 0.0812, 0.0916, 0.0867], device='cuda:3'), in_proj_covar=tensor([0.0380, 0.0262, 0.0441, 0.0559, 0.0340, 0.0437, 0.0384, 0.0381], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 06:37:17,750 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3164, 3.1295, 2.8824, 3.0257, 3.2387, 3.2468, 3.2851, 3.2758], device='cuda:3'), covar=tensor([0.0911, 0.0683, 0.2233, 0.2698, 0.0834, 0.1051, 0.1212, 0.1009], device='cuda:3'), in_proj_covar=tensor([0.0379, 0.0261, 0.0439, 0.0558, 0.0338, 0.0435, 0.0382, 0.0380], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 06:37:17,833 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=105547.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:37:20,318 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8240, 1.6034, 1.7902, 1.9596, 1.4255, 1.7643, 1.7040, 1.9047], device='cuda:3'), covar=tensor([0.0207, 0.0315, 0.0209, 0.0161, 0.0304, 0.0387, 0.0266, 0.0215], device='cuda:3'), in_proj_covar=tensor([0.0294, 0.0260, 0.0377, 0.0329, 0.0271, 0.0304, 0.0310, 0.0280], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 06:37:32,113 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=105563.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:37:51,601 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.085e+02 2.276e+02 2.781e+02 3.572e+02 7.841e+02, threshold=5.562e+02, percent-clipped=2.0 2022-12-08 06:38:10,277 INFO [train.py:873] (3/4) Epoch 14, batch 7300, loss[loss=0.1139, simple_loss=0.1419, pruned_loss=0.04299, over 4969.00 frames. ], tot_loss[loss=0.1159, simple_loss=0.149, pruned_loss=0.04139, over 1966939.88 frames. ], batch size: 100, lr: 5.54e-03, grad_scale: 8.0 2022-12-08 06:39:20,008 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.394e+02 2.190e+02 2.577e+02 3.146e+02 6.129e+02, threshold=5.153e+02, percent-clipped=1.0 2022-12-08 06:39:38,186 INFO [train.py:873] (3/4) Epoch 14, batch 7400, loss[loss=0.11, simple_loss=0.1501, pruned_loss=0.03491, over 14527.00 frames. ], tot_loss[loss=0.117, simple_loss=0.1495, pruned_loss=0.04222, over 1869321.68 frames. ], batch size: 49, lr: 5.54e-03, grad_scale: 8.0 2022-12-08 06:39:58,369 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.73 vs. limit=5.0 2022-12-08 06:40:11,110 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=105744.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 06:40:48,154 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.057e+02 2.114e+02 2.694e+02 3.325e+02 8.535e+02, threshold=5.387e+02, percent-clipped=3.0 2022-12-08 06:40:51,583 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0355, 2.0233, 2.0893, 2.0667, 2.0559, 1.7034, 1.4621, 1.8354], device='cuda:3'), covar=tensor([0.0570, 0.0529, 0.0485, 0.0392, 0.0451, 0.1243, 0.2097, 0.0470], device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0173, 0.0143, 0.0144, 0.0202, 0.0138, 0.0157, 0.0191], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 06:40:53,318 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=105792.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:40:59,953 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=105799.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:41:07,346 INFO [train.py:873] (3/4) Epoch 14, batch 7500, loss[loss=0.138, simple_loss=0.1612, pruned_loss=0.05738, over 10383.00 frames. ], tot_loss[loss=0.1163, simple_loss=0.1491, pruned_loss=0.04173, over 1893130.56 frames. ], batch size: 100, lr: 5.54e-03, grad_scale: 8.0 2022-12-08 06:41:36,973 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=105842.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:41:44,732 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=105853.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:41:47,025 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=105858.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:41:48,041 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=105860.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:41:49,289 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2022-12-08 06:42:35,200 INFO [train.py:873] (3/4) Epoch 15, batch 0, loss[loss=0.1108, simple_loss=0.1533, pruned_loss=0.03415, over 13973.00 frames. ], tot_loss[loss=0.1108, simple_loss=0.1533, pruned_loss=0.03415, over 13973.00 frames. ], batch size: 19, lr: 5.35e-03, grad_scale: 8.0 2022-12-08 06:42:35,201 INFO [train.py:896] (3/4) Computing validation loss 2022-12-08 06:42:42,694 INFO [train.py:905] (3/4) Epoch 15, validation: loss=0.1381, simple_loss=0.1782, pruned_loss=0.049, over 857387.00 frames. 2022-12-08 06:42:42,695 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-08 06:42:58,146 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 5.674e+01 2.091e+02 3.002e+02 4.364e+02 1.099e+03, threshold=6.004e+02, percent-clipped=13.0 2022-12-08 06:44:07,303 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4690, 2.2194, 3.4246, 3.5497, 3.4127, 2.2961, 3.4351, 2.6368], device='cuda:3'), covar=tensor([0.0435, 0.1152, 0.0735, 0.0486, 0.0498, 0.1533, 0.0460, 0.0983], device='cuda:3'), in_proj_covar=tensor([0.0290, 0.0256, 0.0372, 0.0326, 0.0269, 0.0302, 0.0306, 0.0277], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 06:44:11,359 INFO [train.py:873] (3/4) Epoch 15, batch 100, loss[loss=0.1098, simple_loss=0.1516, pruned_loss=0.03397, over 13941.00 frames. ], tot_loss[loss=0.1173, simple_loss=0.1508, pruned_loss=0.04188, over 887929.82 frames. ], batch size: 23, lr: 5.35e-03, grad_scale: 8.0 2022-12-08 06:44:26,575 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.268e+02 2.570e+02 2.979e+02 3.765e+02 9.423e+02, threshold=5.958e+02, percent-clipped=4.0 2022-12-08 06:44:36,259 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.6020, 4.3816, 4.0890, 4.2999, 4.3852, 4.4769, 4.5849, 4.5402], device='cuda:3'), covar=tensor([0.0811, 0.0554, 0.2142, 0.2585, 0.0813, 0.0844, 0.0968, 0.0845], device='cuda:3'), in_proj_covar=tensor([0.0374, 0.0257, 0.0436, 0.0551, 0.0336, 0.0430, 0.0381, 0.0377], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 06:45:18,391 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106044.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:45:26,207 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1002, 3.9815, 3.5567, 2.6871, 3.5350, 3.7968, 4.2024, 3.5411], device='cuda:3'), covar=tensor([0.0467, 0.1028, 0.0749, 0.1209, 0.0672, 0.0605, 0.0576, 0.0873], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0173, 0.0136, 0.0125, 0.0138, 0.0151, 0.0128, 0.0138], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 06:45:38,211 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2022-12-08 06:45:40,386 INFO [train.py:873] (3/4) Epoch 15, batch 200, loss[loss=0.1531, simple_loss=0.1514, pruned_loss=0.07734, over 1253.00 frames. ], tot_loss[loss=0.1168, simple_loss=0.15, pruned_loss=0.04181, over 1318141.48 frames. ], batch size: 100, lr: 5.34e-03, grad_scale: 8.0 2022-12-08 06:45:47,364 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106077.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:45:55,352 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.497e+02 2.218e+02 2.817e+02 3.547e+02 5.569e+02, threshold=5.634e+02, percent-clipped=0.0 2022-12-08 06:46:00,770 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=106092.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:46:41,882 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106138.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:46:45,006 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106142.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:46:50,461 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106148.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:46:56,486 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106155.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:46:59,141 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106158.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:47:08,666 INFO [train.py:873] (3/4) Epoch 15, batch 300, loss[loss=0.08995, simple_loss=0.1301, pruned_loss=0.02488, over 13919.00 frames. ], tot_loss[loss=0.116, simple_loss=0.1492, pruned_loss=0.04135, over 1594201.46 frames. ], batch size: 19, lr: 5.34e-03, grad_scale: 8.0 2022-12-08 06:47:10,585 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9630, 2.0009, 2.2327, 1.4511, 1.5975, 2.1026, 1.3347, 1.9883], device='cuda:3'), covar=tensor([0.1053, 0.1682, 0.0790, 0.2187, 0.2545, 0.0798, 0.3410, 0.1044], device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0099, 0.0090, 0.0097, 0.0116, 0.0086, 0.0119, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 06:47:23,718 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.045e+02 2.130e+02 2.529e+02 3.253e+02 6.005e+02, threshold=5.059e+02, percent-clipped=1.0 2022-12-08 06:47:27,428 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=106190.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:47:37,511 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.5099, 1.6758, 1.8814, 1.8811, 1.8145, 1.8327, 1.5257, 1.4031], device='cuda:3'), covar=tensor([0.1303, 0.1169, 0.0560, 0.0528, 0.1152, 0.0757, 0.2025, 0.1731], device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0085, 0.0067, 0.0070, 0.0096, 0.0083, 0.0098, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0006, 0.0006], device='cuda:3') 2022-12-08 06:47:42,124 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=106206.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:47:52,777 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8020, 1.0542, 1.3081, 1.2010, 0.9484, 1.3018, 1.0039, 0.8702], device='cuda:3'), covar=tensor([0.2062, 0.1262, 0.0382, 0.0435, 0.1604, 0.0947, 0.1782, 0.1438], device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0085, 0.0068, 0.0070, 0.0096, 0.0083, 0.0098, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0006, 0.0006], device='cuda:3') 2022-12-08 06:48:03,871 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.51 vs. limit=5.0 2022-12-08 06:48:36,983 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106268.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:48:37,736 INFO [train.py:873] (3/4) Epoch 15, batch 400, loss[loss=0.1101, simple_loss=0.1354, pruned_loss=0.04238, over 3876.00 frames. ], tot_loss[loss=0.1155, simple_loss=0.149, pruned_loss=0.04099, over 1726699.33 frames. ], batch size: 100, lr: 5.34e-03, grad_scale: 8.0 2022-12-08 06:48:38,163 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9311, 3.0452, 3.1497, 2.9577, 3.0374, 2.8969, 1.5440, 2.8099], device='cuda:3'), covar=tensor([0.0426, 0.0398, 0.0330, 0.0454, 0.0332, 0.0742, 0.2615, 0.0327], device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0171, 0.0142, 0.0143, 0.0200, 0.0138, 0.0156, 0.0190], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 06:48:52,653 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.320e+02 2.076e+02 2.571e+02 3.282e+02 8.496e+02, threshold=5.142e+02, percent-clipped=2.0 2022-12-08 06:49:27,282 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 2022-12-08 06:49:30,189 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106329.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:49:36,165 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106336.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:49:37,031 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2725, 2.3175, 1.8579, 2.3553, 2.2207, 2.2371, 2.0390, 2.0194], device='cuda:3'), covar=tensor([0.1022, 0.1039, 0.2215, 0.0823, 0.1278, 0.0720, 0.1483, 0.1442], device='cuda:3'), in_proj_covar=tensor([0.0274, 0.0290, 0.0262, 0.0275, 0.0319, 0.0296, 0.0255, 0.0244], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 06:49:41,597 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9291, 1.8666, 1.9266, 2.2744, 2.1826, 1.1199, 1.9872, 2.0956], device='cuda:3'), covar=tensor([0.1478, 0.1028, 0.0907, 0.0532, 0.1025, 0.0982, 0.0901, 0.0708], device='cuda:3'), in_proj_covar=tensor([0.0032, 0.0031, 0.0035, 0.0029, 0.0031, 0.0044, 0.0032, 0.0035], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 06:50:06,076 INFO [train.py:873] (3/4) Epoch 15, batch 500, loss[loss=0.1287, simple_loss=0.1502, pruned_loss=0.0536, over 6928.00 frames. ], tot_loss[loss=0.1145, simple_loss=0.1483, pruned_loss=0.04037, over 1834539.62 frames. ], batch size: 100, lr: 5.34e-03, grad_scale: 8.0 2022-12-08 06:50:21,321 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.193e+02 2.064e+02 2.651e+02 3.398e+02 5.648e+02, threshold=5.301e+02, percent-clipped=3.0 2022-12-08 06:50:31,318 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106397.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:50:34,163 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3682, 1.5368, 2.5518, 1.5148, 2.4906, 2.4820, 2.0059, 2.6478], device='cuda:3'), covar=tensor([0.0270, 0.2258, 0.0392, 0.1755, 0.0534, 0.0501, 0.1004, 0.0297], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0156, 0.0159, 0.0168, 0.0166, 0.0177, 0.0132, 0.0148], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 06:50:43,405 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.3733, 5.2516, 4.9578, 5.4099, 5.0342, 4.8223, 5.4992, 5.1814], device='cuda:3'), covar=tensor([0.0619, 0.0655, 0.0727, 0.0543, 0.0569, 0.0458, 0.0490, 0.0735], device='cuda:3'), in_proj_covar=tensor([0.0140, 0.0138, 0.0143, 0.0156, 0.0143, 0.0121, 0.0165, 0.0146], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 06:50:43,763 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.09 vs. limit=2.0 2022-12-08 06:50:47,945 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8384, 1.7080, 1.9708, 1.6955, 1.9873, 1.7795, 1.6902, 1.8765], device='cuda:3'), covar=tensor([0.0600, 0.1254, 0.0420, 0.0487, 0.0446, 0.0733, 0.0339, 0.0383], device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0313, 0.0390, 0.0297, 0.0370, 0.0321, 0.0356, 0.0299], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 06:51:03,994 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106433.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:51:17,610 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106448.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:51:22,884 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106454.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:51:23,739 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106455.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:51:36,096 INFO [train.py:873] (3/4) Epoch 15, batch 600, loss[loss=0.1173, simple_loss=0.1459, pruned_loss=0.04438, over 6905.00 frames. ], tot_loss[loss=0.1141, simple_loss=0.1478, pruned_loss=0.04015, over 1832152.28 frames. ], batch size: 100, lr: 5.33e-03, grad_scale: 8.0 2022-12-08 06:51:39,614 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2022-12-08 06:51:51,214 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.287e+02 2.156e+02 2.802e+02 3.401e+02 5.738e+02, threshold=5.604e+02, percent-clipped=2.0 2022-12-08 06:52:00,105 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=106496.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:52:06,429 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=106503.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:52:07,819 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0 2022-12-08 06:52:17,383 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106515.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:52:31,579 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3682, 2.7033, 4.3955, 4.4679, 4.3788, 2.6010, 4.5289, 3.5141], device='cuda:3'), covar=tensor([0.0347, 0.1064, 0.0768, 0.0492, 0.0412, 0.1698, 0.0349, 0.0827], device='cuda:3'), in_proj_covar=tensor([0.0289, 0.0256, 0.0370, 0.0325, 0.0267, 0.0299, 0.0304, 0.0275], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 06:53:05,160 INFO [train.py:873] (3/4) Epoch 15, batch 700, loss[loss=0.09674, simple_loss=0.1252, pruned_loss=0.03415, over 5010.00 frames. ], tot_loss[loss=0.1144, simple_loss=0.1481, pruned_loss=0.04037, over 1921684.58 frames. ], batch size: 100, lr: 5.33e-03, grad_scale: 8.0 2022-12-08 06:53:08,747 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5025, 1.4036, 3.5949, 1.5681, 3.4466, 3.6344, 2.6146, 3.8886], device='cuda:3'), covar=tensor([0.0240, 0.3201, 0.0387, 0.2246, 0.0692, 0.0367, 0.0892, 0.0195], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0157, 0.0159, 0.0169, 0.0167, 0.0177, 0.0133, 0.0150], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 06:53:19,980 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.992e+01 2.009e+02 2.731e+02 3.273e+02 5.771e+02, threshold=5.463e+02, percent-clipped=3.0 2022-12-08 06:53:53,100 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106624.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:53:59,585 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.99 vs. limit=5.0 2022-12-08 06:54:32,760 INFO [train.py:873] (3/4) Epoch 15, batch 800, loss[loss=0.132, simple_loss=0.1733, pruned_loss=0.04529, over 14229.00 frames. ], tot_loss[loss=0.1146, simple_loss=0.1485, pruned_loss=0.04039, over 1994510.10 frames. ], batch size: 35, lr: 5.33e-03, grad_scale: 8.0 2022-12-08 06:54:47,970 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.357e+02 2.230e+02 2.748e+02 3.448e+02 5.498e+02, threshold=5.496e+02, percent-clipped=1.0 2022-12-08 06:54:53,533 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106692.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:55:17,370 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4976, 1.6386, 1.8905, 1.9070, 1.7856, 1.7842, 1.5260, 1.3738], device='cuda:3'), covar=tensor([0.1144, 0.1297, 0.0484, 0.0586, 0.1094, 0.0806, 0.1610, 0.1962], device='cuda:3'), in_proj_covar=tensor([0.0140, 0.0086, 0.0067, 0.0070, 0.0095, 0.0083, 0.0098, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0006, 0.0006], device='cuda:3') 2022-12-08 06:55:30,040 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106733.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:55:31,773 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8620, 2.4234, 2.6543, 1.7220, 2.3923, 2.6394, 2.8810, 2.3120], device='cuda:3'), covar=tensor([0.0703, 0.0814, 0.0970, 0.1648, 0.1187, 0.0684, 0.0659, 0.1400], device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0176, 0.0138, 0.0127, 0.0141, 0.0153, 0.0130, 0.0141], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 06:55:51,931 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.66 vs. limit=2.0 2022-12-08 06:55:57,431 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.1607, 5.0839, 4.7680, 5.2730, 4.7416, 4.5286, 5.2358, 4.9989], device='cuda:3'), covar=tensor([0.0583, 0.0590, 0.0722, 0.0504, 0.0780, 0.0472, 0.0493, 0.0681], device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0138, 0.0142, 0.0156, 0.0145, 0.0121, 0.0165, 0.0146], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 06:56:02,204 INFO [train.py:873] (3/4) Epoch 15, batch 900, loss[loss=0.1288, simple_loss=0.1335, pruned_loss=0.06201, over 1242.00 frames. ], tot_loss[loss=0.1143, simple_loss=0.1483, pruned_loss=0.04018, over 2002151.48 frames. ], batch size: 100, lr: 5.33e-03, grad_scale: 8.0 2022-12-08 06:56:12,928 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=106781.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:56:16,791 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.105e+02 2.159e+02 2.563e+02 3.204e+02 7.042e+02, threshold=5.126e+02, percent-clipped=4.0 2022-12-08 06:56:31,249 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9791, 2.0049, 3.9599, 2.7288, 3.8359, 2.0164, 2.9104, 3.8143], device='cuda:3'), covar=tensor([0.0608, 0.4202, 0.0547, 0.5492, 0.0632, 0.3314, 0.1283, 0.0405], device='cuda:3'), in_proj_covar=tensor([0.0252, 0.0207, 0.0212, 0.0277, 0.0227, 0.0209, 0.0204, 0.0213], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 06:56:38,485 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106810.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:56:51,945 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.17 vs. limit=5.0 2022-12-08 06:57:27,680 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.2413, 1.4759, 1.7701, 1.6278, 1.6133, 1.6360, 1.3575, 1.3314], device='cuda:3'), covar=tensor([0.1319, 0.1060, 0.0387, 0.0610, 0.1225, 0.0944, 0.1490, 0.1504], device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0085, 0.0067, 0.0071, 0.0096, 0.0083, 0.0099, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0006, 0.0006], device='cuda:3') 2022-12-08 06:57:30,129 INFO [train.py:873] (3/4) Epoch 15, batch 1000, loss[loss=0.09829, simple_loss=0.1428, pruned_loss=0.02688, over 14254.00 frames. ], tot_loss[loss=0.1152, simple_loss=0.1488, pruned_loss=0.04081, over 1953177.17 frames. ], batch size: 89, lr: 5.32e-03, grad_scale: 8.0 2022-12-08 06:57:44,838 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.212e+02 2.224e+02 2.651e+02 3.456e+02 5.513e+02, threshold=5.301e+02, percent-clipped=2.0 2022-12-08 06:58:13,200 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.1302, 4.9518, 4.7875, 5.1434, 4.7305, 4.5066, 5.1783, 5.0162], device='cuda:3'), covar=tensor([0.0578, 0.0754, 0.0572, 0.0532, 0.0649, 0.0472, 0.0540, 0.0606], device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0138, 0.0142, 0.0156, 0.0145, 0.0121, 0.0165, 0.0146], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 06:58:18,710 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106924.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:58:28,613 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8303, 1.0305, 1.3126, 1.2078, 0.9960, 1.2709, 1.0711, 0.8974], device='cuda:3'), covar=tensor([0.1868, 0.1231, 0.0413, 0.0560, 0.1799, 0.1007, 0.1558, 0.1410], device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0086, 0.0068, 0.0071, 0.0097, 0.0084, 0.0100, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0006, 0.0006], device='cuda:3') 2022-12-08 06:58:58,496 INFO [train.py:873] (3/4) Epoch 15, batch 1100, loss[loss=0.1634, simple_loss=0.1765, pruned_loss=0.07512, over 8618.00 frames. ], tot_loss[loss=0.1156, simple_loss=0.1489, pruned_loss=0.0411, over 1982352.72 frames. ], batch size: 100, lr: 5.32e-03, grad_scale: 8.0 2022-12-08 06:59:01,490 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=106972.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:59:13,542 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.148e+02 2.263e+02 2.855e+02 3.604e+02 7.949e+02, threshold=5.710e+02, percent-clipped=5.0 2022-12-08 06:59:19,126 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106992.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 06:59:37,942 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.58 vs. limit=2.0 2022-12-08 07:00:02,195 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=107040.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:00:28,135 INFO [train.py:873] (3/4) Epoch 15, batch 1200, loss[loss=0.113, simple_loss=0.1493, pruned_loss=0.03831, over 14294.00 frames. ], tot_loss[loss=0.116, simple_loss=0.1494, pruned_loss=0.04127, over 2000920.02 frames. ], batch size: 63, lr: 5.32e-03, grad_scale: 8.0 2022-12-08 07:00:42,788 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.313e+02 2.082e+02 2.691e+02 3.251e+02 9.387e+02, threshold=5.382e+02, percent-clipped=1.0 2022-12-08 07:01:04,289 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=107110.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:01:22,958 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9788, 1.9003, 2.1919, 1.8289, 2.1915, 1.7547, 1.6774, 1.5541], device='cuda:3'), covar=tensor([0.0480, 0.0782, 0.0319, 0.0370, 0.0343, 0.0538, 0.0537, 0.0763], device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0020, 0.0018, 0.0019, 0.0019, 0.0031, 0.0025, 0.0030], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 07:01:30,766 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107140.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 07:01:41,339 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107152.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:01:46,369 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=107158.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:01:47,753 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2022-12-08 07:01:56,717 INFO [train.py:873] (3/4) Epoch 15, batch 1300, loss[loss=0.1228, simple_loss=0.1279, pruned_loss=0.05885, over 2605.00 frames. ], tot_loss[loss=0.1157, simple_loss=0.149, pruned_loss=0.0412, over 1959598.78 frames. ], batch size: 100, lr: 5.32e-03, grad_scale: 8.0 2022-12-08 07:02:09,238 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9940, 2.1200, 1.9888, 2.1222, 1.7964, 2.0061, 2.0742, 2.0205], device='cuda:3'), covar=tensor([0.0937, 0.1177, 0.1225, 0.0921, 0.1652, 0.1024, 0.1109, 0.0915], device='cuda:3'), in_proj_covar=tensor([0.0140, 0.0138, 0.0143, 0.0157, 0.0145, 0.0121, 0.0164, 0.0146], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 07:02:12,107 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.125e+02 2.037e+02 2.513e+02 3.265e+02 6.018e+02, threshold=5.026e+02, percent-clipped=1.0 2022-12-08 07:02:20,869 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8730, 1.9295, 1.8038, 1.8473, 2.0241, 1.1976, 1.8946, 2.0421], device='cuda:3'), covar=tensor([0.1247, 0.0599, 0.0821, 0.1121, 0.0823, 0.0964, 0.0811, 0.0506], device='cuda:3'), in_proj_covar=tensor([0.0031, 0.0031, 0.0035, 0.0029, 0.0031, 0.0044, 0.0032, 0.0035], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 07:02:25,645 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107201.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 07:02:36,983 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107213.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 07:02:39,408 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8120, 0.8474, 0.7166, 0.8345, 0.8566, 0.2973, 0.7505, 0.8627], device='cuda:3'), covar=tensor([0.0502, 0.0483, 0.0543, 0.0357, 0.0322, 0.0312, 0.0894, 0.0680], device='cuda:3'), in_proj_covar=tensor([0.0031, 0.0031, 0.0034, 0.0029, 0.0031, 0.0044, 0.0032, 0.0034], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 07:02:49,792 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8276, 4.6805, 4.5185, 4.8317, 4.4599, 4.2725, 4.9117, 4.6591], device='cuda:3'), covar=tensor([0.0606, 0.0776, 0.0712, 0.0574, 0.0592, 0.0564, 0.0533, 0.0689], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0137, 0.0142, 0.0156, 0.0144, 0.0121, 0.0164, 0.0145], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 07:02:55,707 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7934, 2.7638, 2.0468, 2.8657, 2.6694, 2.7241, 2.5101, 2.2057], device='cuda:3'), covar=tensor([0.0956, 0.1045, 0.2606, 0.0801, 0.0970, 0.0934, 0.1281, 0.2399], device='cuda:3'), in_proj_covar=tensor([0.0274, 0.0291, 0.0262, 0.0274, 0.0319, 0.0299, 0.0256, 0.0243], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 07:03:26,075 INFO [train.py:873] (3/4) Epoch 15, batch 1400, loss[loss=0.1212, simple_loss=0.1433, pruned_loss=0.04953, over 5015.00 frames. ], tot_loss[loss=0.116, simple_loss=0.1491, pruned_loss=0.04146, over 1944106.79 frames. ], batch size: 100, lr: 5.31e-03, grad_scale: 4.0 2022-12-08 07:03:29,085 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.88 vs. limit=5.0 2022-12-08 07:03:38,623 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9140, 2.6087, 3.5340, 2.4124, 2.3155, 3.1644, 1.6307, 3.0250], device='cuda:3'), covar=tensor([0.1171, 0.1245, 0.0600, 0.1686, 0.2098, 0.0901, 0.3497, 0.0990], device='cuda:3'), in_proj_covar=tensor([0.0085, 0.0101, 0.0092, 0.0099, 0.0117, 0.0088, 0.0120, 0.0093], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 07:03:41,696 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.373e+02 2.191e+02 2.804e+02 3.719e+02 1.162e+03, threshold=5.608e+02, percent-clipped=5.0 2022-12-08 07:04:00,050 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3348, 3.7674, 3.3252, 3.5801, 2.5855, 3.5992, 3.4135, 1.7948], device='cuda:3'), covar=tensor([0.1818, 0.0551, 0.1165, 0.0717, 0.1064, 0.0651, 0.1015, 0.2360], device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0086, 0.0068, 0.0071, 0.0098, 0.0085, 0.0100, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 07:04:17,434 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1069, 3.0848, 2.9476, 3.2163, 2.7919, 2.8571, 3.1921, 3.0485], device='cuda:3'), covar=tensor([0.0799, 0.1035, 0.0963, 0.0707, 0.1219, 0.0816, 0.0798, 0.0919], device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0139, 0.0144, 0.0158, 0.0146, 0.0122, 0.0165, 0.0147], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 07:04:20,076 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.0144, 1.3106, 1.3804, 0.9543, 0.8492, 1.1383, 0.9357, 1.1765], device='cuda:3'), covar=tensor([0.2652, 0.3269, 0.0972, 0.3052, 0.3675, 0.1389, 0.2019, 0.1519], device='cuda:3'), in_proj_covar=tensor([0.0086, 0.0102, 0.0092, 0.0100, 0.0118, 0.0089, 0.0122, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 07:04:37,586 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 2022-12-08 07:04:54,155 INFO [train.py:873] (3/4) Epoch 15, batch 1500, loss[loss=0.1405, simple_loss=0.1634, pruned_loss=0.05878, over 14359.00 frames. ], tot_loss[loss=0.1147, simple_loss=0.1482, pruned_loss=0.04061, over 1961627.56 frames. ], batch size: 53, lr: 5.31e-03, grad_scale: 4.0 2022-12-08 07:05:10,370 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.112e+02 2.250e+02 2.599e+02 3.364e+02 1.423e+03, threshold=5.197e+02, percent-clipped=1.0 2022-12-08 07:05:48,883 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.1309, 4.7153, 4.6014, 5.1428, 4.7543, 4.4641, 5.1212, 4.2699], device='cuda:3'), covar=tensor([0.0355, 0.0899, 0.0388, 0.0398, 0.0877, 0.0549, 0.0507, 0.0516], device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0273, 0.0197, 0.0195, 0.0184, 0.0155, 0.0286, 0.0168], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 07:06:23,610 INFO [train.py:873] (3/4) Epoch 15, batch 1600, loss[loss=0.1241, simple_loss=0.1491, pruned_loss=0.04958, over 13880.00 frames. ], tot_loss[loss=0.1152, simple_loss=0.148, pruned_loss=0.04127, over 1885605.08 frames. ], batch size: 20, lr: 5.31e-03, grad_scale: 8.0 2022-12-08 07:06:33,490 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.68 vs. limit=5.0 2022-12-08 07:06:39,091 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.559e+01 2.229e+02 2.676e+02 3.216e+02 6.513e+02, threshold=5.353e+02, percent-clipped=3.0 2022-12-08 07:06:43,406 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7077, 3.5918, 3.5023, 3.7556, 3.3246, 3.3041, 3.7512, 3.6382], device='cuda:3'), covar=tensor([0.0595, 0.0848, 0.0858, 0.0611, 0.0920, 0.0710, 0.0668, 0.0716], device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0139, 0.0144, 0.0157, 0.0145, 0.0122, 0.0165, 0.0146], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 07:06:47,603 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=107496.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 07:06:58,120 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=107508.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 07:06:59,888 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107510.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:07:34,154 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9081, 2.7124, 2.7762, 2.9399, 2.7840, 2.8520, 2.9908, 2.4476], device='cuda:3'), covar=tensor([0.0646, 0.1239, 0.0549, 0.0579, 0.0858, 0.0485, 0.0641, 0.0695], device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0274, 0.0196, 0.0195, 0.0183, 0.0155, 0.0286, 0.0169], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 07:07:43,724 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.68 vs. limit=2.0 2022-12-08 07:07:50,963 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.93 vs. limit=5.0 2022-12-08 07:07:51,202 INFO [train.py:873] (3/4) Epoch 15, batch 1700, loss[loss=0.1134, simple_loss=0.1452, pruned_loss=0.04084, over 14257.00 frames. ], tot_loss[loss=0.1153, simple_loss=0.1484, pruned_loss=0.04103, over 1923540.31 frames. ], batch size: 80, lr: 5.31e-03, grad_scale: 8.0 2022-12-08 07:07:53,418 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107571.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:08:07,308 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.130e+02 2.065e+02 2.454e+02 2.883e+02 5.170e+02, threshold=4.908e+02, percent-clipped=0.0 2022-12-08 07:09:11,905 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107659.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:09:20,340 INFO [train.py:873] (3/4) Epoch 15, batch 1800, loss[loss=0.1148, simple_loss=0.1509, pruned_loss=0.03931, over 14262.00 frames. ], tot_loss[loss=0.1149, simple_loss=0.1483, pruned_loss=0.04079, over 1929495.38 frames. ], batch size: 63, lr: 5.30e-03, grad_scale: 8.0 2022-12-08 07:09:35,244 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.000e+02 2.283e+02 3.065e+02 3.850e+02 1.119e+03, threshold=6.129e+02, percent-clipped=8.0 2022-12-08 07:10:03,847 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1340, 4.1676, 4.5636, 3.8511, 4.3773, 4.5927, 1.7322, 4.0865], device='cuda:3'), covar=tensor([0.0321, 0.0344, 0.0358, 0.0432, 0.0301, 0.0220, 0.3139, 0.0309], device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0173, 0.0143, 0.0144, 0.0203, 0.0140, 0.0159, 0.0190], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 07:10:04,696 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107720.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:10:20,190 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107737.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:10:39,619 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.7305, 4.0388, 3.2416, 5.1137, 4.5302, 4.8905, 4.2664, 3.4242], device='cuda:3'), covar=tensor([0.0778, 0.1213, 0.3344, 0.0393, 0.0778, 0.0914, 0.1123, 0.3111], device='cuda:3'), in_proj_covar=tensor([0.0278, 0.0293, 0.0264, 0.0278, 0.0322, 0.0301, 0.0259, 0.0244], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 07:10:48,098 INFO [train.py:873] (3/4) Epoch 15, batch 1900, loss[loss=0.1357, simple_loss=0.1453, pruned_loss=0.06306, over 2677.00 frames. ], tot_loss[loss=0.1145, simple_loss=0.148, pruned_loss=0.0405, over 1906065.24 frames. ], batch size: 100, lr: 5.30e-03, grad_scale: 8.0 2022-12-08 07:11:04,226 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.371e+02 2.201e+02 2.753e+02 3.141e+02 7.986e+02, threshold=5.507e+02, percent-clipped=3.0 2022-12-08 07:11:06,340 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3720, 2.0359, 2.4312, 2.4612, 2.2724, 1.9857, 2.4143, 2.1942], device='cuda:3'), covar=tensor([0.0367, 0.0807, 0.0395, 0.0384, 0.0494, 0.1022, 0.0423, 0.0552], device='cuda:3'), in_proj_covar=tensor([0.0295, 0.0259, 0.0378, 0.0332, 0.0272, 0.0307, 0.0310, 0.0283], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 07:11:06,995 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8131, 2.7308, 2.0735, 2.8295, 2.7122, 2.7290, 2.4860, 2.1683], device='cuda:3'), covar=tensor([0.0948, 0.1233, 0.2774, 0.0856, 0.1007, 0.0884, 0.1385, 0.2752], device='cuda:3'), in_proj_covar=tensor([0.0278, 0.0292, 0.0264, 0.0278, 0.0322, 0.0300, 0.0258, 0.0244], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 07:11:12,112 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=107796.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 07:11:13,994 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107798.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 07:11:22,875 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=107808.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:11:54,592 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=107844.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 07:12:05,288 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=107856.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:12:07,051 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0700, 3.9262, 3.2493, 3.4741, 3.7924, 3.9955, 4.1136, 4.0259], device='cuda:3'), covar=tensor([0.1075, 0.0699, 0.3260, 0.3590, 0.1293, 0.1278, 0.1187, 0.1230], device='cuda:3'), in_proj_covar=tensor([0.0385, 0.0262, 0.0447, 0.0566, 0.0344, 0.0437, 0.0392, 0.0386], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 07:12:07,999 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8571, 2.0043, 1.9266, 1.7976, 1.8800, 1.7038, 1.3906, 1.3734], device='cuda:3'), covar=tensor([0.0201, 0.0326, 0.0342, 0.0361, 0.0257, 0.0304, 0.0298, 0.0458], device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0020, 0.0018, 0.0019, 0.0019, 0.0031, 0.0025, 0.0030], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 07:12:13,853 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=107866.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:12:16,484 INFO [train.py:873] (3/4) Epoch 15, batch 2000, loss[loss=0.09331, simple_loss=0.1267, pruned_loss=0.02998, over 5991.00 frames. ], tot_loss[loss=0.1146, simple_loss=0.1483, pruned_loss=0.04048, over 1931209.03 frames. ], batch size: 100, lr: 5.30e-03, grad_scale: 8.0 2022-12-08 07:12:21,487 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.72 vs. limit=5.0 2022-12-08 07:12:31,885 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.235e+02 1.995e+02 2.775e+02 3.641e+02 9.267e+02, threshold=5.550e+02, percent-clipped=5.0 2022-12-08 07:13:35,984 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.3238, 4.9102, 4.8300, 5.3470, 4.9661, 4.4680, 5.2833, 4.4664], device='cuda:3'), covar=tensor([0.0296, 0.0859, 0.0385, 0.0379, 0.0799, 0.0524, 0.0494, 0.0458], device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0273, 0.0196, 0.0194, 0.0184, 0.0154, 0.0284, 0.0168], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 07:13:43,635 INFO [train.py:873] (3/4) Epoch 15, batch 2100, loss[loss=0.1159, simple_loss=0.1457, pruned_loss=0.04302, over 14205.00 frames. ], tot_loss[loss=0.1146, simple_loss=0.1481, pruned_loss=0.04052, over 1886035.68 frames. ], batch size: 94, lr: 5.30e-03, grad_scale: 8.0 2022-12-08 07:13:50,276 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8522, 2.1411, 2.2610, 2.0469, 1.8049, 1.8321, 1.7331, 1.4854], device='cuda:3'), covar=tensor([0.0376, 0.0295, 0.0353, 0.0391, 0.0489, 0.0397, 0.0346, 0.0489], device='cuda:3'), in_proj_covar=tensor([0.0019, 0.0020, 0.0017, 0.0019, 0.0019, 0.0030, 0.0025, 0.0030], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 07:13:59,441 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.122e+02 1.955e+02 2.546e+02 3.190e+02 6.620e+02, threshold=5.091e+02, percent-clipped=1.0 2022-12-08 07:14:08,488 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=7.01 vs. limit=5.0 2022-12-08 07:14:15,531 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.64 vs. limit=5.0 2022-12-08 07:14:24,593 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108015.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:15:05,924 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9438, 1.9096, 1.9434, 2.0339, 1.9287, 1.6439, 1.3399, 1.7188], device='cuda:3'), covar=tensor([0.0885, 0.0786, 0.0725, 0.0487, 0.0706, 0.1619, 0.2799, 0.0863], device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0170, 0.0142, 0.0142, 0.0200, 0.0138, 0.0156, 0.0187], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 07:15:11,289 INFO [train.py:873] (3/4) Epoch 15, batch 2200, loss[loss=0.1058, simple_loss=0.1397, pruned_loss=0.03597, over 13919.00 frames. ], tot_loss[loss=0.1149, simple_loss=0.1481, pruned_loss=0.04084, over 1900243.81 frames. ], batch size: 23, lr: 5.29e-03, grad_scale: 8.0 2022-12-08 07:15:26,510 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.457e+02 2.145e+02 2.568e+02 3.392e+02 6.463e+02, threshold=5.135e+02, percent-clipped=2.0 2022-12-08 07:15:32,325 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108093.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 07:15:52,821 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1044, 1.9710, 4.0994, 2.6829, 3.9152, 1.8637, 2.8563, 3.8954], device='cuda:3'), covar=tensor([0.0721, 0.4461, 0.0549, 0.6517, 0.0763, 0.3926, 0.1525, 0.0557], device='cuda:3'), in_proj_covar=tensor([0.0254, 0.0205, 0.0213, 0.0278, 0.0231, 0.0208, 0.0204, 0.0215], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 07:16:18,003 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7496, 1.3596, 2.5243, 2.2510, 2.4363, 2.5452, 1.6977, 2.5149], device='cuda:3'), covar=tensor([0.0944, 0.1311, 0.0188, 0.0444, 0.0425, 0.0204, 0.0700, 0.0244], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0158, 0.0130, 0.0170, 0.0147, 0.0140, 0.0123, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 07:16:36,137 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=108166.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:16:39,020 INFO [train.py:873] (3/4) Epoch 15, batch 2300, loss[loss=0.1051, simple_loss=0.1464, pruned_loss=0.03196, over 14287.00 frames. ], tot_loss[loss=0.1142, simple_loss=0.1476, pruned_loss=0.04038, over 1940494.00 frames. ], batch size: 25, lr: 5.29e-03, grad_scale: 8.0 2022-12-08 07:16:43,749 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.6016, 4.4016, 4.2424, 4.6699, 4.2378, 4.0030, 4.6850, 4.5426], device='cuda:3'), covar=tensor([0.0700, 0.0843, 0.0812, 0.0582, 0.0823, 0.0600, 0.0595, 0.0651], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0137, 0.0141, 0.0154, 0.0142, 0.0120, 0.0162, 0.0143], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 07:16:54,803 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.282e+02 2.030e+02 2.596e+02 3.385e+02 5.799e+02, threshold=5.191e+02, percent-clipped=3.0 2022-12-08 07:17:04,514 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5834, 3.4450, 3.1934, 2.4123, 3.1126, 3.4360, 3.7918, 2.9436], device='cuda:3'), covar=tensor([0.0713, 0.0981, 0.0937, 0.1384, 0.0903, 0.0632, 0.0655, 0.1195], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0175, 0.0138, 0.0125, 0.0140, 0.0151, 0.0130, 0.0139], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 07:17:18,546 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=108214.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:18:06,535 INFO [train.py:873] (3/4) Epoch 15, batch 2400, loss[loss=0.1227, simple_loss=0.1576, pruned_loss=0.04388, over 14257.00 frames. ], tot_loss[loss=0.1155, simple_loss=0.1487, pruned_loss=0.04119, over 1937516.25 frames. ], batch size: 57, lr: 5.29e-03, grad_scale: 8.0 2022-12-08 07:18:22,283 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.295e+02 2.197e+02 2.869e+02 3.470e+02 6.788e+02, threshold=5.738e+02, percent-clipped=3.0 2022-12-08 07:18:25,354 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5670, 1.9217, 2.5105, 1.9961, 2.5704, 2.4246, 2.2767, 2.2625], device='cuda:3'), covar=tensor([0.0701, 0.2657, 0.0853, 0.1699, 0.0545, 0.1086, 0.0935, 0.1512], device='cuda:3'), in_proj_covar=tensor([0.0354, 0.0317, 0.0395, 0.0300, 0.0375, 0.0323, 0.0363, 0.0303], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 07:18:34,375 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5875, 2.3723, 3.0453, 1.9014, 2.0060, 2.7502, 1.5103, 2.4971], device='cuda:3'), covar=tensor([0.1018, 0.1391, 0.0565, 0.2151, 0.2485, 0.0921, 0.3708, 0.1245], device='cuda:3'), in_proj_covar=tensor([0.0086, 0.0101, 0.0093, 0.0099, 0.0118, 0.0089, 0.0122, 0.0093], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 07:18:47,660 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=108315.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:18:55,847 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=108325.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:19:22,143 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7562, 1.7468, 1.8728, 1.7319, 1.8249, 1.6533, 1.7351, 1.2872], device='cuda:3'), covar=tensor([0.0190, 0.0358, 0.0281, 0.0238, 0.0220, 0.0315, 0.0250, 0.0479], device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0020, 0.0018, 0.0019, 0.0019, 0.0031, 0.0025, 0.0030], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 07:19:29,551 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=108363.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:19:34,872 INFO [train.py:873] (3/4) Epoch 15, batch 2500, loss[loss=0.1432, simple_loss=0.1396, pruned_loss=0.0734, over 1304.00 frames. ], tot_loss[loss=0.1155, simple_loss=0.1486, pruned_loss=0.04117, over 1944059.53 frames. ], batch size: 100, lr: 5.29e-03, grad_scale: 8.0 2022-12-08 07:19:43,395 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.0863, 1.2859, 1.3774, 1.0084, 0.8581, 1.1214, 0.8815, 1.2184], device='cuda:3'), covar=tensor([0.2060, 0.2533, 0.0891, 0.2485, 0.2865, 0.1081, 0.1429, 0.1059], device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0102, 0.0093, 0.0100, 0.0119, 0.0089, 0.0123, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 07:19:50,376 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=108386.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:19:50,976 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.050e+02 2.088e+02 2.601e+02 3.226e+02 6.971e+02, threshold=5.203e+02, percent-clipped=3.0 2022-12-08 07:19:56,286 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=108393.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:20:31,662 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=108432.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:20:38,937 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=108441.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:20:44,047 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.0365, 1.1178, 0.9384, 1.0568, 1.1875, 0.7376, 0.9181, 1.1032], device='cuda:3'), covar=tensor([0.0428, 0.0545, 0.0578, 0.0408, 0.0340, 0.0596, 0.0983, 0.0665], device='cuda:3'), in_proj_covar=tensor([0.0032, 0.0032, 0.0035, 0.0030, 0.0032, 0.0045, 0.0033, 0.0035], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 07:21:01,704 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 2022-12-08 07:21:03,651 INFO [train.py:873] (3/4) Epoch 15, batch 2600, loss[loss=0.1166, simple_loss=0.1213, pruned_loss=0.05599, over 2629.00 frames. ], tot_loss[loss=0.1148, simple_loss=0.148, pruned_loss=0.04076, over 1892018.29 frames. ], batch size: 100, lr: 5.28e-03, grad_scale: 8.0 2022-12-08 07:21:17,354 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.86 vs. limit=5.0 2022-12-08 07:21:19,264 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.775e+01 2.028e+02 2.663e+02 3.413e+02 6.606e+02, threshold=5.327e+02, percent-clipped=6.0 2022-12-08 07:21:24,916 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=108493.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 07:22:08,777 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9550, 1.8947, 4.5761, 2.3410, 4.5031, 4.9042, 4.5816, 5.3751], device='cuda:3'), covar=tensor([0.0191, 0.2890, 0.0397, 0.2077, 0.0259, 0.0346, 0.0253, 0.0131], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0157, 0.0161, 0.0170, 0.0167, 0.0177, 0.0134, 0.0150], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 07:22:17,005 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9262, 1.6926, 1.9505, 1.7191, 1.9712, 1.7939, 1.6584, 1.9268], device='cuda:3'), covar=tensor([0.0616, 0.1380, 0.0426, 0.0426, 0.0567, 0.0834, 0.0276, 0.0344], device='cuda:3'), in_proj_covar=tensor([0.0357, 0.0317, 0.0397, 0.0301, 0.0377, 0.0324, 0.0366, 0.0305], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 07:22:21,474 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.57 vs. limit=2.0 2022-12-08 07:22:25,304 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8561, 3.9723, 4.2320, 3.7603, 4.0305, 4.1739, 1.5315, 3.7914], device='cuda:3'), covar=tensor([0.0320, 0.0343, 0.0306, 0.0403, 0.0330, 0.0258, 0.3251, 0.0312], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0173, 0.0145, 0.0145, 0.0204, 0.0140, 0.0159, 0.0192], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 07:22:32,113 INFO [train.py:873] (3/4) Epoch 15, batch 2700, loss[loss=0.1073, simple_loss=0.1472, pruned_loss=0.0337, over 14295.00 frames. ], tot_loss[loss=0.1139, simple_loss=0.1479, pruned_loss=0.03999, over 1955096.49 frames. ], batch size: 25, lr: 5.28e-03, grad_scale: 4.0 2022-12-08 07:22:39,972 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=108577.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:22:49,625 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.289e+02 2.154e+02 2.629e+02 3.316e+02 6.561e+02, threshold=5.258e+02, percent-clipped=1.0 2022-12-08 07:23:33,958 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=108638.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:24:01,227 INFO [train.py:873] (3/4) Epoch 15, batch 2800, loss[loss=0.1197, simple_loss=0.1545, pruned_loss=0.04245, over 14241.00 frames. ], tot_loss[loss=0.1131, simple_loss=0.1472, pruned_loss=0.03947, over 1956519.83 frames. ], batch size: 69, lr: 5.28e-03, grad_scale: 8.0 2022-12-08 07:24:12,084 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108681.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:24:17,715 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.202e+02 2.137e+02 2.731e+02 3.357e+02 8.128e+02, threshold=5.462e+02, percent-clipped=3.0 2022-12-08 07:24:50,707 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8305, 1.8163, 1.6214, 1.8836, 1.7302, 1.7731, 1.7762, 1.7371], device='cuda:3'), covar=tensor([0.1246, 0.1036, 0.2053, 0.0918, 0.1203, 0.0757, 0.1711, 0.1197], device='cuda:3'), in_proj_covar=tensor([0.0275, 0.0288, 0.0261, 0.0277, 0.0320, 0.0298, 0.0254, 0.0240], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:3') 2022-12-08 07:25:11,035 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.0114, 4.8682, 4.6589, 5.0884, 4.5871, 4.3940, 5.1075, 4.8766], device='cuda:3'), covar=tensor([0.0687, 0.0805, 0.0752, 0.0601, 0.0749, 0.0589, 0.0574, 0.0728], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0138, 0.0142, 0.0155, 0.0142, 0.0121, 0.0162, 0.0143], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 07:25:27,690 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2022-12-08 07:25:29,354 INFO [train.py:873] (3/4) Epoch 15, batch 2900, loss[loss=0.1049, simple_loss=0.145, pruned_loss=0.03237, over 11281.00 frames. ], tot_loss[loss=0.1131, simple_loss=0.1472, pruned_loss=0.03944, over 1929079.94 frames. ], batch size: 14, lr: 5.28e-03, grad_scale: 8.0 2022-12-08 07:25:33,623 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7752, 3.5135, 3.4219, 3.7893, 3.5158, 3.7597, 3.8456, 3.1637], device='cuda:3'), covar=tensor([0.0577, 0.1034, 0.0526, 0.0513, 0.0902, 0.0352, 0.0555, 0.0614], device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0274, 0.0196, 0.0195, 0.0184, 0.0153, 0.0282, 0.0168], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 07:25:33,641 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3195, 3.0996, 2.7856, 3.0248, 3.2144, 3.2534, 3.2978, 3.3079], device='cuda:3'), covar=tensor([0.0961, 0.0644, 0.2341, 0.2599, 0.0927, 0.1093, 0.1072, 0.0889], device='cuda:3'), in_proj_covar=tensor([0.0383, 0.0260, 0.0444, 0.0558, 0.0338, 0.0437, 0.0385, 0.0379], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 07:25:37,081 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8309, 3.0583, 4.6306, 3.5616, 4.5523, 4.5499, 4.3176, 4.0882], device='cuda:3'), covar=tensor([0.0727, 0.2914, 0.0855, 0.1698, 0.0855, 0.0817, 0.1602, 0.1601], device='cuda:3'), in_proj_covar=tensor([0.0354, 0.0316, 0.0392, 0.0299, 0.0374, 0.0322, 0.0363, 0.0302], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 07:25:45,502 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.212e+02 2.119e+02 2.529e+02 3.118e+02 5.786e+02, threshold=5.058e+02, percent-clipped=1.0 2022-12-08 07:25:45,650 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108788.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 07:25:48,192 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.0770, 1.1986, 1.3303, 0.9611, 0.8609, 1.0967, 0.9428, 1.1821], device='cuda:3'), covar=tensor([0.2188, 0.3587, 0.1387, 0.2689, 0.3818, 0.1522, 0.2197, 0.1443], device='cuda:3'), in_proj_covar=tensor([0.0086, 0.0102, 0.0094, 0.0099, 0.0119, 0.0089, 0.0122, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 07:25:48,214 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=108791.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:25:55,094 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2022-12-08 07:26:00,009 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.29 vs. limit=5.0 2022-12-08 07:26:42,230 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=108852.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:26:56,662 INFO [train.py:873] (3/4) Epoch 15, batch 3000, loss[loss=0.09931, simple_loss=0.1353, pruned_loss=0.03165, over 14285.00 frames. ], tot_loss[loss=0.1127, simple_loss=0.1466, pruned_loss=0.03933, over 1913788.55 frames. ], batch size: 46, lr: 5.27e-03, grad_scale: 8.0 2022-12-08 07:26:56,662 INFO [train.py:896] (3/4) Computing validation loss 2022-12-08 07:27:05,082 INFO [train.py:905] (3/4) Epoch 15, validation: loss=0.1368, simple_loss=0.1737, pruned_loss=0.04998, over 857387.00 frames. 2022-12-08 07:27:05,083 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-08 07:27:21,818 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.554e+02 2.224e+02 2.739e+02 3.295e+02 6.855e+02, threshold=5.478e+02, percent-clipped=4.0 2022-12-08 07:27:48,006 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7793, 1.7216, 1.5639, 1.7975, 1.9097, 1.1993, 1.5052, 1.6168], device='cuda:3'), covar=tensor([0.0597, 0.0730, 0.1013, 0.0559, 0.0608, 0.0784, 0.0749, 0.1072], device='cuda:3'), in_proj_covar=tensor([0.0032, 0.0031, 0.0035, 0.0029, 0.0031, 0.0045, 0.0032, 0.0035], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 07:28:00,920 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108933.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:28:25,945 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.71 vs. limit=2.0 2022-12-08 07:28:32,783 INFO [train.py:873] (3/4) Epoch 15, batch 3100, loss[loss=0.08315, simple_loss=0.1247, pruned_loss=0.02077, over 13997.00 frames. ], tot_loss[loss=0.1128, simple_loss=0.1469, pruned_loss=0.03933, over 1935279.02 frames. ], batch size: 19, lr: 5.27e-03, grad_scale: 8.0 2022-12-08 07:28:42,830 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=108981.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:28:48,816 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.271e+02 2.020e+02 2.542e+02 3.088e+02 8.012e+02, threshold=5.083e+02, percent-clipped=4.0 2022-12-08 07:29:25,239 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=109029.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:29:32,509 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109037.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 07:30:00,245 INFO [train.py:873] (3/4) Epoch 15, batch 3200, loss[loss=0.09414, simple_loss=0.1369, pruned_loss=0.02569, over 14597.00 frames. ], tot_loss[loss=0.1134, simple_loss=0.1473, pruned_loss=0.03972, over 1935438.39 frames. ], batch size: 30, lr: 5.27e-03, grad_scale: 8.0 2022-12-08 07:30:16,994 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.196e+02 2.174e+02 2.531e+02 2.915e+02 6.335e+02, threshold=5.062e+02, percent-clipped=1.0 2022-12-08 07:30:17,207 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109088.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:30:25,655 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109098.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 07:30:26,360 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4343, 2.4319, 2.5471, 2.4244, 2.5001, 2.1149, 1.4224, 2.2247], device='cuda:3'), covar=tensor([0.0590, 0.0528, 0.0470, 0.0446, 0.0456, 0.1422, 0.2521, 0.0468], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0172, 0.0145, 0.0144, 0.0204, 0.0140, 0.0159, 0.0192], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 07:30:31,742 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1564, 2.9384, 2.2682, 3.2341, 3.1249, 3.1401, 2.7958, 2.3647], device='cuda:3'), covar=tensor([0.0993, 0.1475, 0.3460, 0.0773, 0.1033, 0.1152, 0.1570, 0.3163], device='cuda:3'), in_proj_covar=tensor([0.0281, 0.0292, 0.0266, 0.0281, 0.0324, 0.0305, 0.0260, 0.0245], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 07:30:58,808 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=109136.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:31:08,435 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109147.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:31:27,858 INFO [train.py:873] (3/4) Epoch 15, batch 3300, loss[loss=0.1008, simple_loss=0.142, pruned_loss=0.02979, over 14588.00 frames. ], tot_loss[loss=0.1129, simple_loss=0.1471, pruned_loss=0.0394, over 1956975.00 frames. ], batch size: 21, lr: 5.27e-03, grad_scale: 8.0 2022-12-08 07:31:44,097 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.026e+02 2.178e+02 2.677e+02 3.330e+02 7.070e+02, threshold=5.354e+02, percent-clipped=2.0 2022-12-08 07:31:47,495 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2022-12-08 07:31:49,457 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109194.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:32:23,689 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109233.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:32:37,905 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8093, 4.5726, 4.2254, 4.4568, 4.5092, 4.7165, 4.8182, 4.7904], device='cuda:3'), covar=tensor([0.0665, 0.0492, 0.2151, 0.2824, 0.0745, 0.0814, 0.0765, 0.0727], device='cuda:3'), in_proj_covar=tensor([0.0382, 0.0261, 0.0443, 0.0560, 0.0339, 0.0439, 0.0387, 0.0382], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 07:32:38,876 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0431, 2.7035, 2.7930, 1.9171, 2.5171, 2.7290, 3.0342, 2.4230], device='cuda:3'), covar=tensor([0.0547, 0.0652, 0.0837, 0.1292, 0.0891, 0.0655, 0.0599, 0.1130], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0171, 0.0138, 0.0125, 0.0139, 0.0151, 0.0128, 0.0138], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 07:32:43,167 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109255.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:32:55,127 INFO [train.py:873] (3/4) Epoch 15, batch 3400, loss[loss=0.09789, simple_loss=0.1434, pruned_loss=0.0262, over 13930.00 frames. ], tot_loss[loss=0.1129, simple_loss=0.147, pruned_loss=0.03944, over 1985795.05 frames. ], batch size: 23, lr: 5.26e-03, grad_scale: 4.0 2022-12-08 07:33:05,763 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=109281.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:33:12,975 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.873e+01 2.037e+02 2.446e+02 3.255e+02 7.608e+02, threshold=4.893e+02, percent-clipped=4.0 2022-12-08 07:33:17,186 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109294.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:33:38,176 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109318.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:33:58,652 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7757, 2.4583, 3.6214, 2.7667, 3.6210, 3.5454, 3.3326, 2.9637], device='cuda:3'), covar=tensor([0.0871, 0.3180, 0.1196, 0.2151, 0.0902, 0.1070, 0.1470, 0.2079], device='cuda:3'), in_proj_covar=tensor([0.0355, 0.0318, 0.0394, 0.0303, 0.0377, 0.0324, 0.0363, 0.0305], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 07:34:01,104 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8857, 0.8737, 0.7707, 0.8608, 0.8813, 0.5801, 0.6180, 0.6546], device='cuda:3'), covar=tensor([0.0129, 0.0121, 0.0111, 0.0117, 0.0123, 0.0208, 0.0158, 0.0237], device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0020, 0.0018, 0.0020, 0.0019, 0.0031, 0.0025, 0.0030], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 07:34:10,927 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109355.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:34:13,815 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.87 vs. limit=5.0 2022-12-08 07:34:23,256 INFO [train.py:873] (3/4) Epoch 15, batch 3500, loss[loss=0.1034, simple_loss=0.1304, pruned_loss=0.03822, over 4979.00 frames. ], tot_loss[loss=0.1143, simple_loss=0.1478, pruned_loss=0.04041, over 1974794.32 frames. ], batch size: 100, lr: 5.26e-03, grad_scale: 4.0 2022-12-08 07:34:32,299 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109379.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 07:34:40,318 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.249e+02 2.264e+02 2.745e+02 3.644e+02 1.157e+03, threshold=5.490e+02, percent-clipped=6.0 2022-12-08 07:34:44,017 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109393.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 07:35:15,320 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7394, 1.7670, 1.5962, 1.9143, 1.7962, 1.8099, 1.7678, 1.7081], device='cuda:3'), covar=tensor([0.1176, 0.0999, 0.2050, 0.0657, 0.1083, 0.0630, 0.1499, 0.1075], device='cuda:3'), in_proj_covar=tensor([0.0276, 0.0289, 0.0261, 0.0276, 0.0320, 0.0301, 0.0257, 0.0244], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 07:35:20,590 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4170, 1.9786, 1.9474, 2.2369, 2.0905, 1.3002, 2.0915, 2.3213], device='cuda:3'), covar=tensor([0.0808, 0.1111, 0.1163, 0.0986, 0.0868, 0.0987, 0.0793, 0.0503], device='cuda:3'), in_proj_covar=tensor([0.0032, 0.0032, 0.0035, 0.0029, 0.0032, 0.0045, 0.0033, 0.0036], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 07:35:31,949 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109447.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:35:50,856 INFO [train.py:873] (3/4) Epoch 15, batch 3600, loss[loss=0.08631, simple_loss=0.1271, pruned_loss=0.02277, over 14265.00 frames. ], tot_loss[loss=0.1139, simple_loss=0.1472, pruned_loss=0.04033, over 1905692.90 frames. ], batch size: 35, lr: 5.26e-03, grad_scale: 8.0 2022-12-08 07:36:08,874 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.258e+02 2.104e+02 2.724e+02 3.497e+02 9.894e+02, threshold=5.448e+02, percent-clipped=5.0 2022-12-08 07:36:14,324 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=109495.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:37:02,977 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109550.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:37:19,586 INFO [train.py:873] (3/4) Epoch 15, batch 3700, loss[loss=0.1076, simple_loss=0.149, pruned_loss=0.0331, over 14222.00 frames. ], tot_loss[loss=0.1128, simple_loss=0.1466, pruned_loss=0.03952, over 1893098.63 frames. ], batch size: 37, lr: 5.26e-03, grad_scale: 8.0 2022-12-08 07:37:25,788 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.59 vs. limit=2.0 2022-12-08 07:37:28,755 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109579.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:37:37,109 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.397e+02 2.068e+02 2.547e+02 3.153e+02 5.371e+02, threshold=5.093e+02, percent-clipped=0.0 2022-12-08 07:37:51,716 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9278, 1.6322, 4.1194, 3.8880, 3.9261, 4.2551, 3.6949, 4.2380], device='cuda:3'), covar=tensor([0.1752, 0.1801, 0.0165, 0.0276, 0.0268, 0.0191, 0.0306, 0.0174], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0159, 0.0130, 0.0171, 0.0147, 0.0141, 0.0124, 0.0125], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 07:38:23,084 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109640.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:38:23,446 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.66 vs. limit=2.0 2022-12-08 07:38:32,125 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109650.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:38:39,883 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109659.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:38:48,152 INFO [train.py:873] (3/4) Epoch 15, batch 3800, loss[loss=0.1045, simple_loss=0.1447, pruned_loss=0.0321, over 14212.00 frames. ], tot_loss[loss=0.1132, simple_loss=0.1469, pruned_loss=0.03976, over 1949073.89 frames. ], batch size: 35, lr: 5.26e-03, grad_scale: 8.0 2022-12-08 07:38:53,058 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109674.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 07:39:06,072 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.110e+02 2.193e+02 2.777e+02 3.474e+02 7.448e+02, threshold=5.555e+02, percent-clipped=3.0 2022-12-08 07:39:10,158 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109693.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 07:39:31,392 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6692, 1.6943, 1.7645, 1.4890, 1.5683, 1.5238, 1.3208, 1.1922], device='cuda:3'), covar=tensor([0.0185, 0.0294, 0.0202, 0.0212, 0.0223, 0.0299, 0.0236, 0.0372], device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0021, 0.0018, 0.0020, 0.0019, 0.0032, 0.0026, 0.0031], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 07:39:33,979 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109720.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:39:52,263 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=109741.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 07:39:55,398 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2022-12-08 07:40:16,359 INFO [train.py:873] (3/4) Epoch 15, batch 3900, loss[loss=0.1387, simple_loss=0.1619, pruned_loss=0.05776, over 11146.00 frames. ], tot_loss[loss=0.1125, simple_loss=0.1466, pruned_loss=0.03922, over 2017502.92 frames. ], batch size: 100, lr: 5.25e-03, grad_scale: 8.0 2022-12-08 07:40:34,143 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.655e+02 2.235e+02 2.763e+02 3.267e+02 1.043e+03, threshold=5.525e+02, percent-clipped=3.0 2022-12-08 07:40:43,281 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8021, 2.5089, 4.8235, 3.1748, 4.5983, 2.3933, 3.4752, 4.4923], device='cuda:3'), covar=tensor([0.0475, 0.3754, 0.0331, 0.6406, 0.0508, 0.3178, 0.1373, 0.0379], device='cuda:3'), in_proj_covar=tensor([0.0251, 0.0205, 0.0213, 0.0275, 0.0230, 0.0206, 0.0206, 0.0216], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 07:41:28,462 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109850.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:41:45,244 INFO [train.py:873] (3/4) Epoch 15, batch 4000, loss[loss=0.114, simple_loss=0.1479, pruned_loss=0.04004, over 14539.00 frames. ], tot_loss[loss=0.1128, simple_loss=0.1469, pruned_loss=0.03934, over 2001635.37 frames. ], batch size: 49, lr: 5.25e-03, grad_scale: 8.0 2022-12-08 07:41:57,529 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9009, 2.7232, 2.4460, 2.6155, 2.7948, 2.8247, 2.8586, 2.8849], device='cuda:3'), covar=tensor([0.1138, 0.0773, 0.2561, 0.2919, 0.1023, 0.1281, 0.1408, 0.0969], device='cuda:3'), in_proj_covar=tensor([0.0381, 0.0261, 0.0442, 0.0555, 0.0335, 0.0437, 0.0385, 0.0383], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 07:42:02,603 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.030e+02 2.144e+02 2.556e+02 3.390e+02 6.366e+02, threshold=5.112e+02, percent-clipped=4.0 2022-12-08 07:42:10,893 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=109898.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:42:43,622 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109935.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:42:56,984 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109950.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:43:10,926 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2292, 2.6275, 2.5827, 2.6877, 2.1537, 2.6941, 2.4312, 1.3138], device='cuda:3'), covar=tensor([0.0936, 0.0661, 0.0870, 0.0539, 0.0930, 0.0538, 0.0946, 0.2254], device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0087, 0.0068, 0.0072, 0.0098, 0.0086, 0.0100, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 07:43:13,347 INFO [train.py:873] (3/4) Epoch 15, batch 4100, loss[loss=0.1165, simple_loss=0.1549, pruned_loss=0.03907, over 14160.00 frames. ], tot_loss[loss=0.114, simple_loss=0.1479, pruned_loss=0.04009, over 1997814.79 frames. ], batch size: 99, lr: 5.25e-03, grad_scale: 8.0 2022-12-08 07:43:18,160 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109974.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:43:20,137 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4271, 1.4374, 3.4746, 1.6794, 3.3053, 3.5640, 2.4458, 3.7848], device='cuda:3'), covar=tensor([0.0239, 0.3270, 0.0425, 0.2375, 0.0898, 0.0448, 0.0907, 0.0200], device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0157, 0.0160, 0.0170, 0.0169, 0.0179, 0.0134, 0.0152], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 07:43:31,526 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.583e+01 2.156e+02 2.609e+02 3.160e+02 7.009e+02, threshold=5.218e+02, percent-clipped=5.0 2022-12-08 07:43:39,875 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=109998.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:43:59,014 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110015.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:44:00,743 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110017.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:44:05,369 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110022.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:44:06,230 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5294, 3.4524, 3.3080, 3.6288, 3.2106, 3.1216, 3.6030, 3.5078], device='cuda:3'), covar=tensor([0.0725, 0.0924, 0.0890, 0.0608, 0.0950, 0.0759, 0.0640, 0.0690], device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0139, 0.0143, 0.0157, 0.0144, 0.0122, 0.0167, 0.0145], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 07:44:24,350 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110044.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:44:44,326 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.60 vs. limit=5.0 2022-12-08 07:44:46,251 INFO [train.py:873] (3/4) Epoch 15, batch 4200, loss[loss=0.1005, simple_loss=0.1419, pruned_loss=0.02951, over 14266.00 frames. ], tot_loss[loss=0.1135, simple_loss=0.1479, pruned_loss=0.03955, over 2013218.50 frames. ], batch size: 60, lr: 5.25e-03, grad_scale: 8.0 2022-12-08 07:44:54,505 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110078.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:45:03,731 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.112e+02 2.176e+02 2.682e+02 3.403e+02 7.328e+02, threshold=5.365e+02, percent-clipped=6.0 2022-12-08 07:45:18,624 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110105.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:45:22,779 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110110.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:45:27,455 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110115.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:45:38,386 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4134, 3.1350, 2.5326, 3.4853, 3.3506, 3.3692, 2.9630, 2.4948], device='cuda:3'), covar=tensor([0.0768, 0.1294, 0.2812, 0.0645, 0.0841, 0.0925, 0.1340, 0.2928], device='cuda:3'), in_proj_covar=tensor([0.0276, 0.0289, 0.0261, 0.0278, 0.0321, 0.0300, 0.0257, 0.0244], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 07:46:13,634 INFO [train.py:873] (3/4) Epoch 15, batch 4300, loss[loss=0.1115, simple_loss=0.1441, pruned_loss=0.03941, over 14410.00 frames. ], tot_loss[loss=0.114, simple_loss=0.148, pruned_loss=0.03998, over 2037804.89 frames. ], batch size: 73, lr: 5.24e-03, grad_scale: 4.0 2022-12-08 07:46:15,686 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110171.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:46:16,517 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8643, 3.1887, 2.6794, 2.9719, 2.4371, 3.1878, 2.9807, 1.5356], device='cuda:3'), covar=tensor([0.1400, 0.0663, 0.1759, 0.0780, 0.0953, 0.0837, 0.1131, 0.2201], device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0089, 0.0069, 0.0073, 0.0099, 0.0086, 0.0101, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 07:46:20,273 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110176.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:46:32,112 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.012e+02 2.281e+02 2.830e+02 3.301e+02 1.295e+03, threshold=5.661e+02, percent-clipped=5.0 2022-12-08 07:46:45,449 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.74 vs. limit=2.0 2022-12-08 07:46:55,713 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8820, 4.8560, 5.2816, 4.4466, 5.0716, 5.3855, 2.1201, 4.7153], device='cuda:3'), covar=tensor([0.0251, 0.0268, 0.0351, 0.0385, 0.0264, 0.0098, 0.2816, 0.0264], device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0172, 0.0144, 0.0143, 0.0202, 0.0139, 0.0157, 0.0192], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 07:47:12,438 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110235.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:47:24,316 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.7811, 2.6244, 4.8268, 3.3669, 4.5770, 2.3034, 3.6497, 4.6068], device='cuda:3'), covar=tensor([0.0490, 0.3472, 0.0299, 0.5514, 0.0482, 0.3223, 0.1137, 0.0298], device='cuda:3'), in_proj_covar=tensor([0.0255, 0.0206, 0.0213, 0.0277, 0.0232, 0.0206, 0.0207, 0.0217], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 07:47:41,818 INFO [train.py:873] (3/4) Epoch 15, batch 4400, loss[loss=0.1068, simple_loss=0.1513, pruned_loss=0.03113, over 14281.00 frames. ], tot_loss[loss=0.1137, simple_loss=0.1479, pruned_loss=0.03968, over 1987691.86 frames. ], batch size: 44, lr: 5.24e-03, grad_scale: 8.0 2022-12-08 07:47:47,365 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8368, 0.8057, 0.6362, 0.8626, 0.8584, 0.2778, 0.7523, 0.8155], device='cuda:3'), covar=tensor([0.0396, 0.0372, 0.0381, 0.0416, 0.0321, 0.0329, 0.0891, 0.0669], device='cuda:3'), in_proj_covar=tensor([0.0033, 0.0032, 0.0036, 0.0030, 0.0032, 0.0046, 0.0033, 0.0036], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 07:47:54,422 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110283.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:47:54,563 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110283.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:48:00,237 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.123e+01 2.204e+02 2.570e+02 3.174e+02 5.425e+02, threshold=5.140e+02, percent-clipped=0.0 2022-12-08 07:48:09,772 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5648, 4.2616, 4.1122, 4.6019, 4.2542, 3.9965, 4.5878, 3.9480], device='cuda:3'), covar=tensor([0.0421, 0.0923, 0.0433, 0.0397, 0.0832, 0.0824, 0.0542, 0.0502], device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0276, 0.0200, 0.0197, 0.0190, 0.0157, 0.0291, 0.0173], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 07:48:12,298 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110303.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:48:22,581 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110315.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:48:25,414 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110318.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:48:48,565 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110344.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:49:05,183 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110363.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:49:06,017 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1205, 1.9500, 2.0835, 2.1366, 2.0678, 2.0884, 2.2017, 1.8805], device='cuda:3'), covar=tensor([0.0995, 0.1445, 0.0830, 0.0868, 0.1006, 0.0787, 0.0959, 0.0768], device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0276, 0.0201, 0.0197, 0.0189, 0.0158, 0.0291, 0.0173], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 07:49:06,087 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110364.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:49:10,333 INFO [train.py:873] (3/4) Epoch 15, batch 4500, loss[loss=0.1025, simple_loss=0.1422, pruned_loss=0.03144, over 14325.00 frames. ], tot_loss[loss=0.1131, simple_loss=0.1472, pruned_loss=0.03948, over 1967931.12 frames. ], batch size: 60, lr: 5.24e-03, grad_scale: 8.0 2022-12-08 07:49:14,193 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110373.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:49:20,082 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110379.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:49:28,837 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 2022-12-08 07:49:29,134 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.277e+02 2.023e+02 2.685e+02 3.179e+02 5.561e+02, threshold=5.371e+02, percent-clipped=3.0 2022-12-08 07:49:30,362 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.66 vs. limit=5.0 2022-12-08 07:49:38,303 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110400.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:49:58,036 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7587, 1.6889, 3.0509, 2.1404, 2.8964, 1.8041, 2.3491, 2.7352], device='cuda:3'), covar=tensor([0.1158, 0.4214, 0.0715, 0.4107, 0.0897, 0.3393, 0.1388, 0.0922], device='cuda:3'), in_proj_covar=tensor([0.0253, 0.0207, 0.0213, 0.0276, 0.0232, 0.0207, 0.0207, 0.0218], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0005], device='cuda:3') 2022-12-08 07:50:36,574 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110466.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:50:39,117 INFO [train.py:873] (3/4) Epoch 15, batch 4600, loss[loss=0.08579, simple_loss=0.1353, pruned_loss=0.01813, over 14052.00 frames. ], tot_loss[loss=0.1131, simple_loss=0.1473, pruned_loss=0.03939, over 1962818.96 frames. ], batch size: 29, lr: 5.24e-03, grad_scale: 4.0 2022-12-08 07:50:40,981 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110471.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:50:48,755 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.80 vs. limit=5.0 2022-12-08 07:50:58,404 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.443e+02 2.355e+02 2.918e+02 3.425e+02 9.705e+02, threshold=5.836e+02, percent-clipped=5.0 2022-12-08 07:51:23,044 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 2022-12-08 07:51:49,487 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4044, 1.0941, 1.2357, 0.8180, 1.1085, 1.3701, 1.0543, 1.1028], device='cuda:3'), covar=tensor([0.0409, 0.0859, 0.0641, 0.0459, 0.0946, 0.0651, 0.0560, 0.1159], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0171, 0.0136, 0.0125, 0.0139, 0.0151, 0.0129, 0.0139], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 07:52:06,648 INFO [train.py:873] (3/4) Epoch 15, batch 4700, loss[loss=0.1076, simple_loss=0.1534, pruned_loss=0.03085, over 14242.00 frames. ], tot_loss[loss=0.1137, simple_loss=0.1478, pruned_loss=0.03979, over 1912200.85 frames. ], batch size: 35, lr: 5.23e-03, grad_scale: 4.0 2022-12-08 07:52:26,207 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.145e+02 2.034e+02 2.470e+02 3.158e+02 1.519e+03, threshold=4.939e+02, percent-clipped=4.0 2022-12-08 07:52:44,583 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110611.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:52:52,176 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2022-12-08 07:52:57,417 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2022-12-08 07:53:09,054 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110639.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:53:15,753 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0566, 2.1896, 1.9884, 2.1726, 1.8479, 1.9853, 2.1239, 2.0684], device='cuda:3'), covar=tensor([0.0947, 0.1043, 0.1014, 0.0842, 0.1261, 0.0944, 0.0980, 0.1004], device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0140, 0.0144, 0.0157, 0.0145, 0.0122, 0.0167, 0.0147], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 07:53:26,651 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110659.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:53:30,582 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2022-12-08 07:53:34,855 INFO [train.py:873] (3/4) Epoch 15, batch 4800, loss[loss=0.1415, simple_loss=0.1727, pruned_loss=0.05514, over 10377.00 frames. ], tot_loss[loss=0.1132, simple_loss=0.1474, pruned_loss=0.03951, over 1909845.16 frames. ], batch size: 100, lr: 5.23e-03, grad_scale: 8.0 2022-12-08 07:53:37,787 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110672.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:53:38,528 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110673.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:53:39,394 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110674.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:53:53,745 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.088e+02 2.309e+02 2.792e+02 3.592e+02 7.510e+02, threshold=5.583e+02, percent-clipped=5.0 2022-12-08 07:54:01,990 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110700.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:54:20,445 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110721.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:54:24,698 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8594, 1.8593, 1.9777, 1.7373, 1.7392, 1.5981, 1.3742, 1.3244], device='cuda:3'), covar=tensor([0.0178, 0.0365, 0.0208, 0.0263, 0.0240, 0.0300, 0.0261, 0.0437], device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0020, 0.0018, 0.0019, 0.0019, 0.0031, 0.0025, 0.0030], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 07:54:33,589 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110736.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:54:44,446 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110748.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:55:00,691 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110766.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:55:03,181 INFO [train.py:873] (3/4) Epoch 15, batch 4900, loss[loss=0.1127, simple_loss=0.155, pruned_loss=0.0352, over 14376.00 frames. ], tot_loss[loss=0.1133, simple_loss=0.1473, pruned_loss=0.0396, over 1927831.49 frames. ], batch size: 55, lr: 5.23e-03, grad_scale: 8.0 2022-12-08 07:55:04,850 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110771.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:55:22,714 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.185e+02 1.962e+02 2.570e+02 3.049e+02 5.982e+02, threshold=5.139e+02, percent-clipped=2.0 2022-12-08 07:55:28,487 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110797.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 07:55:35,051 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2022-12-08 07:55:43,507 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110814.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:55:48,358 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110819.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:56:31,172 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110868.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:56:31,861 INFO [train.py:873] (3/4) Epoch 15, batch 5000, loss[loss=0.1389, simple_loss=0.1404, pruned_loss=0.06866, over 2674.00 frames. ], tot_loss[loss=0.113, simple_loss=0.147, pruned_loss=0.03949, over 1874444.66 frames. ], batch size: 100, lr: 5.23e-03, grad_scale: 4.0 2022-12-08 07:56:52,356 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.709e+01 2.105e+02 2.624e+02 3.225e+02 7.187e+02, threshold=5.248e+02, percent-clipped=2.0 2022-12-08 07:56:53,182 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2022-12-08 07:57:03,444 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7628, 2.0151, 2.1178, 2.1371, 1.9971, 2.1093, 1.8576, 1.4235], device='cuda:3'), covar=tensor([0.0908, 0.1037, 0.0508, 0.0627, 0.0964, 0.0906, 0.1376, 0.1783], device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0088, 0.0069, 0.0071, 0.0097, 0.0086, 0.0100, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 07:57:25,027 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110929.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:57:33,745 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110939.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:57:50,978 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110958.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:57:51,646 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110959.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:57:58,604 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110967.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:58:00,304 INFO [train.py:873] (3/4) Epoch 15, batch 5100, loss[loss=0.1227, simple_loss=0.1555, pruned_loss=0.04495, over 14278.00 frames. ], tot_loss[loss=0.1131, simple_loss=0.1465, pruned_loss=0.03982, over 1896897.90 frames. ], batch size: 31, lr: 5.22e-03, grad_scale: 4.0 2022-12-08 07:58:04,930 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110974.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:58:16,279 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110987.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:58:16,302 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8351, 4.6531, 4.4386, 4.8815, 4.4065, 4.3332, 4.8843, 4.6760], device='cuda:3'), covar=tensor([0.0574, 0.0808, 0.0800, 0.0490, 0.0718, 0.0525, 0.0506, 0.0640], device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0141, 0.0145, 0.0158, 0.0146, 0.0122, 0.0168, 0.0147], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 07:58:20,532 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.198e+02 2.177e+02 2.703e+02 3.385e+02 6.811e+02, threshold=5.407e+02, percent-clipped=6.0 2022-12-08 07:58:33,414 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=111007.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:58:37,696 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4547, 1.3939, 4.3043, 1.9544, 4.2925, 4.5017, 3.9003, 4.8796], device='cuda:3'), covar=tensor([0.0220, 0.3403, 0.0350, 0.2190, 0.0334, 0.0348, 0.0448, 0.0146], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0157, 0.0161, 0.0170, 0.0169, 0.0180, 0.0133, 0.0151], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 07:58:44,153 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111019.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:58:46,935 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=111022.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 07:59:28,411 INFO [train.py:873] (3/4) Epoch 15, batch 5200, loss[loss=0.1462, simple_loss=0.139, pruned_loss=0.07671, over 1217.00 frames. ], tot_loss[loss=0.1132, simple_loss=0.1465, pruned_loss=0.03997, over 1846220.69 frames. ], batch size: 100, lr: 5.22e-03, grad_scale: 8.0 2022-12-08 07:59:36,278 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2274, 1.8767, 2.1940, 1.5015, 1.9058, 2.2627, 2.1108, 1.9223], device='cuda:3'), covar=tensor([0.0801, 0.0642, 0.0934, 0.1366, 0.1351, 0.0802, 0.0806, 0.1324], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0173, 0.0139, 0.0126, 0.0141, 0.0152, 0.0131, 0.0140], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 07:59:43,330 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9017, 1.9904, 1.9222, 1.9875, 2.0963, 1.7972, 1.6616, 1.3406], device='cuda:3'), covar=tensor([0.0245, 0.0400, 0.0323, 0.0341, 0.0207, 0.0328, 0.0283, 0.0562], device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0020, 0.0018, 0.0019, 0.0019, 0.0031, 0.0025, 0.0030], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 07:59:48,545 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.014e+01 2.007e+02 2.591e+02 3.079e+02 5.022e+02, threshold=5.182e+02, percent-clipped=0.0 2022-12-08 07:59:48,664 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111092.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 08:00:00,494 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7527, 2.1384, 2.1395, 2.2642, 2.0136, 2.2129, 1.9107, 1.4178], device='cuda:3'), covar=tensor([0.1189, 0.1068, 0.0697, 0.0531, 0.0962, 0.0741, 0.1448, 0.2088], device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0088, 0.0069, 0.0072, 0.0098, 0.0086, 0.0100, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 08:00:34,043 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.75 vs. limit=2.0 2022-12-08 08:00:34,337 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.6553, 5.1017, 4.9957, 5.5860, 5.2095, 4.6132, 5.5387, 4.7211], device='cuda:3'), covar=tensor([0.0351, 0.0972, 0.0349, 0.0377, 0.0803, 0.0412, 0.0485, 0.0488], device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0270, 0.0197, 0.0194, 0.0185, 0.0155, 0.0285, 0.0169], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 08:00:57,147 INFO [train.py:873] (3/4) Epoch 15, batch 5300, loss[loss=0.155, simple_loss=0.1498, pruned_loss=0.08009, over 1207.00 frames. ], tot_loss[loss=0.1126, simple_loss=0.1463, pruned_loss=0.03945, over 1831654.27 frames. ], batch size: 100, lr: 5.22e-03, grad_scale: 4.0 2022-12-08 08:01:18,476 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.451e+02 2.026e+02 2.377e+02 2.924e+02 1.186e+03, threshold=4.754e+02, percent-clipped=4.0 2022-12-08 08:01:46,182 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111224.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:01:47,012 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111225.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:01:51,569 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9409, 1.4820, 3.0027, 2.7202, 2.8841, 3.0114, 2.0958, 3.0019], device='cuda:3'), covar=tensor([0.1190, 0.1401, 0.0145, 0.0360, 0.0323, 0.0177, 0.0516, 0.0195], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0159, 0.0130, 0.0170, 0.0147, 0.0141, 0.0122, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 08:02:20,545 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7875, 3.2968, 2.5827, 4.0227, 3.8640, 3.8819, 3.3600, 2.7176], device='cuda:3'), covar=tensor([0.0840, 0.1460, 0.3340, 0.0548, 0.0852, 0.1098, 0.1159, 0.3259], device='cuda:3'), in_proj_covar=tensor([0.0282, 0.0293, 0.0262, 0.0280, 0.0323, 0.0304, 0.0257, 0.0245], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 08:02:23,765 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.64 vs. limit=2.0 2022-12-08 08:02:23,984 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111267.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:02:25,564 INFO [train.py:873] (3/4) Epoch 15, batch 5400, loss[loss=0.1268, simple_loss=0.1317, pruned_loss=0.06096, over 2666.00 frames. ], tot_loss[loss=0.1125, simple_loss=0.1463, pruned_loss=0.03939, over 1824188.60 frames. ], batch size: 100, lr: 5.22e-03, grad_scale: 4.0 2022-12-08 08:02:40,980 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111286.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:02:46,809 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.248e+02 2.071e+02 2.546e+02 3.124e+02 5.888e+02, threshold=5.092e+02, percent-clipped=1.0 2022-12-08 08:03:05,559 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111314.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:03:06,443 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=111315.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:03:53,190 INFO [train.py:873] (3/4) Epoch 15, batch 5500, loss[loss=0.1341, simple_loss=0.1223, pruned_loss=0.07295, over 1289.00 frames. ], tot_loss[loss=0.1127, simple_loss=0.1462, pruned_loss=0.03964, over 1794471.42 frames. ], batch size: 100, lr: 5.21e-03, grad_scale: 4.0 2022-12-08 08:04:13,959 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111392.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 08:04:14,624 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.096e+02 2.065e+02 2.464e+02 3.206e+02 5.854e+02, threshold=4.928e+02, percent-clipped=3.0 2022-12-08 08:04:17,579 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3660, 1.9600, 2.3721, 2.4912, 2.2664, 1.9445, 2.4749, 2.0836], device='cuda:3'), covar=tensor([0.0413, 0.0887, 0.0524, 0.0417, 0.0560, 0.1180, 0.0457, 0.0701], device='cuda:3'), in_proj_covar=tensor([0.0292, 0.0257, 0.0373, 0.0327, 0.0271, 0.0305, 0.0311, 0.0282], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 08:04:32,042 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.64 vs. limit=2.0 2022-12-08 08:04:43,531 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.9588, 0.9462, 0.9219, 0.9593, 0.8576, 0.7983, 0.9668, 0.8363], device='cuda:3'), covar=tensor([0.0227, 0.0227, 0.0202, 0.0246, 0.0228, 0.0376, 0.0228, 0.0355], device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0021, 0.0018, 0.0019, 0.0019, 0.0031, 0.0025, 0.0030], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 08:04:44,607 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.21 vs. limit=5.0 2022-12-08 08:04:56,370 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=111440.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:05:03,490 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.2994, 2.6245, 4.3349, 4.5656, 4.3418, 2.5559, 4.4517, 3.4955], device='cuda:3'), covar=tensor([0.0437, 0.1045, 0.0876, 0.0386, 0.0410, 0.1750, 0.0462, 0.0914], device='cuda:3'), in_proj_covar=tensor([0.0291, 0.0257, 0.0371, 0.0327, 0.0270, 0.0304, 0.0310, 0.0281], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 08:05:20,920 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111468.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:05:21,570 INFO [train.py:873] (3/4) Epoch 15, batch 5600, loss[loss=0.1014, simple_loss=0.1426, pruned_loss=0.03013, over 14335.00 frames. ], tot_loss[loss=0.1128, simple_loss=0.1466, pruned_loss=0.03952, over 1863557.68 frames. ], batch size: 73, lr: 5.21e-03, grad_scale: 8.0 2022-12-08 08:05:22,884 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 2022-12-08 08:05:42,278 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.367e+02 2.013e+02 2.578e+02 3.220e+02 8.105e+02, threshold=5.156e+02, percent-clipped=4.0 2022-12-08 08:05:57,176 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.69 vs. limit=2.0 2022-12-08 08:06:10,303 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111524.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:06:14,528 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111529.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:06:24,560 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0930, 1.7151, 1.8344, 1.8252, 1.9236, 1.1888, 1.6729, 1.7108], device='cuda:3'), covar=tensor([0.0502, 0.1049, 0.0938, 0.1434, 0.0895, 0.0929, 0.0743, 0.0952], device='cuda:3'), in_proj_covar=tensor([0.0033, 0.0033, 0.0036, 0.0031, 0.0032, 0.0046, 0.0034, 0.0037], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 08:06:48,894 INFO [train.py:873] (3/4) Epoch 15, batch 5700, loss[loss=0.1291, simple_loss=0.1399, pruned_loss=0.05917, over 2665.00 frames. ], tot_loss[loss=0.1145, simple_loss=0.148, pruned_loss=0.04051, over 1919925.08 frames. ], batch size: 100, lr: 5.21e-03, grad_scale: 4.0 2022-12-08 08:06:51,466 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=111572.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:06:59,415 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111581.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:07:10,365 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.509e+01 2.357e+02 2.792e+02 3.438e+02 6.748e+02, threshold=5.583e+02, percent-clipped=1.0 2022-12-08 08:07:28,701 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111614.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:08:10,714 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=111662.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:08:17,296 INFO [train.py:873] (3/4) Epoch 15, batch 5800, loss[loss=0.08956, simple_loss=0.1328, pruned_loss=0.02319, over 13987.00 frames. ], tot_loss[loss=0.1141, simple_loss=0.1477, pruned_loss=0.04029, over 1956594.24 frames. ], batch size: 19, lr: 5.21e-03, grad_scale: 4.0 2022-12-08 08:08:38,926 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.252e+02 2.211e+02 2.861e+02 3.266e+02 8.310e+02, threshold=5.722e+02, percent-clipped=4.0 2022-12-08 08:09:16,662 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3418, 3.0311, 3.8179, 2.7387, 2.2898, 3.2527, 1.7900, 3.0283], device='cuda:3'), covar=tensor([0.0718, 0.0856, 0.0406, 0.2153, 0.1995, 0.0746, 0.3172, 0.1135], device='cuda:3'), in_proj_covar=tensor([0.0085, 0.0102, 0.0094, 0.0100, 0.0116, 0.0090, 0.0120, 0.0093], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 08:09:44,964 INFO [train.py:873] (3/4) Epoch 15, batch 5900, loss[loss=0.1143, simple_loss=0.1451, pruned_loss=0.04175, over 11968.00 frames. ], tot_loss[loss=0.1129, simple_loss=0.1469, pruned_loss=0.03946, over 1971161.51 frames. ], batch size: 100, lr: 5.21e-03, grad_scale: 4.0 2022-12-08 08:09:46,035 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7193, 2.2272, 2.2277, 2.2499, 2.0699, 1.5615, 2.6203, 2.4199], device='cuda:3'), covar=tensor([0.0645, 0.0529, 0.0743, 0.1566, 0.1927, 0.0715, 0.0427, 0.0974], device='cuda:3'), in_proj_covar=tensor([0.0033, 0.0033, 0.0036, 0.0031, 0.0033, 0.0046, 0.0034, 0.0037], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 08:09:47,217 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2022-12-08 08:10:04,868 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111791.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 08:10:06,463 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8088, 0.8134, 0.7061, 0.8541, 0.8544, 0.4113, 0.7990, 0.8090], device='cuda:3'), covar=tensor([0.0471, 0.0604, 0.0737, 0.0493, 0.0464, 0.0349, 0.1269, 0.0904], device='cuda:3'), in_proj_covar=tensor([0.0034, 0.0033, 0.0036, 0.0031, 0.0033, 0.0046, 0.0034, 0.0037], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 08:10:07,183 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.002e+02 2.082e+02 2.433e+02 3.039e+02 5.689e+02, threshold=4.867e+02, percent-clipped=0.0 2022-12-08 08:10:33,487 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111824.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:10:57,485 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111852.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 08:11:12,398 INFO [train.py:873] (3/4) Epoch 15, batch 6000, loss[loss=0.1461, simple_loss=0.1663, pruned_loss=0.06292, over 8651.00 frames. ], tot_loss[loss=0.1137, simple_loss=0.1476, pruned_loss=0.03992, over 1939886.68 frames. ], batch size: 100, lr: 5.20e-03, grad_scale: 8.0 2022-12-08 08:11:12,398 INFO [train.py:896] (3/4) Computing validation loss 2022-12-08 08:11:20,778 INFO [train.py:905] (3/4) Epoch 15, validation: loss=0.1363, simple_loss=0.1737, pruned_loss=0.04946, over 857387.00 frames. 2022-12-08 08:11:20,778 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-08 08:11:31,662 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111881.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:11:42,936 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.209e+02 2.150e+02 2.606e+02 3.231e+02 6.699e+02, threshold=5.212e+02, percent-clipped=6.0 2022-12-08 08:12:13,664 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=111929.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:12:48,569 INFO [train.py:873] (3/4) Epoch 15, batch 6100, loss[loss=0.1643, simple_loss=0.1804, pruned_loss=0.07411, over 9533.00 frames. ], tot_loss[loss=0.1133, simple_loss=0.1476, pruned_loss=0.03952, over 1954281.52 frames. ], batch size: 100, lr: 5.20e-03, grad_scale: 4.0 2022-12-08 08:13:07,811 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1519, 2.1418, 3.0430, 2.4000, 3.0112, 3.0225, 2.8794, 2.6135], device='cuda:3'), covar=tensor([0.0997, 0.2602, 0.1186, 0.1668, 0.0732, 0.0885, 0.0996, 0.1687], device='cuda:3'), in_proj_covar=tensor([0.0353, 0.0313, 0.0394, 0.0298, 0.0371, 0.0325, 0.0361, 0.0303], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 08:13:10,956 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.382e+02 2.212e+02 2.710e+02 3.481e+02 4.963e+02, threshold=5.420e+02, percent-clipped=0.0 2022-12-08 08:13:30,829 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4534, 4.5762, 4.8767, 4.0870, 4.6810, 4.9371, 1.9086, 4.3592], device='cuda:3'), covar=tensor([0.0301, 0.0271, 0.0339, 0.0434, 0.0291, 0.0182, 0.3132, 0.0270], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0173, 0.0143, 0.0144, 0.0204, 0.0139, 0.0158, 0.0193], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 08:13:34,876 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.66 vs. limit=2.0 2022-12-08 08:13:40,035 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.0604, 4.6399, 4.4868, 4.9790, 4.6427, 4.3980, 4.9994, 4.2176], device='cuda:3'), covar=tensor([0.0340, 0.0927, 0.0380, 0.0446, 0.0802, 0.0613, 0.0527, 0.0518], device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0270, 0.0195, 0.0193, 0.0186, 0.0156, 0.0283, 0.0169], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 08:13:43,480 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8217, 1.3105, 1.9499, 1.2641, 1.8747, 1.9971, 1.6217, 2.0278], device='cuda:3'), covar=tensor([0.0445, 0.2951, 0.0709, 0.2308, 0.0864, 0.0874, 0.1505, 0.0696], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0157, 0.0160, 0.0170, 0.0169, 0.0180, 0.0133, 0.0152], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 08:14:11,257 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5574, 1.4828, 2.7406, 1.4692, 2.6998, 2.6828, 2.0281, 2.8240], device='cuda:3'), covar=tensor([0.0311, 0.2545, 0.0417, 0.1877, 0.0469, 0.0516, 0.1145, 0.0331], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0156, 0.0159, 0.0169, 0.0168, 0.0180, 0.0132, 0.0151], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 08:14:13,907 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112066.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:14:16,350 INFO [train.py:873] (3/4) Epoch 15, batch 6200, loss[loss=0.1203, simple_loss=0.1519, pruned_loss=0.04439, over 14458.00 frames. ], tot_loss[loss=0.1128, simple_loss=0.1472, pruned_loss=0.0392, over 1941325.74 frames. ], batch size: 51, lr: 5.20e-03, grad_scale: 4.0 2022-12-08 08:14:39,270 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.058e+02 2.103e+02 2.700e+02 3.313e+02 6.071e+02, threshold=5.399e+02, percent-clipped=3.0 2022-12-08 08:15:05,136 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=112124.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:15:07,619 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112127.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:15:25,370 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112147.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 08:15:44,578 INFO [train.py:873] (3/4) Epoch 15, batch 6300, loss[loss=0.1061, simple_loss=0.1467, pruned_loss=0.03273, over 14269.00 frames. ], tot_loss[loss=0.1131, simple_loss=0.1473, pruned_loss=0.03945, over 1960194.26 frames. ], batch size: 76, lr: 5.20e-03, grad_scale: 4.0 2022-12-08 08:15:45,687 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9300, 1.7723, 3.9606, 3.6886, 3.7555, 4.0306, 3.3746, 3.9825], device='cuda:3'), covar=tensor([0.1543, 0.1531, 0.0119, 0.0236, 0.0219, 0.0135, 0.0269, 0.0139], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0158, 0.0130, 0.0169, 0.0145, 0.0141, 0.0123, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 08:15:47,433 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=112172.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:16:07,629 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.242e+02 2.167e+02 2.529e+02 3.043e+02 7.690e+02, threshold=5.057e+02, percent-clipped=2.0 2022-12-08 08:16:16,055 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3422, 4.4707, 4.7250, 3.9994, 4.4644, 4.6948, 1.8604, 4.1382], device='cuda:3'), covar=tensor([0.0283, 0.0293, 0.0316, 0.0437, 0.0314, 0.0272, 0.2910, 0.0313], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0174, 0.0144, 0.0146, 0.0206, 0.0141, 0.0160, 0.0195], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 08:16:34,151 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.30 vs. limit=2.0 2022-12-08 08:17:13,388 INFO [train.py:873] (3/4) Epoch 15, batch 6400, loss[loss=0.1023, simple_loss=0.1484, pruned_loss=0.02806, over 13923.00 frames. ], tot_loss[loss=0.1131, simple_loss=0.1472, pruned_loss=0.03947, over 1962584.56 frames. ], batch size: 26, lr: 5.19e-03, grad_scale: 8.0 2022-12-08 08:17:21,555 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8332, 4.1709, 3.3786, 5.1055, 4.4775, 4.8327, 4.3027, 3.6854], device='cuda:3'), covar=tensor([0.0491, 0.0954, 0.2623, 0.0302, 0.0832, 0.1229, 0.0868, 0.2141], device='cuda:3'), in_proj_covar=tensor([0.0277, 0.0290, 0.0258, 0.0279, 0.0321, 0.0299, 0.0252, 0.0246], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 08:17:35,982 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.228e+02 2.167e+02 2.735e+02 3.656e+02 1.132e+03, threshold=5.470e+02, percent-clipped=11.0 2022-12-08 08:17:39,588 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2848, 3.0479, 2.9741, 3.2930, 2.9991, 3.2237, 3.3441, 2.7264], device='cuda:3'), covar=tensor([0.0578, 0.1153, 0.0598, 0.0521, 0.1087, 0.0495, 0.0580, 0.0617], device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0273, 0.0197, 0.0193, 0.0187, 0.0157, 0.0283, 0.0170], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 08:17:55,704 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2022-12-08 08:18:41,069 INFO [train.py:873] (3/4) Epoch 15, batch 6500, loss[loss=0.1252, simple_loss=0.1419, pruned_loss=0.0542, over 3865.00 frames. ], tot_loss[loss=0.1153, simple_loss=0.1488, pruned_loss=0.04091, over 1888626.43 frames. ], batch size: 100, lr: 5.19e-03, grad_scale: 8.0 2022-12-08 08:18:59,414 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112389.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:19:04,119 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.333e+02 2.163e+02 2.578e+02 3.269e+02 5.158e+02, threshold=5.155e+02, percent-clipped=0.0 2022-12-08 08:19:22,653 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2022-12-08 08:19:28,065 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112422.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:19:28,122 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9165, 1.8372, 4.0999, 3.8556, 3.8449, 4.2600, 3.6391, 4.1859], device='cuda:3'), covar=tensor([0.1557, 0.1424, 0.0135, 0.0238, 0.0258, 0.0128, 0.0261, 0.0139], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0157, 0.0130, 0.0169, 0.0146, 0.0141, 0.0123, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 08:19:50,471 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=112447.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 08:19:52,687 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.67 vs. limit=2.0 2022-12-08 08:19:52,995 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112450.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:20:09,727 INFO [train.py:873] (3/4) Epoch 15, batch 6600, loss[loss=0.136, simple_loss=0.1356, pruned_loss=0.06823, over 1278.00 frames. ], tot_loss[loss=0.1138, simple_loss=0.1478, pruned_loss=0.03997, over 1932640.36 frames. ], batch size: 100, lr: 5.19e-03, grad_scale: 8.0 2022-12-08 08:20:31,715 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2022-12-08 08:20:32,877 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.275e+02 2.043e+02 2.558e+02 3.259e+02 6.600e+02, threshold=5.115e+02, percent-clipped=3.0 2022-12-08 08:20:32,968 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=112495.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 08:21:16,913 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3417, 2.0053, 2.3414, 2.4535, 2.2402, 1.8979, 2.4535, 2.1326], device='cuda:3'), covar=tensor([0.0436, 0.0917, 0.0480, 0.0404, 0.0614, 0.1139, 0.0494, 0.0640], device='cuda:3'), in_proj_covar=tensor([0.0292, 0.0259, 0.0374, 0.0330, 0.0271, 0.0305, 0.0312, 0.0281], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 08:21:32,153 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112562.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:21:35,341 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.2028, 1.3986, 4.0312, 1.7760, 3.9475, 4.1968, 3.3776, 4.4650], device='cuda:3'), covar=tensor([0.0236, 0.3417, 0.0477, 0.2378, 0.0479, 0.0487, 0.0574, 0.0223], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0156, 0.0160, 0.0169, 0.0168, 0.0179, 0.0132, 0.0151], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 08:21:37,750 INFO [train.py:873] (3/4) Epoch 15, batch 6700, loss[loss=0.118, simple_loss=0.1258, pruned_loss=0.05508, over 2617.00 frames. ], tot_loss[loss=0.1134, simple_loss=0.1472, pruned_loss=0.0398, over 1922013.12 frames. ], batch size: 100, lr: 5.19e-03, grad_scale: 8.0 2022-12-08 08:21:44,887 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112577.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:22:01,002 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.453e+02 2.231e+02 2.690e+02 3.411e+02 7.258e+02, threshold=5.380e+02, percent-clipped=8.0 2022-12-08 08:22:06,162 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8063, 1.1454, 1.2724, 1.2214, 0.9768, 1.2841, 1.1243, 0.8921], device='cuda:3'), covar=tensor([0.1628, 0.1098, 0.0448, 0.0484, 0.1987, 0.0911, 0.1536, 0.1637], device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0088, 0.0068, 0.0072, 0.0098, 0.0087, 0.0101, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 08:22:20,643 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112617.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:22:26,056 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112623.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:22:39,582 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112638.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 08:23:06,603 INFO [train.py:873] (3/4) Epoch 15, batch 6800, loss[loss=0.1296, simple_loss=0.1592, pruned_loss=0.04998, over 14593.00 frames. ], tot_loss[loss=0.1135, simple_loss=0.1474, pruned_loss=0.03973, over 1977991.30 frames. ], batch size: 23, lr: 5.18e-03, grad_scale: 8.0 2022-12-08 08:23:14,262 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112678.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 08:23:21,891 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5734, 3.6044, 3.7620, 3.3605, 3.6740, 3.6555, 1.3730, 3.4656], device='cuda:3'), covar=tensor([0.0406, 0.0423, 0.0499, 0.0626, 0.0472, 0.0546, 0.3482, 0.0415], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0171, 0.0142, 0.0145, 0.0204, 0.0139, 0.0158, 0.0192], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 08:23:28,859 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.402e+02 2.249e+02 2.719e+02 3.532e+02 9.106e+02, threshold=5.438e+02, percent-clipped=9.0 2022-12-08 08:23:30,458 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2022-12-08 08:23:31,885 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.84 vs. limit=5.0 2022-12-08 08:23:52,948 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=112722.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:24:08,913 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9961, 1.8489, 4.2118, 3.9773, 3.8891, 4.2859, 3.7476, 4.2409], device='cuda:3'), covar=tensor([0.1562, 0.1521, 0.0130, 0.0246, 0.0260, 0.0149, 0.0305, 0.0148], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0157, 0.0130, 0.0169, 0.0146, 0.0141, 0.0123, 0.0125], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 08:24:13,177 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112745.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:24:16,872 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7487, 2.0007, 2.2187, 2.1531, 1.9488, 2.1701, 1.9214, 1.3847], device='cuda:3'), covar=tensor([0.0797, 0.1084, 0.0474, 0.0649, 0.1006, 0.0726, 0.1119, 0.2229], device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0089, 0.0069, 0.0072, 0.0098, 0.0087, 0.0102, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 08:24:34,754 INFO [train.py:873] (3/4) Epoch 15, batch 6900, loss[loss=0.1345, simple_loss=0.1305, pruned_loss=0.06924, over 1160.00 frames. ], tot_loss[loss=0.1131, simple_loss=0.1475, pruned_loss=0.03933, over 2012454.65 frames. ], batch size: 100, lr: 5.18e-03, grad_scale: 8.0 2022-12-08 08:24:35,722 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=112770.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:24:37,136 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.70 vs. limit=5.0 2022-12-08 08:24:53,491 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2022-12-08 08:24:57,285 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.416e+02 2.169e+02 2.514e+02 3.179e+02 8.231e+02, threshold=5.028e+02, percent-clipped=2.0 2022-12-08 08:25:36,753 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.80 vs. limit=2.0 2022-12-08 08:25:46,768 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.0120, 4.7966, 4.6594, 5.0481, 4.6052, 4.3485, 5.1275, 4.8203], device='cuda:3'), covar=tensor([0.0644, 0.0832, 0.0743, 0.0605, 0.0795, 0.0559, 0.0576, 0.0784], device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0141, 0.0145, 0.0160, 0.0146, 0.0122, 0.0168, 0.0147], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 08:26:02,336 INFO [train.py:873] (3/4) Epoch 15, batch 7000, loss[loss=0.1249, simple_loss=0.164, pruned_loss=0.04289, over 14197.00 frames. ], tot_loss[loss=0.1119, simple_loss=0.1465, pruned_loss=0.03868, over 1980573.55 frames. ], batch size: 80, lr: 5.18e-03, grad_scale: 8.0 2022-12-08 08:26:25,455 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.347e+02 2.195e+02 2.715e+02 3.350e+02 5.262e+02, threshold=5.431e+02, percent-clipped=3.0 2022-12-08 08:26:39,116 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3513, 2.3366, 4.3658, 3.1073, 4.1609, 2.2250, 3.3998, 4.1612], device='cuda:3'), covar=tensor([0.0531, 0.3447, 0.0379, 0.5472, 0.0567, 0.3016, 0.1183, 0.0519], device='cuda:3'), in_proj_covar=tensor([0.0252, 0.0201, 0.0213, 0.0273, 0.0231, 0.0207, 0.0205, 0.0215], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 08:26:46,187 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112918.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:26:49,191 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112921.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:26:59,956 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112933.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 08:27:31,514 INFO [train.py:873] (3/4) Epoch 15, batch 7100, loss[loss=0.07927, simple_loss=0.1216, pruned_loss=0.01848, over 13484.00 frames. ], tot_loss[loss=0.1125, simple_loss=0.1464, pruned_loss=0.03929, over 1954151.51 frames. ], batch size: 17, lr: 5.18e-03, grad_scale: 8.0 2022-12-08 08:27:35,233 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112973.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 08:27:41,350 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0182, 2.0117, 2.2966, 2.3392, 2.1334, 1.8377, 1.7212, 1.4468], device='cuda:3'), covar=tensor([0.0350, 0.0559, 0.0344, 0.0249, 0.0236, 0.0445, 0.0349, 0.0585], device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0021, 0.0018, 0.0020, 0.0019, 0.0032, 0.0026, 0.0031], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 08:27:43,063 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112982.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:27:54,134 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.485e+02 2.181e+02 2.817e+02 3.355e+02 7.638e+02, threshold=5.634e+02, percent-clipped=2.0 2022-12-08 08:28:08,011 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=7.36 vs. limit=5.0 2022-12-08 08:28:38,856 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113045.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:28:59,912 INFO [train.py:873] (3/4) Epoch 15, batch 7200, loss[loss=0.103, simple_loss=0.1428, pruned_loss=0.03156, over 13871.00 frames. ], tot_loss[loss=0.1137, simple_loss=0.1475, pruned_loss=0.03996, over 1992020.11 frames. ], batch size: 23, lr: 5.18e-03, grad_scale: 8.0 2022-12-08 08:29:21,099 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=113093.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:29:23,117 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.106e+02 2.228e+02 2.733e+02 3.389e+02 6.488e+02, threshold=5.466e+02, percent-clipped=1.0 2022-12-08 08:29:24,145 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.3986, 1.0554, 1.2350, 0.8774, 1.1456, 1.4148, 1.0320, 1.1519], device='cuda:3'), covar=tensor([0.0439, 0.0820, 0.0728, 0.0514, 0.0890, 0.0714, 0.0612, 0.1099], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0175, 0.0140, 0.0127, 0.0143, 0.0155, 0.0132, 0.0142], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0007, 0.0006, 0.0006], device='cuda:3') 2022-12-08 08:29:37,636 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3497, 2.1855, 2.6225, 1.6559, 1.7950, 2.3566, 1.4694, 2.3052], device='cuda:3'), covar=tensor([0.0989, 0.1669, 0.0667, 0.2231, 0.2263, 0.0936, 0.3213, 0.1126], device='cuda:3'), in_proj_covar=tensor([0.0085, 0.0101, 0.0093, 0.0099, 0.0116, 0.0089, 0.0120, 0.0092], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 08:30:06,800 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2022-12-08 08:30:28,629 INFO [train.py:873] (3/4) Epoch 15, batch 7300, loss[loss=0.1059, simple_loss=0.1361, pruned_loss=0.03788, over 4969.00 frames. ], tot_loss[loss=0.113, simple_loss=0.1467, pruned_loss=0.03963, over 1998003.59 frames. ], batch size: 100, lr: 5.17e-03, grad_scale: 4.0 2022-12-08 08:30:52,685 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.169e+02 2.057e+02 2.452e+02 3.236e+02 1.011e+03, threshold=4.903e+02, percent-clipped=3.0 2022-12-08 08:31:12,607 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113218.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:31:25,888 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113233.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:31:26,694 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5531, 4.3387, 4.2082, 4.4872, 4.1704, 3.9362, 4.6267, 4.4278], device='cuda:3'), covar=tensor([0.0606, 0.0838, 0.0774, 0.0717, 0.0796, 0.0634, 0.0542, 0.0729], device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0140, 0.0145, 0.0159, 0.0145, 0.0122, 0.0166, 0.0147], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 08:31:26,812 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113234.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:31:31,620 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2257, 3.0939, 2.4704, 3.3404, 3.2426, 3.3067, 2.9519, 2.4859], device='cuda:3'), covar=tensor([0.1025, 0.1451, 0.2959, 0.0750, 0.0916, 0.1050, 0.1317, 0.2913], device='cuda:3'), in_proj_covar=tensor([0.0283, 0.0292, 0.0260, 0.0282, 0.0323, 0.0302, 0.0254, 0.0247], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 08:31:55,299 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=113266.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:31:57,793 INFO [train.py:873] (3/4) Epoch 15, batch 7400, loss[loss=0.09547, simple_loss=0.1389, pruned_loss=0.02605, over 14224.00 frames. ], tot_loss[loss=0.1128, simple_loss=0.1466, pruned_loss=0.03946, over 1985381.59 frames. ], batch size: 35, lr: 5.17e-03, grad_scale: 4.0 2022-12-08 08:32:01,640 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113273.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 08:32:05,483 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=113277.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:32:08,855 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=113281.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:32:21,744 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113295.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:32:22,353 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.108e+02 1.996e+02 2.516e+02 3.094e+02 5.431e+02, threshold=5.033e+02, percent-clipped=1.0 2022-12-08 08:32:44,285 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=113321.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:33:21,026 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2022-12-08 08:33:25,702 INFO [train.py:873] (3/4) Epoch 15, batch 7500, loss[loss=0.116, simple_loss=0.1529, pruned_loss=0.03952, over 14265.00 frames. ], tot_loss[loss=0.1131, simple_loss=0.1471, pruned_loss=0.03956, over 1993238.55 frames. ], batch size: 80, lr: 5.17e-03, grad_scale: 4.0 2022-12-08 08:33:27,409 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4887, 4.1853, 4.0353, 4.5308, 4.1776, 3.9858, 4.5550, 3.8863], device='cuda:3'), covar=tensor([0.0401, 0.1053, 0.0417, 0.0398, 0.0846, 0.0892, 0.0504, 0.0467], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0267, 0.0194, 0.0190, 0.0182, 0.0153, 0.0277, 0.0166], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 08:33:47,113 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8455, 0.8517, 0.6708, 0.8225, 0.8670, 0.3254, 0.8208, 0.8682], device='cuda:3'), covar=tensor([0.0424, 0.0477, 0.0413, 0.0448, 0.0321, 0.0267, 0.1074, 0.0631], device='cuda:3'), in_proj_covar=tensor([0.0033, 0.0033, 0.0036, 0.0031, 0.0032, 0.0046, 0.0034, 0.0036], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 08:33:48,620 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.106e+02 2.151e+02 2.525e+02 3.725e+02 6.084e+02, threshold=5.049e+02, percent-clipped=3.0 2022-12-08 08:33:52,609 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7496, 2.3508, 3.5858, 3.7683, 3.6067, 2.4243, 3.6682, 2.7923], device='cuda:3'), covar=tensor([0.0398, 0.1061, 0.0780, 0.0473, 0.0483, 0.1469, 0.0460, 0.0938], device='cuda:3'), in_proj_covar=tensor([0.0291, 0.0259, 0.0375, 0.0328, 0.0271, 0.0306, 0.0311, 0.0281], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 08:33:57,555 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0370, 2.9547, 2.3293, 3.1371, 2.9909, 3.0566, 2.7333, 2.3518], device='cuda:3'), covar=tensor([0.0877, 0.1206, 0.2700, 0.0678, 0.1126, 0.0850, 0.1304, 0.2489], device='cuda:3'), in_proj_covar=tensor([0.0283, 0.0293, 0.0261, 0.0282, 0.0324, 0.0301, 0.0254, 0.0248], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 08:33:59,862 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0383, 3.8192, 3.7383, 4.0990, 3.6757, 3.4858, 4.1039, 3.9357], device='cuda:3'), covar=tensor([0.0675, 0.1029, 0.0839, 0.0675, 0.0968, 0.0733, 0.0547, 0.0772], device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0139, 0.0144, 0.0158, 0.0145, 0.0122, 0.0165, 0.0146], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 08:34:54,702 INFO [train.py:873] (3/4) Epoch 16, batch 0, loss[loss=0.1098, simple_loss=0.1537, pruned_loss=0.03292, over 13916.00 frames. ], tot_loss[loss=0.1098, simple_loss=0.1537, pruned_loss=0.03292, over 13916.00 frames. ], batch size: 20, lr: 5.00e-03, grad_scale: 8.0 2022-12-08 08:34:54,702 INFO [train.py:896] (3/4) Computing validation loss 2022-12-08 08:35:01,997 INFO [train.py:905] (3/4) Epoch 16, validation: loss=0.1445, simple_loss=0.1858, pruned_loss=0.05158, over 857387.00 frames. 2022-12-08 08:35:01,998 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-08 08:35:15,983 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 2022-12-08 08:35:21,000 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9969, 1.8089, 3.1176, 2.2680, 3.0090, 1.8542, 2.4255, 3.0027], device='cuda:3'), covar=tensor([0.1015, 0.3856, 0.0692, 0.4172, 0.0854, 0.3037, 0.1320, 0.0781], device='cuda:3'), in_proj_covar=tensor([0.0251, 0.0203, 0.0214, 0.0275, 0.0231, 0.0206, 0.0206, 0.0215], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 08:35:22,928 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8668, 3.5278, 3.2864, 2.5695, 3.2256, 3.6036, 3.7926, 3.3376], device='cuda:3'), covar=tensor([0.0461, 0.1124, 0.0856, 0.1346, 0.1033, 0.0559, 0.0909, 0.0877], device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0172, 0.0137, 0.0125, 0.0140, 0.0152, 0.0129, 0.0139], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 08:35:28,116 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8166, 2.5739, 3.3322, 2.4030, 2.1587, 2.8710, 1.6754, 2.8818], device='cuda:3'), covar=tensor([0.0882, 0.1084, 0.0553, 0.1928, 0.1951, 0.0968, 0.3133, 0.0944], device='cuda:3'), in_proj_covar=tensor([0.0086, 0.0102, 0.0093, 0.0100, 0.0116, 0.0089, 0.0121, 0.0093], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 08:35:53,141 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6169, 3.4249, 3.1226, 3.3243, 3.5300, 3.5322, 3.5665, 3.6050], device='cuda:3'), covar=tensor([0.0882, 0.0587, 0.2183, 0.2341, 0.0751, 0.0866, 0.0976, 0.0730], device='cuda:3'), in_proj_covar=tensor([0.0383, 0.0266, 0.0442, 0.0564, 0.0342, 0.0439, 0.0387, 0.0385], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 08:36:01,324 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.044e+01 2.013e+02 2.778e+02 4.051e+02 1.103e+03, threshold=5.556e+02, percent-clipped=15.0 2022-12-08 08:36:31,856 INFO [train.py:873] (3/4) Epoch 16, batch 100, loss[loss=0.1065, simple_loss=0.1449, pruned_loss=0.0341, over 14236.00 frames. ], tot_loss[loss=0.1116, simple_loss=0.1473, pruned_loss=0.03795, over 941144.01 frames. ], batch size: 76, lr: 5.00e-03, grad_scale: 4.0 2022-12-08 08:36:46,607 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1906, 1.5014, 1.7553, 1.6835, 1.6352, 1.6812, 1.3451, 1.2433], device='cuda:3'), covar=tensor([0.1383, 0.1234, 0.0478, 0.0518, 0.1164, 0.1146, 0.1488, 0.1664], device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0087, 0.0068, 0.0072, 0.0098, 0.0086, 0.0100, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 08:37:12,407 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5627, 4.3310, 3.9392, 4.2058, 4.3455, 4.4996, 4.5460, 4.5583], device='cuda:3'), covar=tensor([0.0766, 0.0444, 0.2083, 0.2570, 0.0724, 0.0795, 0.0682, 0.0736], device='cuda:3'), in_proj_covar=tensor([0.0382, 0.0265, 0.0440, 0.0562, 0.0339, 0.0437, 0.0385, 0.0385], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 08:37:13,464 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113577.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:37:25,099 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=113590.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:37:31,129 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.249e+02 2.137e+02 2.757e+02 3.201e+02 6.700e+02, threshold=5.514e+02, percent-clipped=1.0 2022-12-08 08:37:49,439 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113617.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:37:56,679 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=113625.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:38:01,674 INFO [train.py:873] (3/4) Epoch 16, batch 200, loss[loss=0.1285, simple_loss=0.1435, pruned_loss=0.05677, over 6027.00 frames. ], tot_loss[loss=0.1109, simple_loss=0.1458, pruned_loss=0.03803, over 1320771.16 frames. ], batch size: 100, lr: 5.00e-03, grad_scale: 4.0 2022-12-08 08:38:14,836 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113646.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:38:19,017 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.10 vs. limit=2.0 2022-12-08 08:38:30,895 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9047, 1.8630, 4.6533, 4.2683, 4.0475, 4.7877, 4.2812, 4.7465], device='cuda:3'), covar=tensor([0.1516, 0.1468, 0.0099, 0.0193, 0.0224, 0.0095, 0.0152, 0.0115], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0159, 0.0131, 0.0170, 0.0148, 0.0143, 0.0125, 0.0126], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 08:38:43,042 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113678.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:38:52,826 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4906, 3.9054, 3.6823, 3.4159, 2.9326, 3.7098, 3.4986, 2.1098], device='cuda:3'), covar=tensor([0.1404, 0.0432, 0.0498, 0.0952, 0.0802, 0.0542, 0.1055, 0.1856], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0087, 0.0068, 0.0071, 0.0097, 0.0086, 0.0099, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 08:38:59,610 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.233e+02 2.172e+02 2.621e+02 3.364e+02 2.345e+03, threshold=5.243e+02, percent-clipped=6.0 2022-12-08 08:39:09,093 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113707.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:39:17,681 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1834, 2.7878, 2.7215, 1.7210, 2.6551, 3.0120, 3.2253, 2.5000], device='cuda:3'), covar=tensor([0.0646, 0.0931, 0.0959, 0.1722, 0.0990, 0.0580, 0.0692, 0.1255], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0174, 0.0139, 0.0127, 0.0143, 0.0154, 0.0131, 0.0142], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 08:39:21,857 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6562, 3.6907, 4.2828, 3.3090, 2.5743, 3.6304, 2.0553, 3.7037], device='cuda:3'), covar=tensor([0.0561, 0.0675, 0.0417, 0.1501, 0.1877, 0.0696, 0.3050, 0.0915], device='cuda:3'), in_proj_covar=tensor([0.0085, 0.0100, 0.0092, 0.0098, 0.0115, 0.0088, 0.0119, 0.0091], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 08:39:30,011 INFO [train.py:873] (3/4) Epoch 16, batch 300, loss[loss=0.1019, simple_loss=0.1398, pruned_loss=0.03201, over 11979.00 frames. ], tot_loss[loss=0.1122, simple_loss=0.1465, pruned_loss=0.03896, over 1570902.32 frames. ], batch size: 100, lr: 5.00e-03, grad_scale: 4.0 2022-12-08 08:40:28,277 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.134e+02 1.970e+02 2.457e+02 3.061e+02 5.224e+02, threshold=4.913e+02, percent-clipped=0.0 2022-12-08 08:40:57,903 INFO [train.py:873] (3/4) Epoch 16, batch 400, loss[loss=0.09172, simple_loss=0.1423, pruned_loss=0.0206, over 14250.00 frames. ], tot_loss[loss=0.112, simple_loss=0.1465, pruned_loss=0.0388, over 1701695.19 frames. ], batch size: 37, lr: 4.99e-03, grad_scale: 4.0 2022-12-08 08:41:29,641 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.54 vs. limit=5.0 2022-12-08 08:41:50,240 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113890.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:41:56,841 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.079e+02 2.042e+02 2.566e+02 3.444e+02 7.891e+02, threshold=5.133e+02, percent-clipped=8.0 2022-12-08 08:42:25,887 INFO [train.py:873] (3/4) Epoch 16, batch 500, loss[loss=0.1163, simple_loss=0.1568, pruned_loss=0.03789, over 14088.00 frames. ], tot_loss[loss=0.1134, simple_loss=0.1473, pruned_loss=0.03975, over 1789047.07 frames. ], batch size: 29, lr: 4.99e-03, grad_scale: 4.0 2022-12-08 08:42:32,156 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=113938.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:43:02,309 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=113973.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:43:15,255 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.11 vs. limit=2.0 2022-12-08 08:43:22,719 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.5503, 1.9875, 2.0311, 2.1828, 1.8175, 2.0948, 1.9236, 1.1316], device='cuda:3'), covar=tensor([0.1230, 0.1054, 0.0919, 0.0645, 0.1330, 0.0766, 0.1575, 0.2504], device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0088, 0.0069, 0.0072, 0.0099, 0.0087, 0.0101, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 08:43:24,305 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.003e+02 2.124e+02 2.547e+02 2.966e+02 5.657e+02, threshold=5.095e+02, percent-clipped=2.0 2022-12-08 08:43:28,108 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=114002.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:43:40,770 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.6181, 2.5404, 2.0217, 2.6509, 2.4783, 2.5155, 2.3188, 2.1536], device='cuda:3'), covar=tensor([0.0977, 0.1299, 0.3291, 0.0840, 0.1469, 0.1139, 0.1593, 0.2421], device='cuda:3'), in_proj_covar=tensor([0.0282, 0.0292, 0.0261, 0.0281, 0.0321, 0.0299, 0.0253, 0.0245], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 08:43:45,526 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.10 vs. limit=2.0 2022-12-08 08:43:53,991 INFO [train.py:873] (3/4) Epoch 16, batch 600, loss[loss=0.1246, simple_loss=0.1196, pruned_loss=0.0648, over 1240.00 frames. ], tot_loss[loss=0.1125, simple_loss=0.1463, pruned_loss=0.03934, over 1801818.25 frames. ], batch size: 100, lr: 4.99e-03, grad_scale: 4.0 2022-12-08 08:44:19,345 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9811, 1.9431, 2.0525, 2.0355, 1.8283, 1.5805, 1.5590, 1.1518], device='cuda:3'), covar=tensor([0.0234, 0.0318, 0.0223, 0.0222, 0.0278, 0.0307, 0.0249, 0.0486], device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0021, 0.0018, 0.0020, 0.0020, 0.0032, 0.0026, 0.0031], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 08:44:28,228 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1009, 2.4516, 3.7895, 2.8759, 3.9405, 3.6566, 3.6637, 3.2551], device='cuda:3'), covar=tensor([0.0687, 0.3228, 0.1309, 0.1871, 0.0764, 0.1183, 0.1481, 0.1764], device='cuda:3'), in_proj_covar=tensor([0.0353, 0.0310, 0.0390, 0.0296, 0.0367, 0.0320, 0.0360, 0.0300], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 08:44:29,368 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.75 vs. limit=2.0 2022-12-08 08:44:48,769 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.26 vs. limit=5.0 2022-12-08 08:44:52,421 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.322e+02 2.140e+02 2.657e+02 3.260e+02 5.874e+02, threshold=5.314e+02, percent-clipped=5.0 2022-12-08 08:45:18,387 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=114127.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:45:21,946 INFO [train.py:873] (3/4) Epoch 16, batch 700, loss[loss=0.1233, simple_loss=0.138, pruned_loss=0.05431, over 3871.00 frames. ], tot_loss[loss=0.1119, simple_loss=0.1461, pruned_loss=0.03888, over 1827775.15 frames. ], batch size: 100, lr: 4.99e-03, grad_scale: 4.0 2022-12-08 08:46:12,416 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=114188.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:46:20,992 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.316e+02 2.016e+02 2.560e+02 3.238e+02 1.602e+03, threshold=5.120e+02, percent-clipped=5.0 2022-12-08 08:46:50,896 INFO [train.py:873] (3/4) Epoch 16, batch 800, loss[loss=0.09583, simple_loss=0.1382, pruned_loss=0.02672, over 13918.00 frames. ], tot_loss[loss=0.1117, simple_loss=0.1458, pruned_loss=0.0388, over 1858823.06 frames. ], batch size: 26, lr: 4.98e-03, grad_scale: 8.0 2022-12-08 08:46:52,390 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.18 vs. limit=5.0 2022-12-08 08:47:28,633 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=114273.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:47:48,678 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.47 vs. limit=5.0 2022-12-08 08:47:50,274 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.518e+01 2.109e+02 2.499e+02 3.135e+02 7.663e+02, threshold=4.999e+02, percent-clipped=3.0 2022-12-08 08:47:54,018 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=114302.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:48:09,884 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=114321.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:48:18,624 INFO [train.py:873] (3/4) Epoch 16, batch 900, loss[loss=0.0974, simple_loss=0.1369, pruned_loss=0.02892, over 14302.00 frames. ], tot_loss[loss=0.1115, simple_loss=0.1458, pruned_loss=0.03864, over 1885319.24 frames. ], batch size: 60, lr: 4.98e-03, grad_scale: 4.0 2022-12-08 08:48:35,137 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=114350.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:48:39,074 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2022-12-08 08:48:40,916 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4116, 2.0948, 2.5950, 1.7397, 1.7549, 2.2411, 1.2722, 2.4269], device='cuda:3'), covar=tensor([0.0920, 0.2087, 0.0900, 0.2709, 0.2802, 0.1350, 0.4271, 0.0888], device='cuda:3'), in_proj_covar=tensor([0.0086, 0.0101, 0.0093, 0.0099, 0.0116, 0.0089, 0.0120, 0.0092], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 08:48:50,113 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.67 vs. limit=2.0 2022-12-08 08:49:15,178 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2022-12-08 08:49:17,095 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.318e+02 2.233e+02 2.852e+02 3.607e+02 7.620e+02, threshold=5.705e+02, percent-clipped=7.0 2022-12-08 08:49:20,765 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3850, 3.5651, 3.7219, 3.3319, 3.5694, 3.5213, 1.4810, 3.3416], device='cuda:3'), covar=tensor([0.0420, 0.0368, 0.0368, 0.0542, 0.0359, 0.0547, 0.3159, 0.0337], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0174, 0.0146, 0.0147, 0.0207, 0.0142, 0.0159, 0.0194], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 08:49:34,904 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8026, 4.4604, 4.3106, 4.7771, 4.4942, 4.1936, 4.8104, 4.0525], device='cuda:3'), covar=tensor([0.0325, 0.0841, 0.0423, 0.0410, 0.0711, 0.0625, 0.0480, 0.0489], device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0269, 0.0196, 0.0191, 0.0184, 0.0156, 0.0282, 0.0169], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 08:49:45,662 INFO [train.py:873] (3/4) Epoch 16, batch 1000, loss[loss=0.1186, simple_loss=0.148, pruned_loss=0.04457, over 14539.00 frames. ], tot_loss[loss=0.1128, simple_loss=0.1467, pruned_loss=0.03943, over 1931371.84 frames. ], batch size: 24, lr: 4.98e-03, grad_scale: 4.0 2022-12-08 08:50:30,809 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=114483.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:50:45,034 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.412e+02 2.045e+02 2.472e+02 3.169e+02 1.892e+03, threshold=4.944e+02, percent-clipped=6.0 2022-12-08 08:51:13,049 INFO [train.py:873] (3/4) Epoch 16, batch 1100, loss[loss=0.1695, simple_loss=0.1521, pruned_loss=0.09345, over 1258.00 frames. ], tot_loss[loss=0.113, simple_loss=0.1468, pruned_loss=0.03964, over 1893739.20 frames. ], batch size: 100, lr: 4.98e-03, grad_scale: 4.0 2022-12-08 08:51:32,135 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1649, 2.0556, 2.0970, 2.1985, 2.0783, 2.0805, 2.2773, 1.8960], device='cuda:3'), covar=tensor([0.0952, 0.1217, 0.0745, 0.0810, 0.1069, 0.0731, 0.0833, 0.0763], device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0268, 0.0197, 0.0192, 0.0184, 0.0157, 0.0283, 0.0169], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 08:51:35,739 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9072, 1.5575, 1.8851, 1.9808, 1.5026, 1.7484, 1.8089, 1.8383], device='cuda:3'), covar=tensor([0.0206, 0.0350, 0.0207, 0.0174, 0.0354, 0.0397, 0.0268, 0.0190], device='cuda:3'), in_proj_covar=tensor([0.0293, 0.0261, 0.0378, 0.0332, 0.0275, 0.0309, 0.0313, 0.0283], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 08:52:12,293 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.217e+02 2.285e+02 2.811e+02 4.159e+02 1.083e+03, threshold=5.623e+02, percent-clipped=11.0 2022-12-08 08:52:22,610 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2022-12-08 08:52:41,093 INFO [train.py:873] (3/4) Epoch 16, batch 1200, loss[loss=0.1834, simple_loss=0.1692, pruned_loss=0.09881, over 1279.00 frames. ], tot_loss[loss=0.113, simple_loss=0.1472, pruned_loss=0.03947, over 1943921.61 frames. ], batch size: 100, lr: 4.98e-03, grad_scale: 8.0 2022-12-08 08:53:41,425 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.092e+02 1.943e+02 2.456e+02 3.315e+02 7.474e+02, threshold=4.912e+02, percent-clipped=2.0 2022-12-08 08:53:56,735 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=114716.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:54:01,012 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7930, 3.9045, 4.1615, 3.6797, 4.0459, 4.1629, 1.6085, 3.7912], device='cuda:3'), covar=tensor([0.0342, 0.0351, 0.0311, 0.0502, 0.0277, 0.0309, 0.2962, 0.0294], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0172, 0.0143, 0.0144, 0.0203, 0.0140, 0.0156, 0.0192], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 08:54:09,741 INFO [train.py:873] (3/4) Epoch 16, batch 1300, loss[loss=0.1135, simple_loss=0.1498, pruned_loss=0.03859, over 14173.00 frames. ], tot_loss[loss=0.1123, simple_loss=0.1468, pruned_loss=0.03892, over 2024492.77 frames. ], batch size: 84, lr: 4.97e-03, grad_scale: 8.0 2022-12-08 08:54:15,011 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0440, 2.1487, 2.2443, 2.3445, 2.0035, 2.1791, 2.0997, 1.3874], device='cuda:3'), covar=tensor([0.1026, 0.0927, 0.0620, 0.0510, 0.1015, 0.0745, 0.1117, 0.1862], device='cuda:3'), in_proj_covar=tensor([0.0140, 0.0088, 0.0069, 0.0072, 0.0098, 0.0087, 0.0101, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 08:54:50,233 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=114777.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:54:56,059 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=114783.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:55:04,290 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.09 vs. limit=2.0 2022-12-08 08:55:09,769 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.131e+02 2.098e+02 2.728e+02 3.262e+02 8.407e+02, threshold=5.455e+02, percent-clipped=4.0 2022-12-08 08:55:38,357 INFO [train.py:873] (3/4) Epoch 16, batch 1400, loss[loss=0.1115, simple_loss=0.1481, pruned_loss=0.03747, over 14164.00 frames. ], tot_loss[loss=0.1129, simple_loss=0.1472, pruned_loss=0.03933, over 1988941.64 frames. ], batch size: 84, lr: 4.97e-03, grad_scale: 8.0 2022-12-08 08:55:38,728 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=114831.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:56:14,467 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=114871.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:56:31,878 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8897, 1.3948, 3.1065, 2.8027, 2.9800, 3.1101, 2.3680, 3.0752], device='cuda:3'), covar=tensor([0.1421, 0.1650, 0.0187, 0.0419, 0.0378, 0.0233, 0.0491, 0.0214], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0159, 0.0131, 0.0170, 0.0147, 0.0142, 0.0125, 0.0125], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 08:56:38,331 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.172e+02 2.101e+02 2.672e+02 3.477e+02 6.209e+02, threshold=5.345e+02, percent-clipped=3.0 2022-12-08 08:57:06,856 INFO [train.py:873] (3/4) Epoch 16, batch 1500, loss[loss=0.1154, simple_loss=0.1474, pruned_loss=0.04172, over 5897.00 frames. ], tot_loss[loss=0.1119, simple_loss=0.1462, pruned_loss=0.03879, over 1955398.03 frames. ], batch size: 100, lr: 4.97e-03, grad_scale: 8.0 2022-12-08 08:57:07,903 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=114932.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:57:41,485 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5365, 4.6493, 4.8992, 4.2628, 4.7088, 4.9693, 1.6770, 4.4272], device='cuda:3'), covar=tensor([0.0307, 0.0296, 0.0327, 0.0419, 0.0349, 0.0152, 0.3145, 0.0304], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0172, 0.0143, 0.0144, 0.0203, 0.0140, 0.0156, 0.0191], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 08:58:07,016 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.294e+02 2.137e+02 2.682e+02 3.588e+02 7.078e+02, threshold=5.364e+02, percent-clipped=5.0 2022-12-08 08:58:39,010 INFO [train.py:873] (3/4) Epoch 16, batch 1600, loss[loss=0.1239, simple_loss=0.1525, pruned_loss=0.04765, over 14163.00 frames. ], tot_loss[loss=0.1116, simple_loss=0.1457, pruned_loss=0.03879, over 1961903.25 frames. ], batch size: 84, lr: 4.97e-03, grad_scale: 8.0 2022-12-08 08:58:39,134 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9535, 4.7303, 4.2435, 4.5123, 4.6416, 4.7788, 4.8206, 4.8846], device='cuda:3'), covar=tensor([0.0743, 0.0448, 0.2213, 0.2741, 0.0748, 0.0947, 0.0954, 0.0878], device='cuda:3'), in_proj_covar=tensor([0.0390, 0.0267, 0.0447, 0.0570, 0.0347, 0.0444, 0.0393, 0.0392], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 08:59:15,598 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115072.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:59:27,763 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=115086.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 08:59:39,794 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.499e+02 2.275e+02 2.932e+02 3.523e+02 7.695e+02, threshold=5.865e+02, percent-clipped=6.0 2022-12-08 09:00:07,460 INFO [train.py:873] (3/4) Epoch 16, batch 1700, loss[loss=0.1063, simple_loss=0.1424, pruned_loss=0.03516, over 14257.00 frames. ], tot_loss[loss=0.1117, simple_loss=0.1461, pruned_loss=0.03866, over 1985072.81 frames. ], batch size: 80, lr: 4.96e-03, grad_scale: 4.0 2022-12-08 09:00:14,917 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.52 vs. limit=5.0 2022-12-08 09:00:21,856 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=115147.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:00:26,622 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.69 vs. limit=2.0 2022-12-08 09:01:02,112 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0124, 3.8080, 3.7371, 4.0212, 3.6846, 3.4862, 4.0519, 3.8812], device='cuda:3'), covar=tensor([0.0617, 0.0888, 0.0744, 0.0637, 0.0829, 0.0683, 0.0562, 0.0707], device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0141, 0.0144, 0.0159, 0.0146, 0.0123, 0.0166, 0.0146], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 09:01:09,164 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.404e+01 2.088e+02 2.519e+02 3.191e+02 6.447e+02, threshold=5.039e+02, percent-clipped=1.0 2022-12-08 09:01:33,703 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115227.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:01:36,927 INFO [train.py:873] (3/4) Epoch 16, batch 1800, loss[loss=0.1149, simple_loss=0.1497, pruned_loss=0.04006, over 14283.00 frames. ], tot_loss[loss=0.1118, simple_loss=0.1459, pruned_loss=0.03881, over 1959425.37 frames. ], batch size: 57, lr: 4.96e-03, grad_scale: 4.0 2022-12-08 09:02:18,434 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2672, 1.7381, 2.2299, 1.8320, 2.3533, 2.1489, 2.0538, 2.1176], device='cuda:3'), covar=tensor([0.0740, 0.2422, 0.0699, 0.1103, 0.0620, 0.1067, 0.0651, 0.1019], device='cuda:3'), in_proj_covar=tensor([0.0355, 0.0313, 0.0395, 0.0298, 0.0370, 0.0325, 0.0363, 0.0300], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 09:02:38,559 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.091e+02 1.953e+02 2.460e+02 2.998e+02 4.316e+02, threshold=4.920e+02, percent-clipped=0.0 2022-12-08 09:03:06,674 INFO [train.py:873] (3/4) Epoch 16, batch 1900, loss[loss=0.1193, simple_loss=0.1521, pruned_loss=0.04324, over 14277.00 frames. ], tot_loss[loss=0.1113, simple_loss=0.1451, pruned_loss=0.03879, over 1895965.16 frames. ], batch size: 80, lr: 4.96e-03, grad_scale: 4.0 2022-12-08 09:03:07,741 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6195, 1.5083, 1.4377, 1.5992, 1.6198, 1.0987, 1.4769, 1.4332], device='cuda:3'), covar=tensor([0.0711, 0.0825, 0.0627, 0.0574, 0.0641, 0.0876, 0.0664, 0.0820], device='cuda:3'), in_proj_covar=tensor([0.0034, 0.0033, 0.0036, 0.0031, 0.0033, 0.0046, 0.0034, 0.0037], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 09:03:39,704 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3901, 3.2826, 4.1773, 3.0442, 2.4871, 3.4407, 2.2634, 3.5151], device='cuda:3'), covar=tensor([0.1210, 0.1335, 0.0582, 0.1786, 0.1907, 0.1090, 0.2928, 0.1392], device='cuda:3'), in_proj_covar=tensor([0.0085, 0.0100, 0.0092, 0.0099, 0.0114, 0.0087, 0.0118, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 09:03:41,649 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=115370.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:03:42,504 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=115371.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:03:43,329 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=115372.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:04:08,304 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.218e+02 2.103e+02 2.568e+02 3.222e+02 1.079e+03, threshold=5.135e+02, percent-clipped=6.0 2022-12-08 09:04:24,853 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4800, 4.5896, 4.8823, 4.0945, 4.7759, 5.0100, 1.8217, 4.4623], device='cuda:3'), covar=tensor([0.0310, 0.0311, 0.0359, 0.0547, 0.0318, 0.0162, 0.3139, 0.0267], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0173, 0.0145, 0.0145, 0.0204, 0.0142, 0.0157, 0.0192], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 09:04:26,541 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=115420.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:04:36,094 INFO [train.py:873] (3/4) Epoch 16, batch 2000, loss[loss=0.1155, simple_loss=0.1488, pruned_loss=0.04115, over 10318.00 frames. ], tot_loss[loss=0.1117, simple_loss=0.1454, pruned_loss=0.03901, over 1871173.97 frames. ], batch size: 100, lr: 4.96e-03, grad_scale: 8.0 2022-12-08 09:04:36,281 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=115431.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:04:37,419 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=115432.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:04:45,839 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115442.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:05:36,028 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.145e+01 1.975e+02 2.632e+02 3.322e+02 6.213e+02, threshold=5.264e+02, percent-clipped=1.0 2022-12-08 09:05:59,709 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=115527.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:06:03,273 INFO [train.py:873] (3/4) Epoch 16, batch 2100, loss[loss=0.1121, simple_loss=0.1492, pruned_loss=0.03752, over 14205.00 frames. ], tot_loss[loss=0.1116, simple_loss=0.1456, pruned_loss=0.03883, over 1893354.24 frames. ], batch size: 94, lr: 4.96e-03, grad_scale: 4.0 2022-12-08 09:06:41,693 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.57 vs. limit=2.0 2022-12-08 09:06:42,177 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=115575.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:06:47,398 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2022-12-08 09:06:49,050 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0841, 4.1713, 4.4127, 3.7336, 4.2489, 4.3568, 1.6727, 3.9766], device='cuda:3'), covar=tensor([0.0329, 0.0391, 0.0376, 0.0634, 0.0325, 0.0239, 0.3215, 0.0308], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0175, 0.0146, 0.0147, 0.0207, 0.0143, 0.0158, 0.0194], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 09:07:05,104 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.092e+02 1.971e+02 2.573e+02 2.986e+02 1.089e+03, threshold=5.147e+02, percent-clipped=6.0 2022-12-08 09:07:11,119 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7123, 1.8239, 1.9968, 1.9886, 1.7419, 1.6592, 1.4063, 1.0666], device='cuda:3'), covar=tensor([0.0279, 0.0304, 0.0249, 0.0287, 0.0389, 0.0304, 0.0288, 0.0494], device='cuda:3'), in_proj_covar=tensor([0.0021, 0.0021, 0.0018, 0.0020, 0.0020, 0.0032, 0.0026, 0.0031], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 09:07:12,996 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.5444, 1.5191, 1.5770, 1.4411, 1.3405, 1.3879, 1.1873, 0.9902], device='cuda:3'), covar=tensor([0.0200, 0.0234, 0.0201, 0.0273, 0.0265, 0.0332, 0.0264, 0.0431], device='cuda:3'), in_proj_covar=tensor([0.0021, 0.0021, 0.0018, 0.0020, 0.0020, 0.0032, 0.0026, 0.0031], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 09:07:31,195 INFO [train.py:873] (3/4) Epoch 16, batch 2200, loss[loss=0.1171, simple_loss=0.1413, pruned_loss=0.0464, over 4963.00 frames. ], tot_loss[loss=0.1122, simple_loss=0.1461, pruned_loss=0.03916, over 1865539.02 frames. ], batch size: 100, lr: 4.95e-03, grad_scale: 4.0 2022-12-08 09:08:32,753 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.142e+02 2.098e+02 2.631e+02 3.239e+02 6.147e+02, threshold=5.263e+02, percent-clipped=2.0 2022-12-08 09:08:39,686 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 2022-12-08 09:08:54,837 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115726.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:08:55,706 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115727.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:08:58,974 INFO [train.py:873] (3/4) Epoch 16, batch 2300, loss[loss=0.1541, simple_loss=0.1711, pruned_loss=0.06856, over 8633.00 frames. ], tot_loss[loss=0.1119, simple_loss=0.1464, pruned_loss=0.03873, over 1929721.07 frames. ], batch size: 100, lr: 4.95e-03, grad_scale: 4.0 2022-12-08 09:09:08,627 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=115742.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:09:26,364 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7005, 2.2321, 2.3153, 2.5805, 2.1218, 1.3804, 2.4187, 2.7650], device='cuda:3'), covar=tensor([0.1350, 0.0803, 0.0890, 0.0964, 0.1462, 0.0794, 0.0816, 0.0658], device='cuda:3'), in_proj_covar=tensor([0.0034, 0.0033, 0.0036, 0.0031, 0.0033, 0.0046, 0.0034, 0.0037], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 09:09:50,759 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=115790.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:10:00,478 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.978e+01 2.138e+02 2.481e+02 3.114e+02 7.710e+02, threshold=4.963e+02, percent-clipped=3.0 2022-12-08 09:10:26,760 INFO [train.py:873] (3/4) Epoch 16, batch 2400, loss[loss=0.108, simple_loss=0.1469, pruned_loss=0.03455, over 14361.00 frames. ], tot_loss[loss=0.1128, simple_loss=0.1468, pruned_loss=0.03937, over 1896233.42 frames. ], batch size: 73, lr: 4.95e-03, grad_scale: 8.0 2022-12-08 09:10:52,428 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4511, 1.7855, 4.3283, 1.9477, 4.1832, 4.5495, 3.9898, 4.8696], device='cuda:3'), covar=tensor([0.0200, 0.3073, 0.0390, 0.2351, 0.0392, 0.0276, 0.0379, 0.0142], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0154, 0.0159, 0.0167, 0.0166, 0.0177, 0.0133, 0.0151], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 09:11:19,578 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=115892.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 09:11:27,448 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.676e+01 2.182e+02 2.667e+02 3.375e+02 7.141e+02, threshold=5.334e+02, percent-clipped=4.0 2022-12-08 09:11:37,186 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4098, 3.1613, 2.5077, 3.5161, 3.3803, 3.3937, 2.9720, 2.5000], device='cuda:3'), covar=tensor([0.0994, 0.1371, 0.3147, 0.0686, 0.0995, 0.1213, 0.1480, 0.3146], device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0296, 0.0262, 0.0284, 0.0324, 0.0305, 0.0256, 0.0247], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 09:11:54,003 INFO [train.py:873] (3/4) Epoch 16, batch 2500, loss[loss=0.0972, simple_loss=0.1229, pruned_loss=0.03575, over 3850.00 frames. ], tot_loss[loss=0.1116, simple_loss=0.1458, pruned_loss=0.03866, over 1888855.33 frames. ], batch size: 100, lr: 4.95e-03, grad_scale: 8.0 2022-12-08 09:12:13,587 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=115953.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 09:12:56,497 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.515e+02 2.068e+02 2.436e+02 3.138e+02 6.345e+02, threshold=4.873e+02, percent-clipped=1.0 2022-12-08 09:13:07,096 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116014.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:13:17,568 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116026.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:13:18,371 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116027.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:13:21,683 INFO [train.py:873] (3/4) Epoch 16, batch 2600, loss[loss=0.1043, simple_loss=0.1481, pruned_loss=0.0302, over 14339.00 frames. ], tot_loss[loss=0.112, simple_loss=0.146, pruned_loss=0.03894, over 1872447.39 frames. ], batch size: 31, lr: 4.95e-03, grad_scale: 4.0 2022-12-08 09:13:59,636 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=116074.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:14:00,549 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=116075.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:14:00,631 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116075.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 09:14:17,253 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7036, 2.3217, 2.5439, 1.6435, 2.2666, 2.5038, 2.6493, 2.2677], device='cuda:3'), covar=tensor([0.0734, 0.0928, 0.0845, 0.1556, 0.1236, 0.0859, 0.0679, 0.1360], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0174, 0.0139, 0.0126, 0.0142, 0.0153, 0.0133, 0.0142], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 09:14:23,795 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.166e+02 2.134e+02 2.737e+02 3.520e+02 9.731e+02, threshold=5.473e+02, percent-clipped=5.0 2022-12-08 09:14:26,811 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116105.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:14:36,339 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116116.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:14:49,176 INFO [train.py:873] (3/4) Epoch 16, batch 2700, loss[loss=0.1143, simple_loss=0.1284, pruned_loss=0.05008, over 2620.00 frames. ], tot_loss[loss=0.1129, simple_loss=0.1467, pruned_loss=0.03951, over 1920283.14 frames. ], batch size: 100, lr: 4.94e-03, grad_scale: 4.0 2022-12-08 09:15:20,035 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116166.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:15:29,463 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116177.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:15:51,299 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.057e+02 2.028e+02 2.481e+02 3.066e+02 5.287e+02, threshold=4.961e+02, percent-clipped=0.0 2022-12-08 09:16:16,290 INFO [train.py:873] (3/4) Epoch 16, batch 2800, loss[loss=0.1144, simple_loss=0.1513, pruned_loss=0.03881, over 12739.00 frames. ], tot_loss[loss=0.1121, simple_loss=0.1465, pruned_loss=0.0389, over 1942087.96 frames. ], batch size: 100, lr: 4.94e-03, grad_scale: 8.0 2022-12-08 09:16:22,279 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0184, 3.4180, 3.2642, 3.1979, 2.5306, 3.3820, 3.2577, 1.7019], device='cuda:3'), covar=tensor([0.1478, 0.1047, 0.1080, 0.0718, 0.0986, 0.0587, 0.0847, 0.2166], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0087, 0.0069, 0.0072, 0.0097, 0.0086, 0.0099, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 09:16:31,675 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116248.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 09:16:46,134 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 2022-12-08 09:17:19,503 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.418e+02 2.003e+02 2.440e+02 3.142e+02 5.921e+02, threshold=4.880e+02, percent-clipped=3.0 2022-12-08 09:17:19,783 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2930, 2.1937, 3.2503, 2.4751, 3.1470, 3.1279, 3.0159, 2.7437], device='cuda:3'), covar=tensor([0.1039, 0.3061, 0.0945, 0.1934, 0.0828, 0.1081, 0.1380, 0.1644], device='cuda:3'), in_proj_covar=tensor([0.0357, 0.0315, 0.0396, 0.0302, 0.0372, 0.0326, 0.0366, 0.0303], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 09:17:31,532 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2050, 2.0749, 2.1573, 2.2440, 2.1228, 2.1292, 2.3164, 1.9402], device='cuda:3'), covar=tensor([0.1096, 0.1408, 0.0795, 0.0908, 0.1206, 0.0829, 0.0870, 0.0757], device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0274, 0.0199, 0.0195, 0.0185, 0.0158, 0.0282, 0.0171], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 09:17:37,577 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116323.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:17:44,502 INFO [train.py:873] (3/4) Epoch 16, batch 2900, loss[loss=0.1022, simple_loss=0.145, pruned_loss=0.02976, over 14191.00 frames. ], tot_loss[loss=0.1122, simple_loss=0.1463, pruned_loss=0.03907, over 1938475.77 frames. ], batch size: 84, lr: 4.94e-03, grad_scale: 4.0 2022-12-08 09:18:18,453 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116370.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 09:18:30,463 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116384.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:18:47,320 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.284e+02 2.101e+02 2.585e+02 3.248e+02 5.004e+02, threshold=5.170e+02, percent-clipped=1.0 2022-12-08 09:19:07,261 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 2022-12-08 09:19:11,753 INFO [train.py:873] (3/4) Epoch 16, batch 3000, loss[loss=0.1146, simple_loss=0.1529, pruned_loss=0.03814, over 14269.00 frames. ], tot_loss[loss=0.1118, simple_loss=0.1463, pruned_loss=0.03863, over 1931231.44 frames. ], batch size: 60, lr: 4.94e-03, grad_scale: 4.0 2022-12-08 09:19:11,753 INFO [train.py:896] (3/4) Computing validation loss 2022-12-08 09:19:18,099 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5835, 2.7193, 4.3703, 3.4295, 4.4358, 4.2183, 4.2074, 3.7471], device='cuda:3'), covar=tensor([0.0595, 0.3349, 0.0948, 0.1725, 0.0591, 0.0910, 0.1349, 0.2010], device='cuda:3'), in_proj_covar=tensor([0.0353, 0.0312, 0.0393, 0.0299, 0.0370, 0.0323, 0.0362, 0.0300], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 09:19:20,204 INFO [train.py:905] (3/4) Epoch 16, validation: loss=0.1369, simple_loss=0.1741, pruned_loss=0.04986, over 857387.00 frames. 2022-12-08 09:19:20,204 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-08 09:19:26,731 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2022-12-08 09:19:32,988 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116446.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:19:45,781 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116461.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:19:55,309 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116472.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:19:55,390 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4127, 1.4419, 1.3374, 1.4509, 1.5096, 1.0618, 1.2427, 1.3773], device='cuda:3'), covar=tensor([0.0587, 0.0581, 0.0709, 0.0563, 0.0528, 0.0879, 0.0792, 0.0586], device='cuda:3'), in_proj_covar=tensor([0.0035, 0.0033, 0.0037, 0.0032, 0.0033, 0.0047, 0.0035, 0.0038], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 09:20:22,410 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.224e+02 2.145e+02 2.640e+02 3.351e+02 6.789e+02, threshold=5.280e+02, percent-clipped=3.0 2022-12-08 09:20:24,260 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116505.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:20:26,009 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116507.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 09:20:45,069 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.6085, 5.1413, 4.9471, 5.6001, 5.1193, 4.7109, 5.5207, 4.5361], device='cuda:3'), covar=tensor([0.0299, 0.0933, 0.0353, 0.0347, 0.0843, 0.0364, 0.0490, 0.0524], device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0271, 0.0198, 0.0194, 0.0184, 0.0156, 0.0282, 0.0169], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 09:20:46,690 INFO [train.py:873] (3/4) Epoch 16, batch 3100, loss[loss=0.1474, simple_loss=0.1387, pruned_loss=0.07799, over 1255.00 frames. ], tot_loss[loss=0.1111, simple_loss=0.1456, pruned_loss=0.03827, over 1979723.71 frames. ], batch size: 100, lr: 4.93e-03, grad_scale: 4.0 2022-12-08 09:21:01,856 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116548.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 09:21:02,836 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5796, 1.9417, 2.5130, 2.0666, 2.5301, 2.3929, 2.3523, 2.3098], device='cuda:3'), covar=tensor([0.0695, 0.2326, 0.0754, 0.1394, 0.0628, 0.0997, 0.0708, 0.1159], device='cuda:3'), in_proj_covar=tensor([0.0353, 0.0311, 0.0393, 0.0298, 0.0371, 0.0324, 0.0361, 0.0302], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 09:21:17,444 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116566.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:21:43,617 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=116596.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 09:21:49,521 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.191e+02 2.181e+02 2.695e+02 3.453e+02 7.270e+02, threshold=5.390e+02, percent-clipped=4.0 2022-12-08 09:21:55,918 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.5217, 1.1544, 2.0081, 1.7750, 1.8741, 2.0856, 1.3230, 2.0616], device='cuda:3'), covar=tensor([0.0840, 0.1389, 0.0335, 0.0622, 0.0636, 0.0311, 0.0788, 0.0351], device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0157, 0.0130, 0.0168, 0.0145, 0.0140, 0.0123, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 09:21:56,837 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4006, 1.4841, 1.5656, 1.4167, 1.3823, 1.2467, 1.0951, 1.1339], device='cuda:3'), covar=tensor([0.0245, 0.0208, 0.0159, 0.0204, 0.0219, 0.0415, 0.0313, 0.0486], device='cuda:3'), in_proj_covar=tensor([0.0021, 0.0021, 0.0019, 0.0020, 0.0020, 0.0032, 0.0026, 0.0031], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 09:22:02,710 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116618.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:22:10,397 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.06 vs. limit=5.0 2022-12-08 09:22:13,977 INFO [train.py:873] (3/4) Epoch 16, batch 3200, loss[loss=0.08817, simple_loss=0.1376, pruned_loss=0.01939, over 14228.00 frames. ], tot_loss[loss=0.1107, simple_loss=0.1449, pruned_loss=0.03819, over 1886307.73 frames. ], batch size: 39, lr: 4.93e-03, grad_scale: 8.0 2022-12-08 09:22:35,865 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8637, 4.0375, 4.2624, 3.7138, 4.0923, 4.2111, 1.7058, 3.8635], device='cuda:3'), covar=tensor([0.0397, 0.0355, 0.0331, 0.0451, 0.0332, 0.0318, 0.3005, 0.0282], device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0172, 0.0145, 0.0145, 0.0205, 0.0142, 0.0156, 0.0192], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 09:22:44,615 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4267, 2.5695, 4.3373, 4.5533, 4.2732, 2.5803, 4.4541, 3.3954], device='cuda:3'), covar=tensor([0.0381, 0.1142, 0.0990, 0.0368, 0.0457, 0.1770, 0.0435, 0.0878], device='cuda:3'), in_proj_covar=tensor([0.0293, 0.0260, 0.0377, 0.0331, 0.0274, 0.0307, 0.0311, 0.0282], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 09:22:47,965 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116670.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 09:22:55,610 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116679.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:22:55,711 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116679.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:22:56,555 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8809, 1.5552, 2.0701, 1.6066, 1.8660, 1.4733, 1.6554, 1.9321], device='cuda:3'), covar=tensor([0.2210, 0.2663, 0.0599, 0.1879, 0.1464, 0.1298, 0.1108, 0.1013], device='cuda:3'), in_proj_covar=tensor([0.0250, 0.0201, 0.0213, 0.0271, 0.0232, 0.0203, 0.0203, 0.0216], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 09:23:16,520 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.234e+02 2.227e+02 2.604e+02 3.188e+02 6.019e+02, threshold=5.207e+02, percent-clipped=1.0 2022-12-08 09:23:17,645 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116704.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:23:29,697 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=116718.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:23:37,942 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6118, 1.5805, 1.7942, 1.5168, 1.4840, 1.3803, 1.5035, 1.1819], device='cuda:3'), covar=tensor([0.0179, 0.0167, 0.0137, 0.0143, 0.0173, 0.0294, 0.0206, 0.0323], device='cuda:3'), in_proj_covar=tensor([0.0021, 0.0021, 0.0019, 0.0020, 0.0020, 0.0032, 0.0026, 0.0031], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 09:23:41,060 INFO [train.py:873] (3/4) Epoch 16, batch 3300, loss[loss=0.1057, simple_loss=0.1354, pruned_loss=0.03804, over 4957.00 frames. ], tot_loss[loss=0.1108, simple_loss=0.1446, pruned_loss=0.03847, over 1847050.79 frames. ], batch size: 100, lr: 4.93e-03, grad_scale: 8.0 2022-12-08 09:23:47,936 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1534, 1.1549, 0.9897, 1.1486, 1.1611, 0.8008, 1.0658, 1.1093], device='cuda:3'), covar=tensor([0.0407, 0.0639, 0.0449, 0.0510, 0.0376, 0.0452, 0.0788, 0.0810], device='cuda:3'), in_proj_covar=tensor([0.0034, 0.0033, 0.0036, 0.0031, 0.0033, 0.0046, 0.0035, 0.0037], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 09:23:48,870 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0362, 2.0214, 2.2081, 2.3751, 2.1130, 1.7781, 1.6844, 1.5068], device='cuda:3'), covar=tensor([0.0220, 0.0448, 0.0348, 0.0147, 0.0362, 0.0392, 0.0259, 0.0436], device='cuda:3'), in_proj_covar=tensor([0.0021, 0.0021, 0.0019, 0.0020, 0.0020, 0.0032, 0.0026, 0.0031], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 09:24:07,177 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116761.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:24:10,700 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116765.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:24:16,774 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116772.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:24:28,103 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7827, 1.4113, 2.5219, 2.2610, 2.3891, 2.5591, 1.7877, 2.5437], device='cuda:3'), covar=tensor([0.0920, 0.1291, 0.0242, 0.0468, 0.0477, 0.0239, 0.0671, 0.0246], device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0156, 0.0130, 0.0167, 0.0145, 0.0140, 0.0123, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 09:24:43,050 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116802.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 09:24:43,099 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116802.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:24:43,678 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.185e+02 1.987e+02 2.396e+02 2.955e+02 5.766e+02, threshold=4.792e+02, percent-clipped=2.0 2022-12-08 09:24:48,968 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=116809.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:24:58,346 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=116820.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:25:08,248 INFO [train.py:873] (3/4) Epoch 16, batch 3400, loss[loss=0.176, simple_loss=0.1638, pruned_loss=0.09408, over 1248.00 frames. ], tot_loss[loss=0.1098, simple_loss=0.1445, pruned_loss=0.03757, over 1879135.12 frames. ], batch size: 100, lr: 4.93e-03, grad_scale: 8.0 2022-12-08 09:25:34,067 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116861.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:25:35,844 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116863.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:26:10,487 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.102e+02 2.045e+02 2.562e+02 3.256e+02 5.626e+02, threshold=5.125e+02, percent-clipped=2.0 2022-12-08 09:26:24,388 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2785, 3.0592, 3.0271, 3.2610, 3.1197, 3.2452, 3.3490, 2.7244], device='cuda:3'), covar=tensor([0.0591, 0.0953, 0.0566, 0.0595, 0.0794, 0.0424, 0.0560, 0.0634], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0268, 0.0195, 0.0192, 0.0181, 0.0155, 0.0279, 0.0168], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 09:26:29,294 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.81 vs. limit=2.0 2022-12-08 09:26:35,748 INFO [train.py:873] (3/4) Epoch 16, batch 3500, loss[loss=0.1628, simple_loss=0.1799, pruned_loss=0.0729, over 8609.00 frames. ], tot_loss[loss=0.1111, simple_loss=0.1455, pruned_loss=0.03838, over 1902627.50 frames. ], batch size: 100, lr: 4.93e-03, grad_scale: 8.0 2022-12-08 09:26:44,143 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8594, 2.8233, 2.1389, 2.9147, 2.7565, 2.8229, 2.5447, 2.2787], device='cuda:3'), covar=tensor([0.0984, 0.1493, 0.2970, 0.0900, 0.1244, 0.1129, 0.1441, 0.2575], device='cuda:3'), in_proj_covar=tensor([0.0283, 0.0295, 0.0262, 0.0285, 0.0323, 0.0304, 0.0257, 0.0245], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 09:26:59,108 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.5101, 1.0913, 1.9966, 1.7772, 1.9048, 2.1002, 1.4450, 2.0518], device='cuda:3'), covar=tensor([0.0803, 0.1470, 0.0343, 0.0581, 0.0659, 0.0287, 0.0755, 0.0310], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0159, 0.0131, 0.0170, 0.0147, 0.0141, 0.0125, 0.0125], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 09:27:12,616 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116974.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:27:16,874 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116979.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:27:20,298 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5414, 3.6902, 3.8817, 3.3643, 3.7233, 3.7195, 1.4828, 3.5211], device='cuda:3'), covar=tensor([0.0393, 0.0415, 0.0332, 0.0634, 0.0364, 0.0403, 0.3164, 0.0310], device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0173, 0.0146, 0.0147, 0.0206, 0.0141, 0.0157, 0.0193], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 09:27:38,633 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.091e+02 2.141e+02 2.605e+02 3.184e+02 8.184e+02, threshold=5.210e+02, percent-clipped=2.0 2022-12-08 09:27:58,644 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=117027.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:28:02,011 INFO [train.py:873] (3/4) Epoch 16, batch 3600, loss[loss=0.1109, simple_loss=0.1467, pruned_loss=0.03752, over 14458.00 frames. ], tot_loss[loss=0.1107, simple_loss=0.1457, pruned_loss=0.03784, over 1957361.84 frames. ], batch size: 51, lr: 4.92e-03, grad_scale: 8.0 2022-12-08 09:28:27,279 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117060.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:28:32,878 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.6237, 1.7184, 4.5449, 2.3391, 4.2698, 4.7332, 4.5530, 5.0595], device='cuda:3'), covar=tensor([0.0283, 0.4307, 0.0538, 0.2643, 0.0397, 0.0417, 0.0283, 0.0305], device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0157, 0.0161, 0.0170, 0.0169, 0.0182, 0.0135, 0.0154], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 09:29:03,731 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117102.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:29:06,091 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.352e+02 2.154e+02 2.650e+02 3.296e+02 7.753e+02, threshold=5.300e+02, percent-clipped=2.0 2022-12-08 09:29:28,981 INFO [train.py:873] (3/4) Epoch 16, batch 3700, loss[loss=0.1191, simple_loss=0.1546, pruned_loss=0.0418, over 14085.00 frames. ], tot_loss[loss=0.1109, simple_loss=0.1458, pruned_loss=0.03802, over 2020740.70 frames. ], batch size: 26, lr: 4.92e-03, grad_scale: 4.0 2022-12-08 09:29:45,683 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=117150.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:29:52,522 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117158.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:29:55,075 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117161.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:29:55,901 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9383, 1.7342, 3.0716, 2.2919, 2.9719, 1.7904, 2.4407, 2.9507], device='cuda:3'), covar=tensor([0.1069, 0.4271, 0.0861, 0.4394, 0.1010, 0.3391, 0.1215, 0.0959], device='cuda:3'), in_proj_covar=tensor([0.0252, 0.0202, 0.0213, 0.0272, 0.0232, 0.0205, 0.0202, 0.0217], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0005], device='cuda:3') 2022-12-08 09:30:34,102 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.316e+02 2.079e+02 2.495e+02 3.268e+02 6.334e+02, threshold=4.990e+02, percent-clipped=4.0 2022-12-08 09:30:38,003 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=117209.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:30:48,143 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117221.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:30:57,051 INFO [train.py:873] (3/4) Epoch 16, batch 3800, loss[loss=0.1561, simple_loss=0.1494, pruned_loss=0.08144, over 1225.00 frames. ], tot_loss[loss=0.1119, simple_loss=0.1463, pruned_loss=0.03875, over 2009370.08 frames. ], batch size: 100, lr: 4.92e-03, grad_scale: 4.0 2022-12-08 09:31:13,114 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3001, 2.9415, 3.0262, 2.0921, 2.7677, 2.9078, 3.3061, 2.6735], device='cuda:3'), covar=tensor([0.0673, 0.1049, 0.0809, 0.1350, 0.1034, 0.0753, 0.0641, 0.1173], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0173, 0.0139, 0.0125, 0.0141, 0.0153, 0.0132, 0.0142], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 09:31:18,383 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.61 vs. limit=5.0 2022-12-08 09:31:34,956 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117274.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:31:41,910 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117282.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:32:02,068 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.166e+02 2.002e+02 2.660e+02 3.371e+02 8.415e+02, threshold=5.321e+02, percent-clipped=3.0 2022-12-08 09:32:17,225 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=117322.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:32:23,711 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117329.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:32:25,197 INFO [train.py:873] (3/4) Epoch 16, batch 3900, loss[loss=0.1354, simple_loss=0.1632, pruned_loss=0.05381, over 10334.00 frames. ], tot_loss[loss=0.1112, simple_loss=0.1455, pruned_loss=0.03844, over 2007700.71 frames. ], batch size: 100, lr: 4.92e-03, grad_scale: 4.0 2022-12-08 09:32:51,272 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117360.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:33:17,638 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117390.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:33:30,802 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.106e+02 1.882e+02 2.302e+02 2.791e+02 6.246e+02, threshold=4.605e+02, percent-clipped=1.0 2022-12-08 09:33:33,456 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=117408.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:33:43,140 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8262, 1.6302, 1.8955, 1.6806, 2.0001, 1.8193, 1.5989, 1.8865], device='cuda:3'), covar=tensor([0.0616, 0.1372, 0.0476, 0.0504, 0.0476, 0.0724, 0.0340, 0.0383], device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0312, 0.0390, 0.0297, 0.0367, 0.0322, 0.0362, 0.0299], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 09:33:53,717 INFO [train.py:873] (3/4) Epoch 16, batch 4000, loss[loss=0.1298, simple_loss=0.1302, pruned_loss=0.06469, over 1261.00 frames. ], tot_loss[loss=0.1108, simple_loss=0.1455, pruned_loss=0.038, over 2004035.89 frames. ], batch size: 100, lr: 4.92e-03, grad_scale: 8.0 2022-12-08 09:34:09,675 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4046, 3.1790, 2.4258, 3.5004, 3.3830, 3.4048, 3.0159, 2.5052], device='cuda:3'), covar=tensor([0.0846, 0.1305, 0.3124, 0.0622, 0.1015, 0.1461, 0.1284, 0.2984], device='cuda:3'), in_proj_covar=tensor([0.0287, 0.0296, 0.0265, 0.0288, 0.0327, 0.0305, 0.0259, 0.0247], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 09:34:17,439 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117458.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:34:58,776 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.434e+02 2.165e+02 2.594e+02 3.323e+02 1.177e+03, threshold=5.188e+02, percent-clipped=6.0 2022-12-08 09:35:00,064 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=117506.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:35:22,147 INFO [train.py:873] (3/4) Epoch 16, batch 4100, loss[loss=0.1755, simple_loss=0.1836, pruned_loss=0.08368, over 7770.00 frames. ], tot_loss[loss=0.1111, simple_loss=0.1457, pruned_loss=0.03826, over 1987675.34 frames. ], batch size: 100, lr: 4.91e-03, grad_scale: 8.0 2022-12-08 09:35:34,943 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1267, 1.1099, 1.1856, 1.0123, 1.0701, 0.8151, 0.8270, 0.8370], device='cuda:3'), covar=tensor([0.0176, 0.0171, 0.0146, 0.0166, 0.0175, 0.0333, 0.0230, 0.0294], device='cuda:3'), in_proj_covar=tensor([0.0021, 0.0021, 0.0019, 0.0020, 0.0020, 0.0032, 0.0026, 0.0031], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 09:35:40,892 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.26 vs. limit=5.0 2022-12-08 09:36:02,202 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117577.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:36:27,068 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.327e+02 2.013e+02 2.651e+02 3.289e+02 6.397e+02, threshold=5.302e+02, percent-clipped=7.0 2022-12-08 09:36:37,967 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.12 vs. limit=2.0 2022-12-08 09:36:38,928 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2022-12-08 09:36:49,546 INFO [train.py:873] (3/4) Epoch 16, batch 4200, loss[loss=0.1166, simple_loss=0.1536, pruned_loss=0.0398, over 14266.00 frames. ], tot_loss[loss=0.1125, simple_loss=0.1467, pruned_loss=0.03912, over 1976545.92 frames. ], batch size: 60, lr: 4.91e-03, grad_scale: 8.0 2022-12-08 09:36:50,770 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3917, 2.6669, 4.4350, 4.5159, 4.3426, 2.7359, 4.6268, 3.6935], device='cuda:3'), covar=tensor([0.0383, 0.1168, 0.0739, 0.0387, 0.0435, 0.1673, 0.0283, 0.0807], device='cuda:3'), in_proj_covar=tensor([0.0292, 0.0259, 0.0374, 0.0327, 0.0271, 0.0304, 0.0307, 0.0279], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 09:36:51,533 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2022-12-08 09:37:37,455 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117685.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:37:54,402 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.263e+02 2.184e+02 2.504e+02 3.069e+02 5.464e+02, threshold=5.008e+02, percent-clipped=1.0 2022-12-08 09:37:54,583 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.2847, 4.4417, 4.7032, 3.9735, 4.4692, 4.7288, 1.7666, 4.1980], device='cuda:3'), covar=tensor([0.0376, 0.0392, 0.0344, 0.0501, 0.0325, 0.0306, 0.3286, 0.0300], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0173, 0.0146, 0.0146, 0.0205, 0.0141, 0.0157, 0.0194], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 09:38:17,333 INFO [train.py:873] (3/4) Epoch 16, batch 4300, loss[loss=0.1439, simple_loss=0.1367, pruned_loss=0.07555, over 1281.00 frames. ], tot_loss[loss=0.1116, simple_loss=0.1463, pruned_loss=0.03848, over 1959332.82 frames. ], batch size: 100, lr: 4.91e-03, grad_scale: 8.0 2022-12-08 09:39:08,732 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.7324, 4.5111, 4.3654, 4.7040, 4.3626, 4.0553, 4.7976, 4.5468], device='cuda:3'), covar=tensor([0.0599, 0.0768, 0.0792, 0.0560, 0.0725, 0.0726, 0.0515, 0.0726], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0143, 0.0145, 0.0160, 0.0146, 0.0124, 0.0168, 0.0147], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 09:39:23,470 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.271e+02 2.083e+02 2.539e+02 3.211e+02 7.278e+02, threshold=5.077e+02, percent-clipped=7.0 2022-12-08 09:39:46,101 INFO [train.py:873] (3/4) Epoch 16, batch 4400, loss[loss=0.1243, simple_loss=0.1286, pruned_loss=0.05996, over 2688.00 frames. ], tot_loss[loss=0.1108, simple_loss=0.1459, pruned_loss=0.03786, over 2011479.00 frames. ], batch size: 100, lr: 4.91e-03, grad_scale: 8.0 2022-12-08 09:40:26,825 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117877.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:40:31,763 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.57 vs. limit=2.0 2022-12-08 09:40:50,806 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.193e+02 2.093e+02 2.556e+02 3.184e+02 6.806e+02, threshold=5.111e+02, percent-clipped=1.0 2022-12-08 09:40:51,905 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.6968, 1.4525, 4.5092, 2.1522, 4.2936, 4.6846, 4.3366, 5.0991], device='cuda:3'), covar=tensor([0.0262, 0.3807, 0.0505, 0.2457, 0.0400, 0.0476, 0.0341, 0.0247], device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0156, 0.0160, 0.0169, 0.0167, 0.0180, 0.0133, 0.0153], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 09:41:09,114 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=117925.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:41:14,050 INFO [train.py:873] (3/4) Epoch 16, batch 4500, loss[loss=0.1262, simple_loss=0.1497, pruned_loss=0.05134, over 7771.00 frames. ], tot_loss[loss=0.1105, simple_loss=0.1455, pruned_loss=0.03777, over 1970164.65 frames. ], batch size: 100, lr: 4.91e-03, grad_scale: 8.0 2022-12-08 09:41:37,066 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7160, 4.0112, 3.5300, 3.7109, 2.9542, 3.9677, 3.8971, 2.1801], device='cuda:3'), covar=tensor([0.1384, 0.0794, 0.1582, 0.0491, 0.0754, 0.0695, 0.0793, 0.1823], device='cuda:3'), in_proj_covar=tensor([0.0138, 0.0088, 0.0068, 0.0072, 0.0097, 0.0087, 0.0100, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 09:42:01,233 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117985.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:42:15,727 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0497, 2.3312, 2.3324, 2.4290, 2.0281, 2.4489, 2.2616, 1.3730], device='cuda:3'), covar=tensor([0.0793, 0.0711, 0.0782, 0.0530, 0.0899, 0.0522, 0.0936, 0.1882], device='cuda:3'), in_proj_covar=tensor([0.0138, 0.0088, 0.0069, 0.0072, 0.0098, 0.0087, 0.0100, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 09:42:18,931 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.027e+02 2.156e+02 2.695e+02 3.844e+02 9.084e+02, threshold=5.389e+02, percent-clipped=7.0 2022-12-08 09:42:38,649 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2022-12-08 09:42:41,604 INFO [train.py:873] (3/4) Epoch 16, batch 4600, loss[loss=0.1077, simple_loss=0.1411, pruned_loss=0.03717, over 14040.00 frames. ], tot_loss[loss=0.1105, simple_loss=0.1456, pruned_loss=0.03769, over 2022678.78 frames. ], batch size: 19, lr: 4.90e-03, grad_scale: 8.0 2022-12-08 09:42:43,371 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=118033.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:42:49,983 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118040.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:43:42,510 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118101.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:43:45,464 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.349e+02 2.071e+02 2.553e+02 3.122e+02 5.122e+02, threshold=5.107e+02, percent-clipped=0.0 2022-12-08 09:44:08,312 INFO [train.py:873] (3/4) Epoch 16, batch 4700, loss[loss=0.1059, simple_loss=0.1476, pruned_loss=0.03212, over 14242.00 frames. ], tot_loss[loss=0.1097, simple_loss=0.1452, pruned_loss=0.03713, over 2057016.76 frames. ], batch size: 35, lr: 4.90e-03, grad_scale: 8.0 2022-12-08 09:44:40,815 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118168.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:45:13,137 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.145e+02 2.277e+02 2.708e+02 3.256e+02 5.977e+02, threshold=5.416e+02, percent-clipped=1.0 2022-12-08 09:45:21,980 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.49 vs. limit=5.0 2022-12-08 09:45:34,162 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118229.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 09:45:35,650 INFO [train.py:873] (3/4) Epoch 16, batch 4800, loss[loss=0.1235, simple_loss=0.1514, pruned_loss=0.04784, over 6943.00 frames. ], tot_loss[loss=0.1107, simple_loss=0.1455, pruned_loss=0.03794, over 1977778.43 frames. ], batch size: 100, lr: 4.90e-03, grad_scale: 8.0 2022-12-08 09:46:08,773 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118269.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:46:26,208 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118289.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:46:30,428 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.2529, 4.4463, 4.6864, 3.9142, 4.4873, 4.7557, 1.6646, 4.1618], device='cuda:3'), covar=tensor([0.0330, 0.0356, 0.0341, 0.0487, 0.0309, 0.0204, 0.3182, 0.0304], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0170, 0.0144, 0.0144, 0.0201, 0.0139, 0.0154, 0.0191], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 09:46:39,569 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.170e+02 2.128e+02 2.595e+02 3.127e+02 6.386e+02, threshold=5.189e+02, percent-clipped=3.0 2022-12-08 09:47:00,725 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118329.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:47:01,617 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118330.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:47:02,256 INFO [train.py:873] (3/4) Epoch 16, batch 4900, loss[loss=0.1241, simple_loss=0.1508, pruned_loss=0.0487, over 14533.00 frames. ], tot_loss[loss=0.1119, simple_loss=0.1464, pruned_loss=0.03872, over 1999918.86 frames. ], batch size: 49, lr: 4.90e-03, grad_scale: 8.0 2022-12-08 09:47:18,891 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118350.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:47:53,488 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118390.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:47:58,374 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118396.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:48:06,579 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.054e+02 2.113e+02 2.635e+02 3.192e+02 6.518e+02, threshold=5.270e+02, percent-clipped=3.0 2022-12-08 09:48:29,344 INFO [train.py:873] (3/4) Epoch 16, batch 5000, loss[loss=0.1334, simple_loss=0.1556, pruned_loss=0.05566, over 9502.00 frames. ], tot_loss[loss=0.1115, simple_loss=0.146, pruned_loss=0.03847, over 2064166.57 frames. ], batch size: 100, lr: 4.90e-03, grad_scale: 8.0 2022-12-08 09:49:34,276 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.516e+01 2.019e+02 2.454e+02 2.999e+02 6.382e+02, threshold=4.907e+02, percent-clipped=1.0 2022-12-08 09:49:45,487 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2022-12-08 09:49:50,416 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118524.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 09:49:56,305 INFO [train.py:873] (3/4) Epoch 16, batch 5100, loss[loss=0.1135, simple_loss=0.1449, pruned_loss=0.04104, over 14380.00 frames. ], tot_loss[loss=0.1111, simple_loss=0.1455, pruned_loss=0.03829, over 1981187.91 frames. ], batch size: 53, lr: 4.89e-03, grad_scale: 4.0 2022-12-08 09:50:20,333 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.5874, 1.1696, 2.0095, 1.7209, 1.8652, 2.0967, 1.3742, 2.0688], device='cuda:3'), covar=tensor([0.0827, 0.1383, 0.0280, 0.0597, 0.0667, 0.0298, 0.0766, 0.0285], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0161, 0.0132, 0.0172, 0.0150, 0.0144, 0.0126, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 09:51:01,826 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.319e+02 2.028e+02 2.469e+02 3.033e+02 6.307e+02, threshold=4.938e+02, percent-clipped=1.0 2022-12-08 09:51:12,304 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9230, 2.7541, 3.6155, 2.5043, 2.2432, 2.7747, 1.6627, 3.0547], device='cuda:3'), covar=tensor([0.1299, 0.1345, 0.0655, 0.1835, 0.2202, 0.1114, 0.3316, 0.1019], device='cuda:3'), in_proj_covar=tensor([0.0088, 0.0102, 0.0096, 0.0102, 0.0117, 0.0090, 0.0121, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 09:51:18,448 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118625.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:51:23,927 INFO [train.py:873] (3/4) Epoch 16, batch 5200, loss[loss=0.09441, simple_loss=0.1235, pruned_loss=0.03268, over 3848.00 frames. ], tot_loss[loss=0.1109, simple_loss=0.1457, pruned_loss=0.03808, over 1999535.01 frames. ], batch size: 100, lr: 4.89e-03, grad_scale: 8.0 2022-12-08 09:51:24,068 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118631.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:51:26,010 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.41 vs. limit=2.0 2022-12-08 09:51:36,587 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118645.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:51:38,406 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5431, 1.3655, 3.5706, 1.6065, 3.4465, 3.6645, 2.4925, 3.8952], device='cuda:3'), covar=tensor([0.0248, 0.3224, 0.0387, 0.2208, 0.0664, 0.0377, 0.0901, 0.0196], device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0156, 0.0160, 0.0169, 0.0165, 0.0180, 0.0133, 0.0153], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 09:51:45,459 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2022-12-08 09:52:12,086 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118685.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:52:17,667 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2022-12-08 09:52:18,079 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118692.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:52:21,211 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118696.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:52:29,671 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.100e+02 1.977e+02 2.482e+02 3.057e+02 6.335e+02, threshold=4.964e+02, percent-clipped=4.0 2022-12-08 09:52:50,032 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.12 vs. limit=5.0 2022-12-08 09:52:51,982 INFO [train.py:873] (3/4) Epoch 16, batch 5300, loss[loss=0.1462, simple_loss=0.1628, pruned_loss=0.06483, over 7780.00 frames. ], tot_loss[loss=0.1105, simple_loss=0.1451, pruned_loss=0.03796, over 1934192.40 frames. ], batch size: 100, lr: 4.89e-03, grad_scale: 8.0 2022-12-08 09:53:03,351 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=118744.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:53:38,937 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3242, 3.5465, 4.1235, 3.2670, 2.6510, 3.5312, 1.8334, 3.4099], device='cuda:3'), covar=tensor([0.1296, 0.0713, 0.0545, 0.1396, 0.1889, 0.0951, 0.3624, 0.1588], device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0102, 0.0095, 0.0102, 0.0117, 0.0090, 0.0121, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 09:53:58,970 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.455e+02 1.964e+02 2.394e+02 2.893e+02 4.670e+02, threshold=4.788e+02, percent-clipped=0.0 2022-12-08 09:54:14,787 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118824.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:54:20,952 INFO [train.py:873] (3/4) Epoch 16, batch 5400, loss[loss=0.109, simple_loss=0.1435, pruned_loss=0.03728, over 11157.00 frames. ], tot_loss[loss=0.1096, simple_loss=0.1442, pruned_loss=0.03748, over 1891613.85 frames. ], batch size: 100, lr: 4.89e-03, grad_scale: 8.0 2022-12-08 09:54:39,830 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2905, 1.4215, 3.3891, 1.5447, 3.1879, 3.3477, 2.4115, 3.6103], device='cuda:3'), covar=tensor([0.0248, 0.3050, 0.0398, 0.2231, 0.0857, 0.0424, 0.1012, 0.0207], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0155, 0.0159, 0.0168, 0.0165, 0.0179, 0.0132, 0.0152], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 09:54:57,518 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=118872.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:55:27,015 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.618e+01 2.141e+02 2.825e+02 3.414e+02 6.875e+02, threshold=5.649e+02, percent-clipped=6.0 2022-12-08 09:55:39,425 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0022, 2.4401, 3.8295, 4.1099, 3.8557, 2.4235, 4.0138, 3.0928], device='cuda:3'), covar=tensor([0.0409, 0.1169, 0.1014, 0.0487, 0.0491, 0.1725, 0.0439, 0.0930], device='cuda:3'), in_proj_covar=tensor([0.0292, 0.0259, 0.0373, 0.0327, 0.0271, 0.0304, 0.0308, 0.0280], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 09:55:44,037 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118925.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:55:49,199 INFO [train.py:873] (3/4) Epoch 16, batch 5500, loss[loss=0.1077, simple_loss=0.131, pruned_loss=0.04213, over 4932.00 frames. ], tot_loss[loss=0.1089, simple_loss=0.1442, pruned_loss=0.03683, over 1969824.40 frames. ], batch size: 100, lr: 4.88e-03, grad_scale: 8.0 2022-12-08 09:56:01,709 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118945.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:56:26,055 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=118973.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:56:36,636 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118985.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:56:38,468 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118987.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:56:43,690 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=118993.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:56:55,520 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.279e+02 2.181e+02 2.740e+02 3.306e+02 6.665e+02, threshold=5.480e+02, percent-clipped=3.0 2022-12-08 09:56:55,807 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.1953, 2.7222, 5.2347, 3.5743, 4.9222, 2.5800, 3.8033, 5.0901], device='cuda:3'), covar=tensor([0.0358, 0.3423, 0.0257, 0.5950, 0.0488, 0.2904, 0.1140, 0.0218], device='cuda:3'), in_proj_covar=tensor([0.0255, 0.0203, 0.0216, 0.0275, 0.0236, 0.0206, 0.0202, 0.0219], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0005], device='cuda:3') 2022-12-08 09:57:17,517 INFO [train.py:873] (3/4) Epoch 16, batch 5600, loss[loss=0.1078, simple_loss=0.1433, pruned_loss=0.0361, over 11164.00 frames. ], tot_loss[loss=0.1103, simple_loss=0.1453, pruned_loss=0.03769, over 1975269.44 frames. ], batch size: 100, lr: 4.88e-03, grad_scale: 8.0 2022-12-08 09:57:19,244 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=119033.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:57:51,185 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0677, 1.6928, 2.0918, 1.5001, 1.7887, 2.1061, 1.8986, 1.7976], device='cuda:3'), covar=tensor([0.0946, 0.0697, 0.0811, 0.1431, 0.1635, 0.0846, 0.0773, 0.1488], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0172, 0.0139, 0.0125, 0.0142, 0.0154, 0.0132, 0.0140], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0007, 0.0006, 0.0006], device='cuda:3') 2022-12-08 09:57:55,382 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119075.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:57:59,715 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2022-12-08 09:58:11,180 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119093.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:58:21,972 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.337e+02 2.186e+02 2.774e+02 3.458e+02 6.793e+02, threshold=5.547e+02, percent-clipped=4.0 2022-12-08 09:58:43,944 INFO [train.py:873] (3/4) Epoch 16, batch 5700, loss[loss=0.09984, simple_loss=0.1361, pruned_loss=0.03181, over 13610.00 frames. ], tot_loss[loss=0.1109, simple_loss=0.1454, pruned_loss=0.03822, over 1947475.19 frames. ], batch size: 17, lr: 4.88e-03, grad_scale: 8.0 2022-12-08 09:58:48,256 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119136.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 09:59:04,314 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119154.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 09:59:15,515 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.75 vs. limit=5.0 2022-12-08 09:59:32,302 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9544, 3.4293, 2.6822, 4.1136, 3.8992, 4.0091, 3.5600, 2.8519], device='cuda:3'), covar=tensor([0.0728, 0.1294, 0.3283, 0.0533, 0.0726, 0.0963, 0.1003, 0.2937], device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0292, 0.0262, 0.0284, 0.0323, 0.0300, 0.0254, 0.0244], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 09:59:39,196 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4467, 3.2378, 3.1870, 3.5144, 3.2714, 3.4187, 3.5324, 2.9479], device='cuda:3'), covar=tensor([0.0624, 0.1112, 0.0554, 0.0511, 0.0928, 0.0459, 0.0643, 0.0657], device='cuda:3'), in_proj_covar=tensor([0.0179, 0.0275, 0.0199, 0.0192, 0.0186, 0.0156, 0.0286, 0.0169], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 09:59:50,094 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.363e+02 2.314e+02 2.766e+02 3.252e+02 7.674e+02, threshold=5.531e+02, percent-clipped=3.0 2022-12-08 10:00:11,371 INFO [train.py:873] (3/4) Epoch 16, batch 5800, loss[loss=0.1182, simple_loss=0.1548, pruned_loss=0.04077, over 14289.00 frames. ], tot_loss[loss=0.1113, simple_loss=0.1457, pruned_loss=0.0385, over 1947988.91 frames. ], batch size: 35, lr: 4.88e-03, grad_scale: 8.0 2022-12-08 10:01:00,861 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=119287.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:01:17,095 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.032e+02 2.269e+02 2.799e+02 3.367e+02 7.852e+02, threshold=5.598e+02, percent-clipped=4.0 2022-12-08 10:01:38,988 INFO [train.py:873] (3/4) Epoch 16, batch 5900, loss[loss=0.1197, simple_loss=0.1502, pruned_loss=0.04457, over 9494.00 frames. ], tot_loss[loss=0.1108, simple_loss=0.1453, pruned_loss=0.03817, over 1948008.02 frames. ], batch size: 100, lr: 4.88e-03, grad_scale: 4.0 2022-12-08 10:01:42,503 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=119335.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:02:24,882 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5786, 1.8944, 2.4582, 2.0475, 2.5067, 2.4275, 2.3370, 2.2327], device='cuda:3'), covar=tensor([0.0777, 0.2590, 0.0819, 0.1554, 0.0595, 0.1077, 0.0882, 0.1352], device='cuda:3'), in_proj_covar=tensor([0.0357, 0.0312, 0.0397, 0.0301, 0.0372, 0.0325, 0.0362, 0.0302], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 10:02:45,650 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.338e+02 2.147e+02 2.622e+02 3.177e+02 5.866e+02, threshold=5.244e+02, percent-clipped=1.0 2022-12-08 10:02:46,601 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.45 vs. limit=5.0 2022-12-08 10:03:05,592 INFO [train.py:873] (3/4) Epoch 16, batch 6000, loss[loss=0.1128, simple_loss=0.1436, pruned_loss=0.041, over 6969.00 frames. ], tot_loss[loss=0.1104, simple_loss=0.1452, pruned_loss=0.03786, over 2017220.58 frames. ], batch size: 100, lr: 4.87e-03, grad_scale: 4.0 2022-12-08 10:03:05,592 INFO [train.py:896] (3/4) Computing validation loss 2022-12-08 10:03:14,045 INFO [train.py:905] (3/4) Epoch 16, validation: loss=0.1378, simple_loss=0.175, pruned_loss=0.05031, over 857387.00 frames. 2022-12-08 10:03:14,046 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-08 10:03:14,140 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=119431.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:03:14,254 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8250, 1.8043, 1.9978, 1.8312, 1.8165, 1.6829, 1.5075, 1.2805], device='cuda:3'), covar=tensor([0.0187, 0.0365, 0.0221, 0.0201, 0.0230, 0.0284, 0.0272, 0.0395], device='cuda:3'), in_proj_covar=tensor([0.0021, 0.0021, 0.0019, 0.0020, 0.0020, 0.0032, 0.0027, 0.0031], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 10:03:29,659 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=119449.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 10:03:53,025 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8460, 3.9020, 3.1553, 3.3478, 3.7692, 3.9094, 4.0362, 3.9727], device='cuda:3'), covar=tensor([0.1587, 0.0713, 0.3554, 0.4420, 0.1456, 0.1477, 0.1340, 0.1335], device='cuda:3'), in_proj_covar=tensor([0.0392, 0.0269, 0.0451, 0.0567, 0.0350, 0.0450, 0.0393, 0.0388], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 10:03:56,912 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.55 vs. limit=5.0 2022-12-08 10:04:01,846 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2022-12-08 10:04:20,754 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.903e+01 2.074e+02 2.533e+02 3.058e+02 7.007e+02, threshold=5.066e+02, percent-clipped=2.0 2022-12-08 10:04:26,192 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119514.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:04:28,426 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.0705, 2.6359, 5.0954, 3.4143, 4.7099, 2.2605, 3.6831, 4.8254], device='cuda:3'), covar=tensor([0.0593, 0.3787, 0.0302, 0.6239, 0.0582, 0.3569, 0.1309, 0.0377], device='cuda:3'), in_proj_covar=tensor([0.0253, 0.0202, 0.0215, 0.0274, 0.0236, 0.0207, 0.0200, 0.0218], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0005], device='cuda:3') 2022-12-08 10:04:41,255 INFO [train.py:873] (3/4) Epoch 16, batch 6100, loss[loss=0.1119, simple_loss=0.1473, pruned_loss=0.03828, over 12783.00 frames. ], tot_loss[loss=0.11, simple_loss=0.1451, pruned_loss=0.03748, over 2027237.74 frames. ], batch size: 100, lr: 4.87e-03, grad_scale: 4.0 2022-12-08 10:04:49,149 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.3852, 1.4483, 1.4840, 1.3102, 1.2568, 1.1129, 1.1307, 1.0390], device='cuda:3'), covar=tensor([0.0234, 0.0236, 0.0195, 0.0233, 0.0233, 0.0396, 0.0302, 0.0440], device='cuda:3'), in_proj_covar=tensor([0.0021, 0.0021, 0.0019, 0.0020, 0.0020, 0.0032, 0.0027, 0.0031], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 10:05:19,860 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119575.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:05:48,894 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.289e+02 1.899e+02 2.518e+02 3.457e+02 7.704e+02, threshold=5.035e+02, percent-clipped=3.0 2022-12-08 10:06:05,313 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.83 vs. limit=5.0 2022-12-08 10:06:09,184 INFO [train.py:873] (3/4) Epoch 16, batch 6200, loss[loss=0.1049, simple_loss=0.1406, pruned_loss=0.03463, over 14268.00 frames. ], tot_loss[loss=0.1097, simple_loss=0.1446, pruned_loss=0.03741, over 1966222.65 frames. ], batch size: 89, lr: 4.87e-03, grad_scale: 4.0 2022-12-08 10:06:31,173 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3606, 2.9932, 2.8585, 1.9442, 2.8470, 3.1773, 3.4130, 2.6166], device='cuda:3'), covar=tensor([0.0589, 0.1131, 0.0885, 0.1558, 0.0760, 0.0604, 0.0546, 0.1155], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0172, 0.0140, 0.0125, 0.0142, 0.0154, 0.0132, 0.0141], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0007, 0.0006, 0.0006], device='cuda:3') 2022-12-08 10:06:38,645 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119665.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:06:55,965 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9083, 3.4524, 2.8262, 4.1367, 3.9753, 4.0212, 3.4803, 2.8436], device='cuda:3'), covar=tensor([0.0767, 0.1238, 0.3027, 0.0463, 0.0666, 0.1059, 0.1059, 0.2980], device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0292, 0.0261, 0.0285, 0.0323, 0.0300, 0.0255, 0.0242], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 10:07:16,891 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.045e+02 2.165e+02 2.712e+02 3.301e+02 6.973e+02, threshold=5.424e+02, percent-clipped=2.0 2022-12-08 10:07:32,863 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119726.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:07:37,350 INFO [train.py:873] (3/4) Epoch 16, batch 6300, loss[loss=0.114, simple_loss=0.1517, pruned_loss=0.03817, over 14453.00 frames. ], tot_loss[loss=0.1098, simple_loss=0.1446, pruned_loss=0.0375, over 1936999.34 frames. ], batch size: 51, lr: 4.87e-03, grad_scale: 4.0 2022-12-08 10:07:37,508 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=119731.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:07:53,606 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=119749.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 10:08:02,361 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8628, 3.6441, 3.3165, 2.7512, 3.3088, 3.5849, 3.9307, 3.2076], device='cuda:3'), covar=tensor([0.0527, 0.1013, 0.0901, 0.1137, 0.0743, 0.0660, 0.0608, 0.0915], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0172, 0.0140, 0.0125, 0.0142, 0.0154, 0.0133, 0.0140], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0007, 0.0006, 0.0006], device='cuda:3') 2022-12-08 10:08:11,954 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8860, 4.9908, 5.3626, 4.5313, 5.0806, 5.4145, 2.0241, 4.7696], device='cuda:3'), covar=tensor([0.0284, 0.0232, 0.0273, 0.0386, 0.0260, 0.0133, 0.2866, 0.0266], device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0173, 0.0145, 0.0144, 0.0202, 0.0139, 0.0157, 0.0192], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 10:08:19,729 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=119779.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:08:35,919 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=119797.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:08:46,036 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.350e+02 2.194e+02 2.675e+02 3.528e+02 1.237e+03, threshold=5.350e+02, percent-clipped=2.0 2022-12-08 10:09:05,869 INFO [train.py:873] (3/4) Epoch 16, batch 6400, loss[loss=0.1288, simple_loss=0.1387, pruned_loss=0.0594, over 3870.00 frames. ], tot_loss[loss=0.1093, simple_loss=0.1443, pruned_loss=0.03719, over 1928720.44 frames. ], batch size: 100, lr: 4.87e-03, grad_scale: 8.0 2022-12-08 10:09:40,546 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=119870.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:09:55,359 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.2287, 5.0933, 4.8136, 5.2749, 4.8058, 4.7482, 5.3457, 5.0184], device='cuda:3'), covar=tensor([0.0609, 0.0771, 0.0676, 0.0520, 0.0769, 0.0416, 0.0454, 0.0692], device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0141, 0.0145, 0.0158, 0.0146, 0.0121, 0.0166, 0.0146], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 10:10:13,252 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.254e+02 2.000e+02 2.417e+02 3.320e+02 6.714e+02, threshold=4.834e+02, percent-clipped=5.0 2022-12-08 10:10:22,376 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2022-12-08 10:10:33,940 INFO [train.py:873] (3/4) Epoch 16, batch 6500, loss[loss=0.1166, simple_loss=0.1329, pruned_loss=0.0502, over 3865.00 frames. ], tot_loss[loss=0.1093, simple_loss=0.1441, pruned_loss=0.03724, over 1899824.17 frames. ], batch size: 100, lr: 4.86e-03, grad_scale: 8.0 2022-12-08 10:11:01,991 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9906, 2.0957, 1.9165, 2.1437, 1.7125, 2.0043, 2.0749, 2.0186], device='cuda:3'), covar=tensor([0.0996, 0.1127, 0.1214, 0.0883, 0.2078, 0.0868, 0.1107, 0.1054], device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0141, 0.0146, 0.0158, 0.0146, 0.0121, 0.0167, 0.0147], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 10:11:44,556 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.130e+02 1.947e+02 2.582e+02 3.066e+02 4.807e+02, threshold=5.164e+02, percent-clipped=0.0 2022-12-08 10:11:55,928 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=120021.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:12:04,195 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2022-12-08 10:12:04,536 INFO [train.py:873] (3/4) Epoch 16, batch 6600, loss[loss=0.1852, simple_loss=0.1759, pruned_loss=0.09722, over 1210.00 frames. ], tot_loss[loss=0.1092, simple_loss=0.1443, pruned_loss=0.03704, over 1885367.74 frames. ], batch size: 100, lr: 4.86e-03, grad_scale: 8.0 2022-12-08 10:12:32,425 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2022-12-08 10:12:59,305 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.55 vs. limit=2.0 2022-12-08 10:13:12,086 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.190e+01 2.035e+02 2.583e+02 3.251e+02 5.143e+02, threshold=5.166e+02, percent-clipped=0.0 2022-12-08 10:13:14,931 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9874, 2.5043, 3.8238, 2.8969, 3.8586, 3.7062, 3.6728, 3.2362], device='cuda:3'), covar=tensor([0.0890, 0.3114, 0.1145, 0.1826, 0.0784, 0.1043, 0.1532, 0.1674], device='cuda:3'), in_proj_covar=tensor([0.0355, 0.0311, 0.0394, 0.0301, 0.0369, 0.0324, 0.0361, 0.0299], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 10:13:32,126 INFO [train.py:873] (3/4) Epoch 16, batch 6700, loss[loss=0.1094, simple_loss=0.1151, pruned_loss=0.05188, over 1249.00 frames. ], tot_loss[loss=0.11, simple_loss=0.145, pruned_loss=0.03744, over 1901721.23 frames. ], batch size: 100, lr: 4.86e-03, grad_scale: 8.0 2022-12-08 10:13:34,250 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.73 vs. limit=2.0 2022-12-08 10:14:05,969 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=120170.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:14:39,089 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.095e+02 2.165e+02 2.458e+02 2.899e+02 9.707e+02, threshold=4.916e+02, percent-clipped=4.0 2022-12-08 10:14:48,305 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=120218.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:14:59,585 INFO [train.py:873] (3/4) Epoch 16, batch 6800, loss[loss=0.1535, simple_loss=0.1555, pruned_loss=0.07576, over 1238.00 frames. ], tot_loss[loss=0.1103, simple_loss=0.1451, pruned_loss=0.03775, over 1847185.41 frames. ], batch size: 100, lr: 4.86e-03, grad_scale: 8.0 2022-12-08 10:16:06,648 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.522e+02 2.118e+02 2.648e+02 3.259e+02 6.407e+02, threshold=5.296e+02, percent-clipped=3.0 2022-12-08 10:16:14,472 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2022-12-08 10:16:18,542 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=120321.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:16:27,163 INFO [train.py:873] (3/4) Epoch 16, batch 6900, loss[loss=0.1455, simple_loss=0.1405, pruned_loss=0.07526, over 1218.00 frames. ], tot_loss[loss=0.1097, simple_loss=0.1448, pruned_loss=0.03734, over 1892643.90 frames. ], batch size: 100, lr: 4.86e-03, grad_scale: 8.0 2022-12-08 10:17:00,690 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=120369.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:17:06,202 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2079, 2.9819, 2.3116, 3.2271, 3.0847, 3.1373, 2.6868, 2.3471], device='cuda:3'), covar=tensor([0.0854, 0.1335, 0.3047, 0.0802, 0.0960, 0.1063, 0.1534, 0.2979], device='cuda:3'), in_proj_covar=tensor([0.0282, 0.0292, 0.0260, 0.0283, 0.0323, 0.0301, 0.0254, 0.0242], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 10:17:32,661 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2022-12-08 10:17:34,669 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.039e+02 2.083e+02 2.449e+02 3.128e+02 5.961e+02, threshold=4.898e+02, percent-clipped=1.0 2022-12-08 10:17:46,524 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.7374, 1.7402, 4.5372, 2.4260, 4.3750, 4.6493, 4.3788, 5.1392], device='cuda:3'), covar=tensor([0.0226, 0.3099, 0.0363, 0.2032, 0.0337, 0.0449, 0.0282, 0.0190], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0153, 0.0158, 0.0167, 0.0166, 0.0178, 0.0131, 0.0151], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 10:17:55,022 INFO [train.py:873] (3/4) Epoch 16, batch 7000, loss[loss=0.1013, simple_loss=0.1462, pruned_loss=0.02819, over 14197.00 frames. ], tot_loss[loss=0.1097, simple_loss=0.145, pruned_loss=0.03725, over 1960941.62 frames. ], batch size: 80, lr: 4.85e-03, grad_scale: 8.0 2022-12-08 10:18:59,681 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.40 vs. limit=5.0 2022-12-08 10:19:01,658 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.060e+01 2.199e+02 2.820e+02 3.385e+02 6.936e+02, threshold=5.640e+02, percent-clipped=6.0 2022-12-08 10:19:21,718 INFO [train.py:873] (3/4) Epoch 16, batch 7100, loss[loss=0.107, simple_loss=0.145, pruned_loss=0.03447, over 14277.00 frames. ], tot_loss[loss=0.1093, simple_loss=0.1446, pruned_loss=0.03704, over 1904157.19 frames. ], batch size: 66, lr: 4.85e-03, grad_scale: 8.0 2022-12-08 10:20:28,194 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.979e+01 2.056e+02 2.495e+02 3.157e+02 6.004e+02, threshold=4.990e+02, percent-clipped=1.0 2022-12-08 10:20:48,804 INFO [train.py:873] (3/4) Epoch 16, batch 7200, loss[loss=0.1174, simple_loss=0.1524, pruned_loss=0.04122, over 14072.00 frames. ], tot_loss[loss=0.1099, simple_loss=0.1448, pruned_loss=0.03746, over 1940465.06 frames. ], batch size: 22, lr: 4.85e-03, grad_scale: 8.0 2022-12-08 10:21:27,882 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=120676.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:21:55,119 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.284e+02 2.056e+02 2.553e+02 2.947e+02 6.025e+02, threshold=5.105e+02, percent-clipped=4.0 2022-12-08 10:22:12,824 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4141, 2.4922, 2.4807, 2.2314, 2.2298, 1.3947, 2.2886, 2.4897], device='cuda:3'), covar=tensor([0.1944, 0.0606, 0.0699, 0.1509, 0.0755, 0.0857, 0.1401, 0.0592], device='cuda:3'), in_proj_covar=tensor([0.0036, 0.0035, 0.0038, 0.0033, 0.0034, 0.0047, 0.0036, 0.0038], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 10:22:15,427 INFO [train.py:873] (3/4) Epoch 16, batch 7300, loss[loss=0.1227, simple_loss=0.1455, pruned_loss=0.04995, over 3836.00 frames. ], tot_loss[loss=0.1089, simple_loss=0.1441, pruned_loss=0.03681, over 1950955.58 frames. ], batch size: 100, lr: 4.85e-03, grad_scale: 8.0 2022-12-08 10:22:21,073 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=120737.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:22:38,593 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3075, 3.9825, 3.9374, 4.3198, 4.0954, 3.8499, 4.3561, 3.6375], device='cuda:3'), covar=tensor([0.0450, 0.1031, 0.0470, 0.0475, 0.0724, 0.1081, 0.0618, 0.0578], device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0276, 0.0199, 0.0193, 0.0185, 0.0157, 0.0287, 0.0170], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 10:23:20,523 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.73 vs. limit=2.0 2022-12-08 10:23:22,447 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.271e+02 2.181e+02 2.578e+02 3.347e+02 7.648e+02, threshold=5.157e+02, percent-clipped=5.0 2022-12-08 10:23:30,166 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.89 vs. limit=5.0 2022-12-08 10:23:31,728 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4375, 4.1932, 3.9947, 4.5132, 4.2372, 3.9280, 4.5181, 3.8360], device='cuda:3'), covar=tensor([0.0413, 0.0893, 0.0447, 0.0378, 0.0746, 0.1049, 0.0515, 0.0454], device='cuda:3'), in_proj_covar=tensor([0.0179, 0.0275, 0.0199, 0.0192, 0.0185, 0.0157, 0.0287, 0.0169], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 10:23:37,598 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.70 vs. limit=2.0 2022-12-08 10:23:38,891 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6337, 1.6115, 1.6686, 1.5366, 1.5078, 1.4064, 1.2304, 1.0829], device='cuda:3'), covar=tensor([0.0196, 0.0330, 0.0272, 0.0219, 0.0232, 0.0373, 0.0316, 0.0476], device='cuda:3'), in_proj_covar=tensor([0.0021, 0.0022, 0.0019, 0.0020, 0.0020, 0.0033, 0.0027, 0.0032], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 10:23:42,646 INFO [train.py:873] (3/4) Epoch 16, batch 7400, loss[loss=0.1183, simple_loss=0.1404, pruned_loss=0.04813, over 3830.00 frames. ], tot_loss[loss=0.1095, simple_loss=0.1447, pruned_loss=0.03711, over 1996595.10 frames. ], batch size: 100, lr: 4.85e-03, grad_scale: 8.0 2022-12-08 10:24:01,168 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.2782, 3.7021, 2.8860, 4.4683, 4.0425, 4.2335, 3.8208, 2.9270], device='cuda:3'), covar=tensor([0.0549, 0.1187, 0.3267, 0.0437, 0.0897, 0.1494, 0.0953, 0.2923], device='cuda:3'), in_proj_covar=tensor([0.0282, 0.0290, 0.0259, 0.0284, 0.0325, 0.0300, 0.0254, 0.0243], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 10:24:30,303 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.72 vs. limit=2.0 2022-12-08 10:24:49,562 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.390e+02 2.203e+02 2.672e+02 3.215e+02 8.256e+02, threshold=5.345e+02, percent-clipped=4.0 2022-12-08 10:25:09,744 INFO [train.py:873] (3/4) Epoch 16, batch 7500, loss[loss=0.09248, simple_loss=0.134, pruned_loss=0.02548, over 14330.00 frames. ], tot_loss[loss=0.1095, simple_loss=0.1451, pruned_loss=0.03697, over 2056617.12 frames. ], batch size: 44, lr: 4.84e-03, grad_scale: 8.0 2022-12-08 10:25:31,472 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2022-12-08 10:25:43,978 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8287, 1.5610, 2.0706, 1.5962, 1.9077, 1.3943, 1.6407, 1.9358], device='cuda:3'), covar=tensor([0.3154, 0.3026, 0.0747, 0.1760, 0.1460, 0.1589, 0.1114, 0.0912], device='cuda:3'), in_proj_covar=tensor([0.0253, 0.0204, 0.0218, 0.0272, 0.0236, 0.0206, 0.0202, 0.0218], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0005], device='cuda:3') 2022-12-08 10:26:35,450 INFO [train.py:873] (3/4) Epoch 17, batch 0, loss[loss=0.1286, simple_loss=0.1582, pruned_loss=0.04954, over 14227.00 frames. ], tot_loss[loss=0.1286, simple_loss=0.1582, pruned_loss=0.04954, over 14227.00 frames. ], batch size: 80, lr: 4.70e-03, grad_scale: 8.0 2022-12-08 10:26:35,450 INFO [train.py:896] (3/4) Computing validation loss 2022-12-08 10:26:39,175 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5244, 2.2915, 3.3893, 3.5819, 3.4232, 2.3601, 3.4561, 2.7139], device='cuda:3'), covar=tensor([0.0389, 0.1164, 0.0678, 0.0415, 0.0523, 0.1658, 0.0507, 0.1012], device='cuda:3'), in_proj_covar=tensor([0.0294, 0.0260, 0.0376, 0.0333, 0.0275, 0.0307, 0.0313, 0.0282], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 10:26:39,278 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.9392, 5.4497, 5.4997, 5.8918, 5.3365, 4.9738, 5.7368, 5.3369], device='cuda:3'), covar=tensor([0.0146, 0.0403, 0.0222, 0.0177, 0.0711, 0.0147, 0.0344, 0.0307], device='cuda:3'), in_proj_covar=tensor([0.0181, 0.0279, 0.0200, 0.0194, 0.0187, 0.0157, 0.0291, 0.0171], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 10:26:40,262 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4485, 2.3792, 2.3560, 2.3085, 2.4074, 1.3607, 2.4413, 2.3758], device='cuda:3'), covar=tensor([0.0584, 0.0456, 0.0709, 0.1569, 0.0782, 0.0824, 0.0764, 0.1077], device='cuda:3'), in_proj_covar=tensor([0.0035, 0.0034, 0.0037, 0.0032, 0.0033, 0.0046, 0.0035, 0.0037], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 10:26:42,840 INFO [train.py:905] (3/4) Epoch 17, validation: loss=0.1441, simple_loss=0.1813, pruned_loss=0.05348, over 857387.00 frames. 2022-12-08 10:26:42,841 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-08 10:26:52,040 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=121003.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:26:56,390 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.236e+01 1.830e+02 2.606e+02 3.490e+02 1.014e+03, threshold=5.212e+02, percent-clipped=7.0 2022-12-08 10:27:17,874 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=121032.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:27:39,343 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.7913, 3.0267, 4.4732, 3.3790, 4.5707, 4.3698, 4.2765, 3.8355], device='cuda:3'), covar=tensor([0.0685, 0.3039, 0.1147, 0.1700, 0.0742, 0.0925, 0.1692, 0.1718], device='cuda:3'), in_proj_covar=tensor([0.0354, 0.0313, 0.0392, 0.0302, 0.0369, 0.0322, 0.0360, 0.0300], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 10:27:46,277 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=121064.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:27:58,126 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.52 vs. limit=2.0 2022-12-08 10:28:02,849 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1789, 1.6792, 2.1441, 1.3638, 1.9638, 2.1759, 2.0697, 1.8620], device='cuda:3'), covar=tensor([0.0948, 0.0695, 0.0935, 0.1549, 0.1342, 0.0947, 0.0714, 0.1567], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0171, 0.0140, 0.0126, 0.0143, 0.0154, 0.0133, 0.0143], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0007, 0.0006, 0.0006], device='cuda:3') 2022-12-08 10:28:10,669 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.01 vs. limit=5.0 2022-12-08 10:28:11,824 INFO [train.py:873] (3/4) Epoch 17, batch 100, loss[loss=0.1097, simple_loss=0.1467, pruned_loss=0.03638, over 14209.00 frames. ], tot_loss[loss=0.1085, simple_loss=0.1438, pruned_loss=0.0366, over 821279.65 frames. ], batch size: 57, lr: 4.69e-03, grad_scale: 8.0 2022-12-08 10:28:24,822 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.457e+02 2.352e+02 2.943e+02 3.420e+02 5.915e+02, threshold=5.886e+02, percent-clipped=2.0 2022-12-08 10:28:37,077 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9604, 1.6646, 3.1016, 2.7520, 2.9360, 3.1095, 2.2520, 3.0704], device='cuda:3'), covar=tensor([0.1333, 0.1447, 0.0193, 0.0513, 0.0415, 0.0231, 0.0700, 0.0221], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0161, 0.0131, 0.0170, 0.0149, 0.0144, 0.0126, 0.0125], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 10:29:10,588 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.15 vs. limit=5.0 2022-12-08 10:29:13,698 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=121164.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:29:19,940 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0676, 1.6036, 3.9552, 1.9327, 4.0103, 4.1350, 3.2133, 4.4906], device='cuda:3'), covar=tensor([0.0217, 0.3216, 0.0413, 0.2139, 0.0386, 0.0359, 0.0710, 0.0163], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0154, 0.0158, 0.0166, 0.0165, 0.0178, 0.0132, 0.0151], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 10:29:39,131 INFO [train.py:873] (3/4) Epoch 17, batch 200, loss[loss=0.1535, simple_loss=0.1723, pruned_loss=0.06736, over 8626.00 frames. ], tot_loss[loss=0.1088, simple_loss=0.144, pruned_loss=0.03684, over 1300011.22 frames. ], batch size: 100, lr: 4.69e-03, grad_scale: 8.0 2022-12-08 10:29:52,502 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.399e+02 2.132e+02 2.532e+02 3.029e+02 6.779e+02, threshold=5.065e+02, percent-clipped=1.0 2022-12-08 10:29:57,828 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8031, 1.8356, 1.6260, 1.8804, 1.7437, 1.7841, 1.8108, 1.6877], device='cuda:3'), covar=tensor([0.1398, 0.1048, 0.1879, 0.0890, 0.1388, 0.0678, 0.1560, 0.1164], device='cuda:3'), in_proj_covar=tensor([0.0283, 0.0290, 0.0258, 0.0284, 0.0323, 0.0299, 0.0253, 0.0242], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 10:30:06,995 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=121225.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:30:47,176 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 2022-12-08 10:30:48,447 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1706, 1.2186, 0.8353, 1.1391, 1.1618, 0.7171, 0.9662, 1.1328], device='cuda:3'), covar=tensor([0.0468, 0.0504, 0.0486, 0.0485, 0.0400, 0.0429, 0.1059, 0.0846], device='cuda:3'), in_proj_covar=tensor([0.0035, 0.0034, 0.0037, 0.0032, 0.0033, 0.0046, 0.0035, 0.0037], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 10:31:06,419 INFO [train.py:873] (3/4) Epoch 17, batch 300, loss[loss=0.1075, simple_loss=0.1082, pruned_loss=0.05341, over 1244.00 frames. ], tot_loss[loss=0.1087, simple_loss=0.1439, pruned_loss=0.03678, over 1508748.40 frames. ], batch size: 100, lr: 4.69e-03, grad_scale: 8.0 2022-12-08 10:31:19,308 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.075e+02 2.023e+02 2.561e+02 3.068e+02 6.366e+02, threshold=5.121e+02, percent-clipped=2.0 2022-12-08 10:31:22,075 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=121311.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 10:31:34,633 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.45 vs. limit=5.0 2022-12-08 10:31:37,048 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.32 vs. limit=5.0 2022-12-08 10:31:40,193 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=121332.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:31:41,930 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7371, 1.5386, 2.9109, 1.5475, 2.9946, 2.9224, 2.2059, 3.0974], device='cuda:3'), covar=tensor([0.0278, 0.2718, 0.0394, 0.2052, 0.0338, 0.0462, 0.1029, 0.0255], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0155, 0.0159, 0.0168, 0.0167, 0.0179, 0.0133, 0.0153], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 10:32:02,931 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1998, 1.9817, 2.2057, 2.2543, 2.0345, 2.0024, 2.2743, 2.1577], device='cuda:3'), covar=tensor([0.0342, 0.0669, 0.0390, 0.0324, 0.0518, 0.0745, 0.0388, 0.0406], device='cuda:3'), in_proj_covar=tensor([0.0293, 0.0258, 0.0375, 0.0331, 0.0273, 0.0305, 0.0311, 0.0280], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 10:32:03,755 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=121359.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:32:06,134 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.63 vs. limit=2.0 2022-12-08 10:32:09,745 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2040, 2.9444, 2.9033, 2.1224, 2.6926, 2.9350, 3.3514, 2.5252], device='cuda:3'), covar=tensor([0.0630, 0.0762, 0.0775, 0.1200, 0.0838, 0.0709, 0.0475, 0.1100], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0172, 0.0142, 0.0127, 0.0143, 0.0154, 0.0133, 0.0144], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0007, 0.0006, 0.0006], device='cuda:3') 2022-12-08 10:32:14,903 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=121372.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 10:32:22,001 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=121380.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:32:33,193 INFO [train.py:873] (3/4) Epoch 17, batch 400, loss[loss=0.1218, simple_loss=0.1293, pruned_loss=0.05715, over 2601.00 frames. ], tot_loss[loss=0.1082, simple_loss=0.1436, pruned_loss=0.03641, over 1680905.41 frames. ], batch size: 100, lr: 4.69e-03, grad_scale: 8.0 2022-12-08 10:32:46,077 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=121407.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:32:47,376 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.084e+02 2.168e+02 2.580e+02 3.311e+02 5.353e+02, threshold=5.159e+02, percent-clipped=2.0 2022-12-08 10:33:38,750 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3944, 2.1911, 2.6392, 1.6617, 1.7304, 2.4381, 1.3450, 2.3315], device='cuda:3'), covar=tensor([0.0854, 0.1500, 0.0629, 0.2127, 0.2326, 0.0739, 0.3334, 0.1015], device='cuda:3'), in_proj_covar=tensor([0.0086, 0.0102, 0.0096, 0.0100, 0.0117, 0.0091, 0.0120, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 10:33:38,796 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=121468.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:34:00,288 INFO [train.py:873] (3/4) Epoch 17, batch 500, loss[loss=0.09808, simple_loss=0.1435, pruned_loss=0.02631, over 14297.00 frames. ], tot_loss[loss=0.1082, simple_loss=0.1437, pruned_loss=0.03636, over 1752352.94 frames. ], batch size: 60, lr: 4.69e-03, grad_scale: 8.0 2022-12-08 10:34:14,595 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.238e+02 2.218e+02 2.633e+02 3.024e+02 9.780e+02, threshold=5.266e+02, percent-clipped=2.0 2022-12-08 10:34:18,398 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 2022-12-08 10:34:23,951 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=121520.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:34:24,380 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2022-12-08 10:35:17,950 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8994, 1.3681, 2.0010, 1.2761, 1.9316, 2.0531, 1.6964, 2.1540], device='cuda:3'), covar=tensor([0.0315, 0.2248, 0.0578, 0.2007, 0.0672, 0.0741, 0.1204, 0.0467], device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0157, 0.0160, 0.0168, 0.0168, 0.0180, 0.0133, 0.0153], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 10:35:27,462 INFO [train.py:873] (3/4) Epoch 17, batch 600, loss[loss=0.1007, simple_loss=0.1339, pruned_loss=0.03373, over 6905.00 frames. ], tot_loss[loss=0.1077, simple_loss=0.1433, pruned_loss=0.0361, over 1823119.84 frames. ], batch size: 100, lr: 4.69e-03, grad_scale: 8.0 2022-12-08 10:35:41,066 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.467e+02 2.288e+02 2.719e+02 3.282e+02 7.294e+02, threshold=5.437e+02, percent-clipped=3.0 2022-12-08 10:36:25,601 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=121659.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:36:28,260 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.91 vs. limit=5.0 2022-12-08 10:36:32,519 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=121667.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 10:36:33,597 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8343, 3.7115, 3.3444, 2.8264, 3.3007, 3.6163, 3.9491, 3.2348], device='cuda:3'), covar=tensor([0.0527, 0.0969, 0.0841, 0.1043, 0.0819, 0.0613, 0.0503, 0.0919], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0170, 0.0140, 0.0126, 0.0143, 0.0153, 0.0133, 0.0142], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 10:36:55,474 INFO [train.py:873] (3/4) Epoch 17, batch 700, loss[loss=0.1321, simple_loss=0.1605, pruned_loss=0.05189, over 9467.00 frames. ], tot_loss[loss=0.1087, simple_loss=0.1439, pruned_loss=0.03676, over 1884994.83 frames. ], batch size: 100, lr: 4.68e-03, grad_scale: 8.0 2022-12-08 10:37:08,192 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=121707.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:37:09,978 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.050e+02 2.051e+02 2.543e+02 3.291e+02 5.675e+02, threshold=5.087e+02, percent-clipped=1.0 2022-12-08 10:37:57,028 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=121763.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:38:23,112 INFO [train.py:873] (3/4) Epoch 17, batch 800, loss[loss=0.09333, simple_loss=0.1379, pruned_loss=0.02436, over 14263.00 frames. ], tot_loss[loss=0.1086, simple_loss=0.1441, pruned_loss=0.03657, over 1948987.60 frames. ], batch size: 63, lr: 4.68e-03, grad_scale: 8.0 2022-12-08 10:38:37,354 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.375e+02 2.062e+02 2.469e+02 3.248e+02 9.190e+02, threshold=4.938e+02, percent-clipped=3.0 2022-12-08 10:38:47,643 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=121820.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:39:29,524 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=121868.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:39:51,405 INFO [train.py:873] (3/4) Epoch 17, batch 900, loss[loss=0.1026, simple_loss=0.1437, pruned_loss=0.03073, over 14215.00 frames. ], tot_loss[loss=0.1088, simple_loss=0.1443, pruned_loss=0.03671, over 1985081.59 frames. ], batch size: 69, lr: 4.68e-03, grad_scale: 8.0 2022-12-08 10:39:55,750 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1372, 2.4026, 4.2059, 4.2561, 4.1329, 2.5563, 4.3679, 3.0879], device='cuda:3'), covar=tensor([0.0491, 0.1362, 0.0815, 0.0532, 0.0452, 0.1863, 0.0401, 0.1082], device='cuda:3'), in_proj_covar=tensor([0.0289, 0.0256, 0.0371, 0.0329, 0.0270, 0.0304, 0.0308, 0.0277], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 10:40:04,967 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.307e+02 2.025e+02 2.484e+02 3.273e+02 4.983e+02, threshold=4.968e+02, percent-clipped=1.0 2022-12-08 10:40:40,857 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=121950.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:40:55,451 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=121967.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 10:41:07,483 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4104, 2.2683, 2.7683, 1.9648, 1.8588, 2.4813, 1.5250, 2.4810], device='cuda:3'), covar=tensor([0.0965, 0.1314, 0.0856, 0.1593, 0.2241, 0.0880, 0.3372, 0.0954], device='cuda:3'), in_proj_covar=tensor([0.0084, 0.0100, 0.0095, 0.0098, 0.0115, 0.0090, 0.0118, 0.0093], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 10:41:17,349 INFO [train.py:873] (3/4) Epoch 17, batch 1000, loss[loss=0.1289, simple_loss=0.1554, pruned_loss=0.05121, over 14330.00 frames. ], tot_loss[loss=0.1089, simple_loss=0.1444, pruned_loss=0.03667, over 2009397.75 frames. ], batch size: 55, lr: 4.68e-03, grad_scale: 8.0 2022-12-08 10:41:22,455 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.37 vs. limit=2.0 2022-12-08 10:41:31,552 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.080e+02 2.007e+02 2.510e+02 3.013e+02 6.170e+02, threshold=5.019e+02, percent-clipped=2.0 2022-12-08 10:41:33,489 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122011.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:41:36,909 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=122015.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 10:41:59,497 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.92 vs. limit=5.0 2022-12-08 10:42:18,941 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122063.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:42:18,990 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7407, 1.5550, 1.5879, 1.7828, 1.7129, 1.1396, 1.5587, 1.5593], device='cuda:3'), covar=tensor([0.0559, 0.0546, 0.0783, 0.0766, 0.0506, 0.0787, 0.0657, 0.0615], device='cuda:3'), in_proj_covar=tensor([0.0035, 0.0033, 0.0037, 0.0032, 0.0033, 0.0046, 0.0035, 0.0037], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 10:42:30,902 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1155, 2.9282, 3.6403, 2.6243, 2.2576, 3.0935, 1.8582, 3.1563], device='cuda:3'), covar=tensor([0.0865, 0.1184, 0.0632, 0.1701, 0.2169, 0.0890, 0.3146, 0.1146], device='cuda:3'), in_proj_covar=tensor([0.0085, 0.0100, 0.0095, 0.0099, 0.0116, 0.0090, 0.0118, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 10:42:36,189 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0178, 2.0289, 2.1784, 1.5534, 1.5839, 2.0361, 1.3296, 2.0359], device='cuda:3'), covar=tensor([0.1013, 0.1936, 0.0974, 0.2383, 0.2854, 0.1172, 0.3565, 0.1058], device='cuda:3'), in_proj_covar=tensor([0.0085, 0.0100, 0.0095, 0.0099, 0.0116, 0.0090, 0.0118, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 10:42:44,899 INFO [train.py:873] (3/4) Epoch 17, batch 1100, loss[loss=0.1051, simple_loss=0.1419, pruned_loss=0.03412, over 14207.00 frames. ], tot_loss[loss=0.1086, simple_loss=0.1442, pruned_loss=0.03645, over 2009618.55 frames. ], batch size: 25, lr: 4.68e-03, grad_scale: 8.0 2022-12-08 10:42:51,706 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5883, 3.6709, 3.6632, 3.6261, 3.6227, 3.6767, 1.4958, 3.4548], device='cuda:3'), covar=tensor([0.0425, 0.0430, 0.0555, 0.0411, 0.0439, 0.0462, 0.3836, 0.0427], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0174, 0.0146, 0.0147, 0.0206, 0.0142, 0.0159, 0.0193], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 10:42:59,064 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.272e+02 2.098e+02 2.553e+02 3.146e+02 5.520e+02, threshold=5.106e+02, percent-clipped=3.0 2022-12-08 10:43:00,805 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=122111.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:43:05,783 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122117.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:43:59,190 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122178.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 10:44:04,539 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8945, 1.6913, 1.7790, 1.8370, 1.7386, 1.1802, 1.7536, 1.7023], device='cuda:3'), covar=tensor([0.0719, 0.0731, 0.0660, 0.0649, 0.0803, 0.1013, 0.0665, 0.0728], device='cuda:3'), in_proj_covar=tensor([0.0035, 0.0034, 0.0038, 0.0032, 0.0033, 0.0046, 0.0035, 0.0037], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 10:44:12,206 INFO [train.py:873] (3/4) Epoch 17, batch 1200, loss[loss=0.1013, simple_loss=0.1457, pruned_loss=0.02848, over 14399.00 frames. ], tot_loss[loss=0.1093, simple_loss=0.1447, pruned_loss=0.03698, over 1928077.55 frames. ], batch size: 41, lr: 4.67e-03, grad_scale: 8.0 2022-12-08 10:44:13,856 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122195.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:44:26,232 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.259e+02 1.994e+02 2.399e+02 3.114e+02 6.684e+02, threshold=4.798e+02, percent-clipped=5.0 2022-12-08 10:44:52,420 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2022-12-08 10:45:02,294 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0609, 2.1744, 1.9477, 2.1774, 1.7998, 2.0129, 2.1265, 2.0541], device='cuda:3'), covar=tensor([0.0969, 0.1120, 0.1147, 0.0856, 0.1557, 0.0840, 0.1069, 0.0972], device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0142, 0.0144, 0.0158, 0.0145, 0.0121, 0.0166, 0.0147], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 10:45:07,696 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122256.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:45:29,605 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3790, 4.3810, 4.7042, 4.0221, 4.4452, 4.7622, 1.7970, 4.2439], device='cuda:3'), covar=tensor([0.0311, 0.0321, 0.0320, 0.0582, 0.0315, 0.0198, 0.3089, 0.0302], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0174, 0.0145, 0.0146, 0.0205, 0.0141, 0.0158, 0.0193], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 10:45:33,139 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5671, 1.9173, 2.5139, 2.0937, 2.5534, 2.3949, 2.3434, 2.3338], device='cuda:3'), covar=tensor([0.0723, 0.2686, 0.0846, 0.1609, 0.0546, 0.1156, 0.0860, 0.1338], device='cuda:3'), in_proj_covar=tensor([0.0354, 0.0314, 0.0393, 0.0301, 0.0368, 0.0325, 0.0363, 0.0301], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 10:45:39,707 INFO [train.py:873] (3/4) Epoch 17, batch 1300, loss[loss=0.1036, simple_loss=0.1414, pruned_loss=0.03288, over 14280.00 frames. ], tot_loss[loss=0.1087, simple_loss=0.144, pruned_loss=0.03668, over 1871905.20 frames. ], batch size: 63, lr: 4.67e-03, grad_scale: 8.0 2022-12-08 10:45:52,044 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122306.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:45:54,349 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.486e+02 2.064e+02 2.522e+02 2.867e+02 6.773e+02, threshold=5.044e+02, percent-clipped=2.0 2022-12-08 10:46:02,979 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122319.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:46:17,641 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.0064, 1.3058, 1.3039, 1.0235, 0.8044, 1.0958, 0.9069, 1.2330], device='cuda:3'), covar=tensor([0.2730, 0.2569, 0.1218, 0.2588, 0.3380, 0.1391, 0.1922, 0.1268], device='cuda:3'), in_proj_covar=tensor([0.0085, 0.0100, 0.0094, 0.0098, 0.0115, 0.0089, 0.0118, 0.0093], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 10:46:53,887 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.59 vs. limit=2.0 2022-12-08 10:46:56,018 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122380.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:47:07,175 INFO [train.py:873] (3/4) Epoch 17, batch 1400, loss[loss=0.119, simple_loss=0.1553, pruned_loss=0.04131, over 14593.00 frames. ], tot_loss[loss=0.1093, simple_loss=0.1446, pruned_loss=0.03701, over 1909294.69 frames. ], batch size: 34, lr: 4.67e-03, grad_scale: 8.0 2022-12-08 10:47:14,220 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2022-12-08 10:47:21,305 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.812e+01 2.172e+02 2.630e+02 3.463e+02 6.435e+02, threshold=5.259e+02, percent-clipped=4.0 2022-12-08 10:47:22,280 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5176, 4.1606, 4.0525, 4.5091, 4.1910, 3.9380, 4.5226, 3.7896], device='cuda:3'), covar=tensor([0.0355, 0.0947, 0.0431, 0.0387, 0.0813, 0.1007, 0.0476, 0.0534], device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0272, 0.0198, 0.0192, 0.0184, 0.0155, 0.0284, 0.0169], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 10:47:49,201 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.58 vs. limit=5.0 2022-12-08 10:47:51,248 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.2455, 5.0765, 4.7985, 5.3168, 4.7833, 4.4999, 5.2934, 5.0062], device='cuda:3'), covar=tensor([0.0569, 0.0762, 0.0785, 0.0452, 0.0730, 0.0491, 0.0523, 0.0635], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0144, 0.0147, 0.0161, 0.0148, 0.0123, 0.0169, 0.0149], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 10:47:54,932 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8608, 2.1300, 2.2635, 2.3282, 2.0533, 2.3134, 2.1405, 1.4678], device='cuda:3'), covar=tensor([0.0875, 0.1080, 0.0643, 0.0623, 0.0981, 0.0587, 0.0967, 0.1916], device='cuda:3'), in_proj_covar=tensor([0.0137, 0.0088, 0.0069, 0.0074, 0.0099, 0.0087, 0.0100, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 10:48:13,772 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122468.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:48:17,834 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122473.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 10:48:35,289 INFO [train.py:873] (3/4) Epoch 17, batch 1500, loss[loss=0.09139, simple_loss=0.1376, pruned_loss=0.02257, over 14215.00 frames. ], tot_loss[loss=0.1087, simple_loss=0.1442, pruned_loss=0.03664, over 1957049.95 frames. ], batch size: 57, lr: 4.67e-03, grad_scale: 8.0 2022-12-08 10:48:49,379 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.412e+02 2.116e+02 2.482e+02 3.540e+02 1.353e+03, threshold=4.965e+02, percent-clipped=3.0 2022-12-08 10:49:06,558 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122529.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:49:26,284 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122551.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:50:03,169 INFO [train.py:873] (3/4) Epoch 17, batch 1600, loss[loss=0.1084, simple_loss=0.1252, pruned_loss=0.04579, over 3865.00 frames. ], tot_loss[loss=0.1081, simple_loss=0.1434, pruned_loss=0.03638, over 1889402.56 frames. ], batch size: 100, lr: 4.67e-03, grad_scale: 8.0 2022-12-08 10:50:14,941 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122606.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:50:17,232 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.386e+01 1.886e+02 2.427e+02 3.060e+02 1.117e+03, threshold=4.854e+02, percent-clipped=4.0 2022-12-08 10:50:25,412 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2812, 2.1687, 1.9781, 2.3803, 2.3240, 1.2588, 2.2869, 2.4134], device='cuda:3'), covar=tensor([0.1193, 0.0725, 0.0850, 0.1534, 0.0719, 0.0837, 0.0896, 0.0624], device='cuda:3'), in_proj_covar=tensor([0.0035, 0.0034, 0.0038, 0.0032, 0.0033, 0.0046, 0.0035, 0.0037], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 10:50:28,948 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5368, 3.0586, 2.3425, 3.5872, 3.4222, 3.5232, 2.9964, 2.3010], device='cuda:3'), covar=tensor([0.0908, 0.1714, 0.4193, 0.0689, 0.1095, 0.1092, 0.1808, 0.4493], device='cuda:3'), in_proj_covar=tensor([0.0282, 0.0292, 0.0259, 0.0286, 0.0322, 0.0301, 0.0254, 0.0243], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 10:50:56,744 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=122654.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:51:15,254 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122675.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:51:22,617 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.28 vs. limit=5.0 2022-12-08 10:51:22,947 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8481, 0.7579, 0.8059, 0.7974, 0.7828, 0.5139, 0.6606, 0.7776], device='cuda:3'), covar=tensor([0.0196, 0.0167, 0.0144, 0.0162, 0.0170, 0.0326, 0.0222, 0.0262], device='cuda:3'), in_proj_covar=tensor([0.0022, 0.0022, 0.0019, 0.0021, 0.0021, 0.0033, 0.0027, 0.0032], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 10:51:30,441 INFO [train.py:873] (3/4) Epoch 17, batch 1700, loss[loss=0.1311, simple_loss=0.1537, pruned_loss=0.05424, over 8648.00 frames. ], tot_loss[loss=0.1085, simple_loss=0.1437, pruned_loss=0.0366, over 1914051.12 frames. ], batch size: 100, lr: 4.66e-03, grad_scale: 4.0 2022-12-08 10:51:32,140 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.62 vs. limit=5.0 2022-12-08 10:51:45,734 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.060e+02 2.028e+02 2.462e+02 2.959e+02 4.889e+02, threshold=4.923e+02, percent-clipped=1.0 2022-12-08 10:51:50,138 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9490, 3.2104, 3.0247, 3.0886, 2.4832, 3.2604, 3.1025, 1.7796], device='cuda:3'), covar=tensor([0.1253, 0.0669, 0.0824, 0.0827, 0.0995, 0.0504, 0.0925, 0.2033], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0089, 0.0069, 0.0075, 0.0100, 0.0088, 0.0101, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 10:52:39,990 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122773.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 10:52:41,273 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2056, 2.1276, 2.0253, 2.3504, 2.4591, 1.9004, 1.6043, 1.7256], device='cuda:3'), covar=tensor([0.0276, 0.0419, 0.0503, 0.0191, 0.0199, 0.0468, 0.0477, 0.0501], device='cuda:3'), in_proj_covar=tensor([0.0021, 0.0021, 0.0019, 0.0020, 0.0020, 0.0032, 0.0027, 0.0031], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 10:52:57,842 INFO [train.py:873] (3/4) Epoch 17, batch 1800, loss[loss=0.09291, simple_loss=0.1351, pruned_loss=0.02535, over 14293.00 frames. ], tot_loss[loss=0.1088, simple_loss=0.1443, pruned_loss=0.03663, over 1974048.75 frames. ], batch size: 39, lr: 4.66e-03, grad_scale: 4.0 2022-12-08 10:53:12,811 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.140e+02 2.194e+02 2.593e+02 3.400e+02 7.475e+02, threshold=5.185e+02, percent-clipped=4.0 2022-12-08 10:53:22,445 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=122821.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:53:25,008 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122824.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:53:48,753 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122851.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:54:01,026 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8310, 1.9440, 2.1026, 1.4612, 1.5307, 1.8526, 1.2220, 1.9382], device='cuda:3'), covar=tensor([0.1634, 0.1891, 0.0947, 0.2663, 0.3080, 0.1102, 0.2887, 0.1046], device='cuda:3'), in_proj_covar=tensor([0.0086, 0.0102, 0.0096, 0.0101, 0.0117, 0.0091, 0.0119, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 10:54:07,259 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122872.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:54:16,161 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.8898, 5.3944, 5.2258, 5.8321, 5.4671, 4.8830, 5.7767, 4.6720], device='cuda:3'), covar=tensor([0.0290, 0.0815, 0.0365, 0.0339, 0.0724, 0.0350, 0.0476, 0.0533], device='cuda:3'), in_proj_covar=tensor([0.0181, 0.0278, 0.0202, 0.0195, 0.0187, 0.0157, 0.0290, 0.0172], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 10:54:25,553 INFO [train.py:873] (3/4) Epoch 17, batch 1900, loss[loss=0.1008, simple_loss=0.1356, pruned_loss=0.03301, over 14257.00 frames. ], tot_loss[loss=0.1085, simple_loss=0.1439, pruned_loss=0.03655, over 1970197.61 frames. ], batch size: 57, lr: 4.66e-03, grad_scale: 4.0 2022-12-08 10:54:30,922 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=122899.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:54:32,240 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2022-12-08 10:54:40,370 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.411e+02 2.015e+02 2.576e+02 3.102e+02 5.463e+02, threshold=5.151e+02, percent-clipped=2.0 2022-12-08 10:55:00,483 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122933.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:55:06,347 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7631, 1.8304, 1.9839, 1.3724, 1.4685, 1.7842, 1.2286, 1.8289], device='cuda:3'), covar=tensor([0.1392, 0.1764, 0.1075, 0.2917, 0.3123, 0.1138, 0.3227, 0.1112], device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0102, 0.0097, 0.0102, 0.0118, 0.0092, 0.0119, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 10:55:20,224 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2022-12-08 10:55:37,074 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122975.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:55:52,711 INFO [train.py:873] (3/4) Epoch 17, batch 2000, loss[loss=0.1212, simple_loss=0.1567, pruned_loss=0.04288, over 11174.00 frames. ], tot_loss[loss=0.1089, simple_loss=0.1441, pruned_loss=0.03683, over 1923591.02 frames. ], batch size: 100, lr: 4.66e-03, grad_scale: 8.0 2022-12-08 10:56:08,074 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.251e+02 2.103e+02 2.584e+02 3.281e+02 5.069e+02, threshold=5.168e+02, percent-clipped=0.0 2022-12-08 10:56:19,439 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=123023.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:57:20,549 INFO [train.py:873] (3/4) Epoch 17, batch 2100, loss[loss=0.08313, simple_loss=0.1297, pruned_loss=0.01829, over 14391.00 frames. ], tot_loss[loss=0.1088, simple_loss=0.1441, pruned_loss=0.03681, over 1944010.29 frames. ], batch size: 41, lr: 4.66e-03, grad_scale: 4.0 2022-12-08 10:57:32,669 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.7589, 4.5210, 4.3525, 4.7507, 4.3377, 4.1625, 4.8289, 4.5362], device='cuda:3'), covar=tensor([0.0550, 0.0807, 0.0793, 0.0530, 0.0811, 0.0642, 0.0506, 0.0681], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0143, 0.0146, 0.0161, 0.0148, 0.0123, 0.0169, 0.0149], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 10:57:36,150 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.184e+02 2.207e+02 2.592e+02 3.533e+02 7.488e+02, threshold=5.184e+02, percent-clipped=8.0 2022-12-08 10:57:47,440 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=123124.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:58:13,625 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9455, 1.8565, 2.0472, 2.0316, 2.0254, 1.7363, 1.7462, 1.3041], device='cuda:3'), covar=tensor([0.0281, 0.0502, 0.0263, 0.0297, 0.0225, 0.0334, 0.0273, 0.0477], device='cuda:3'), in_proj_covar=tensor([0.0022, 0.0022, 0.0020, 0.0021, 0.0021, 0.0033, 0.0027, 0.0032], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-08 10:58:15,836 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=7.49 vs. limit=5.0 2022-12-08 10:58:29,174 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=123172.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 10:58:46,985 INFO [train.py:873] (3/4) Epoch 17, batch 2200, loss[loss=0.1162, simple_loss=0.1466, pruned_loss=0.04292, over 11133.00 frames. ], tot_loss[loss=0.1093, simple_loss=0.1443, pruned_loss=0.03717, over 1935542.91 frames. ], batch size: 100, lr: 4.65e-03, grad_scale: 4.0 2022-12-08 10:59:03,099 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.154e+02 2.073e+02 2.495e+02 2.880e+02 5.648e+02, threshold=4.991e+02, percent-clipped=1.0 2022-12-08 10:59:15,494 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.96 vs. limit=5.0 2022-12-08 10:59:17,876 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=123228.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:00:13,630 INFO [train.py:873] (3/4) Epoch 17, batch 2300, loss[loss=0.1178, simple_loss=0.1522, pruned_loss=0.04172, over 14287.00 frames. ], tot_loss[loss=0.1077, simple_loss=0.1433, pruned_loss=0.03604, over 1966687.42 frames. ], batch size: 44, lr: 4.65e-03, grad_scale: 4.0 2022-12-08 11:00:29,910 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.110e+02 2.209e+02 2.627e+02 3.184e+02 5.736e+02, threshold=5.255e+02, percent-clipped=2.0 2022-12-08 11:00:32,599 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=123314.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:00:46,381 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2022-12-08 11:01:05,515 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.61 vs. limit=2.0 2022-12-08 11:01:26,438 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=123375.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:01:41,812 INFO [train.py:873] (3/4) Epoch 17, batch 2400, loss[loss=0.1154, simple_loss=0.1499, pruned_loss=0.04044, over 14291.00 frames. ], tot_loss[loss=0.1086, simple_loss=0.1439, pruned_loss=0.03665, over 1937007.85 frames. ], batch size: 76, lr: 4.65e-03, grad_scale: 8.0 2022-12-08 11:01:58,032 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.281e+02 2.088e+02 2.563e+02 3.123e+02 4.912e+02, threshold=5.125e+02, percent-clipped=0.0 2022-12-08 11:02:12,667 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=123428.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:03:01,444 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0 2022-12-08 11:03:07,273 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=123489.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:03:10,658 INFO [train.py:873] (3/4) Epoch 17, batch 2500, loss[loss=0.1033, simple_loss=0.1415, pruned_loss=0.03253, over 14178.00 frames. ], tot_loss[loss=0.1075, simple_loss=0.1434, pruned_loss=0.0358, over 1975290.39 frames. ], batch size: 99, lr: 4.65e-03, grad_scale: 8.0 2022-12-08 11:03:26,844 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.447e+02 2.132e+02 2.678e+02 3.392e+02 6.335e+02, threshold=5.356e+02, percent-clipped=3.0 2022-12-08 11:03:42,083 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=123528.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:04:13,718 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.51 vs. limit=2.0 2022-12-08 11:04:23,968 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=123576.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:04:39,204 INFO [train.py:873] (3/4) Epoch 17, batch 2600, loss[loss=0.09062, simple_loss=0.1356, pruned_loss=0.02282, over 14311.00 frames. ], tot_loss[loss=0.1089, simple_loss=0.144, pruned_loss=0.03685, over 1913133.30 frames. ], batch size: 28, lr: 4.65e-03, grad_scale: 8.0 2022-12-08 11:04:55,081 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.145e+02 1.985e+02 2.407e+02 3.072e+02 5.323e+02, threshold=4.814e+02, percent-clipped=0.0 2022-12-08 11:05:46,077 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=123670.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:06:06,728 INFO [train.py:873] (3/4) Epoch 17, batch 2700, loss[loss=0.08903, simple_loss=0.131, pruned_loss=0.02354, over 14653.00 frames. ], tot_loss[loss=0.1083, simple_loss=0.1439, pruned_loss=0.03638, over 1963246.95 frames. ], batch size: 33, lr: 4.65e-03, grad_scale: 8.0 2022-12-08 11:06:22,182 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.075e+02 2.020e+02 2.571e+02 3.169e+02 5.832e+02, threshold=5.143e+02, percent-clipped=4.0 2022-12-08 11:06:26,077 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7222, 1.6383, 1.8727, 1.6885, 1.5736, 1.5543, 1.2763, 1.0716], device='cuda:3'), covar=tensor([0.0207, 0.0327, 0.0183, 0.0181, 0.0220, 0.0374, 0.0291, 0.0478], device='cuda:3'), in_proj_covar=tensor([0.0022, 0.0022, 0.0019, 0.0021, 0.0021, 0.0033, 0.0027, 0.0032], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-08 11:07:25,876 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=123784.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:07:33,393 INFO [train.py:873] (3/4) Epoch 17, batch 2800, loss[loss=0.1193, simple_loss=0.1507, pruned_loss=0.04401, over 14198.00 frames. ], tot_loss[loss=0.1087, simple_loss=0.1444, pruned_loss=0.03654, over 2018070.50 frames. ], batch size: 80, lr: 4.64e-03, grad_scale: 8.0 2022-12-08 11:07:49,825 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.157e+02 2.052e+02 2.657e+02 3.438e+02 8.382e+02, threshold=5.313e+02, percent-clipped=4.0 2022-12-08 11:08:02,343 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.10 vs. limit=2.0 2022-12-08 11:08:43,005 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=123872.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:09:01,389 INFO [train.py:873] (3/4) Epoch 17, batch 2900, loss[loss=0.1002, simple_loss=0.1117, pruned_loss=0.04436, over 2593.00 frames. ], tot_loss[loss=0.1079, simple_loss=0.1439, pruned_loss=0.03598, over 2002856.46 frames. ], batch size: 100, lr: 4.64e-03, grad_scale: 8.0 2022-12-08 11:09:13,196 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2022-12-08 11:09:16,751 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.276e+02 2.204e+02 2.576e+02 3.173e+02 6.433e+02, threshold=5.152e+02, percent-clipped=4.0 2022-12-08 11:09:25,387 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2022-12-08 11:09:36,545 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=123933.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:10:08,136 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=123970.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:10:17,951 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5330, 2.6269, 4.3759, 4.5846, 4.4077, 2.5798, 4.6132, 3.4969], device='cuda:3'), covar=tensor([0.0333, 0.1130, 0.0680, 0.0333, 0.0432, 0.1800, 0.0386, 0.0830], device='cuda:3'), in_proj_covar=tensor([0.0295, 0.0258, 0.0375, 0.0332, 0.0271, 0.0305, 0.0313, 0.0280], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 11:10:21,074 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5282, 4.3129, 4.0364, 4.2002, 4.3894, 4.4393, 4.5212, 4.5367], device='cuda:3'), covar=tensor([0.0811, 0.0481, 0.1846, 0.2393, 0.0595, 0.0772, 0.0754, 0.0682], device='cuda:3'), in_proj_covar=tensor([0.0391, 0.0274, 0.0453, 0.0568, 0.0349, 0.0449, 0.0393, 0.0399], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 11:10:27,940 INFO [train.py:873] (3/4) Epoch 17, batch 3000, loss[loss=0.09, simple_loss=0.1218, pruned_loss=0.02908, over 4989.00 frames. ], tot_loss[loss=0.1082, simple_loss=0.1441, pruned_loss=0.03614, over 2014076.86 frames. ], batch size: 100, lr: 4.64e-03, grad_scale: 8.0 2022-12-08 11:10:27,940 INFO [train.py:896] (3/4) Computing validation loss 2022-12-08 11:10:36,502 INFO [train.py:905] (3/4) Epoch 17, validation: loss=0.1392, simple_loss=0.1759, pruned_loss=0.05127, over 857387.00 frames. 2022-12-08 11:10:36,503 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-08 11:10:52,778 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.289e+02 2.049e+02 2.779e+02 3.287e+02 7.415e+02, threshold=5.559e+02, percent-clipped=4.0 2022-12-08 11:10:56,819 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3351, 2.2096, 3.1577, 2.4310, 3.2188, 3.1130, 2.9826, 2.6312], device='cuda:3'), covar=tensor([0.1180, 0.3281, 0.1129, 0.2134, 0.0870, 0.1249, 0.1448, 0.1953], device='cuda:3'), in_proj_covar=tensor([0.0353, 0.0312, 0.0392, 0.0300, 0.0367, 0.0321, 0.0361, 0.0296], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 11:10:58,402 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=124018.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:11:14,484 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124036.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:11:37,601 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124062.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:11:57,232 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124084.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:12:04,347 INFO [train.py:873] (3/4) Epoch 17, batch 3100, loss[loss=0.08444, simple_loss=0.125, pruned_loss=0.02195, over 13997.00 frames. ], tot_loss[loss=0.1083, simple_loss=0.1439, pruned_loss=0.0364, over 1997936.71 frames. ], batch size: 19, lr: 4.64e-03, grad_scale: 4.0 2022-12-08 11:12:08,239 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124097.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:12:20,720 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.093e+02 2.198e+02 2.691e+02 3.318e+02 1.640e+03, threshold=5.381e+02, percent-clipped=4.0 2022-12-08 11:12:24,141 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1635, 1.1236, 0.9034, 1.1212, 1.1619, 0.7828, 0.9929, 1.1117], device='cuda:3'), covar=tensor([0.0620, 0.0571, 0.0505, 0.0442, 0.0369, 0.0460, 0.0627, 0.0740], device='cuda:3'), in_proj_covar=tensor([0.0035, 0.0034, 0.0038, 0.0031, 0.0033, 0.0046, 0.0035, 0.0037], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 11:12:31,256 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124123.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:12:38,800 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=124132.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:13:15,549 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4010, 1.4078, 3.4952, 1.5534, 3.3379, 3.5560, 2.5103, 3.7854], device='cuda:3'), covar=tensor([0.0292, 0.3360, 0.0434, 0.2446, 0.0804, 0.0428, 0.0925, 0.0221], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0155, 0.0159, 0.0168, 0.0165, 0.0179, 0.0133, 0.0151], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 11:13:31,918 INFO [train.py:873] (3/4) Epoch 17, batch 3200, loss[loss=0.1016, simple_loss=0.138, pruned_loss=0.03266, over 10307.00 frames. ], tot_loss[loss=0.1075, simple_loss=0.1435, pruned_loss=0.03572, over 2064258.46 frames. ], batch size: 100, lr: 4.64e-03, grad_scale: 8.0 2022-12-08 11:13:42,598 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.67 vs. limit=2.0 2022-12-08 11:13:48,754 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.337e+02 1.975e+02 2.521e+02 3.067e+02 5.500e+02, threshold=5.042e+02, percent-clipped=1.0 2022-12-08 11:14:02,392 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124228.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:14:40,389 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124272.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:14:58,461 INFO [train.py:873] (3/4) Epoch 17, batch 3300, loss[loss=0.1029, simple_loss=0.1399, pruned_loss=0.03295, over 14214.00 frames. ], tot_loss[loss=0.1085, simple_loss=0.1438, pruned_loss=0.03655, over 1982822.27 frames. ], batch size: 94, lr: 4.63e-03, grad_scale: 8.0 2022-12-08 11:15:14,505 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.244e+02 2.029e+02 2.735e+02 3.344e+02 8.067e+02, threshold=5.469e+02, percent-clipped=3.0 2022-12-08 11:15:33,700 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124333.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:15:47,094 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1364, 2.1290, 2.3599, 1.5912, 1.6864, 2.1638, 1.3629, 2.2440], device='cuda:3'), covar=tensor([0.1182, 0.1582, 0.0790, 0.2564, 0.2687, 0.0940, 0.3318, 0.0991], device='cuda:3'), in_proj_covar=tensor([0.0086, 0.0101, 0.0095, 0.0100, 0.0116, 0.0090, 0.0118, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 11:15:47,119 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124348.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 11:16:03,929 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 2022-12-08 11:16:17,668 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8651, 1.8935, 1.6509, 1.9417, 1.6823, 1.8573, 1.8184, 1.7171], device='cuda:3'), covar=tensor([0.0959, 0.0876, 0.1720, 0.0679, 0.1082, 0.0615, 0.1381, 0.1112], device='cuda:3'), in_proj_covar=tensor([0.0284, 0.0291, 0.0260, 0.0288, 0.0325, 0.0303, 0.0258, 0.0244], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 11:16:25,320 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124392.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:16:26,084 INFO [train.py:873] (3/4) Epoch 17, batch 3400, loss[loss=0.119, simple_loss=0.1555, pruned_loss=0.04124, over 14247.00 frames. ], tot_loss[loss=0.1074, simple_loss=0.1435, pruned_loss=0.03567, over 2026274.09 frames. ], batch size: 80, lr: 4.63e-03, grad_scale: 4.0 2022-12-08 11:16:40,667 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124409.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 11:16:43,979 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.203e+02 2.119e+02 2.705e+02 3.387e+02 6.415e+02, threshold=5.409e+02, percent-clipped=2.0 2022-12-08 11:16:48,615 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124418.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:17:00,909 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124432.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:17:27,960 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4175, 1.4104, 1.4030, 1.6021, 1.6110, 1.0626, 1.3113, 1.3673], device='cuda:3'), covar=tensor([0.0634, 0.0822, 0.0678, 0.0633, 0.0402, 0.0805, 0.0833, 0.0681], device='cuda:3'), in_proj_covar=tensor([0.0035, 0.0034, 0.0038, 0.0032, 0.0033, 0.0046, 0.0035, 0.0037], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 11:17:28,948 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.11 vs. limit=2.0 2022-12-08 11:17:54,883 INFO [train.py:873] (3/4) Epoch 17, batch 3500, loss[loss=0.1238, simple_loss=0.1542, pruned_loss=0.04671, over 14649.00 frames. ], tot_loss[loss=0.107, simple_loss=0.1427, pruned_loss=0.03564, over 2032821.16 frames. ], batch size: 23, lr: 4.63e-03, grad_scale: 4.0 2022-12-08 11:17:55,050 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124493.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:17:55,072 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8885, 2.8358, 2.1818, 2.9442, 2.7637, 2.8322, 2.5528, 2.2706], device='cuda:3'), covar=tensor([0.1016, 0.1310, 0.2658, 0.1062, 0.1164, 0.1213, 0.1415, 0.2633], device='cuda:3'), in_proj_covar=tensor([0.0282, 0.0290, 0.0257, 0.0287, 0.0323, 0.0301, 0.0256, 0.0242], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 11:18:12,216 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.247e+02 2.179e+02 2.842e+02 3.919e+02 2.207e+03, threshold=5.685e+02, percent-clipped=7.0 2022-12-08 11:18:17,186 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5231, 2.2783, 2.8876, 1.8643, 2.0418, 2.5510, 1.4485, 2.6719], device='cuda:3'), covar=tensor([0.1016, 0.1210, 0.0658, 0.2077, 0.2032, 0.0724, 0.3277, 0.0813], device='cuda:3'), in_proj_covar=tensor([0.0085, 0.0101, 0.0095, 0.0100, 0.0115, 0.0090, 0.0118, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 11:18:25,902 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124528.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:19:03,686 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0379, 3.2708, 3.1495, 3.2779, 2.5295, 3.3784, 3.1328, 1.7700], device='cuda:3'), covar=tensor([0.1312, 0.0771, 0.0850, 0.0526, 0.0980, 0.0587, 0.0896, 0.2035], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0090, 0.0071, 0.0076, 0.0101, 0.0089, 0.0102, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 11:19:08,274 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=124576.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:19:22,686 INFO [train.py:873] (3/4) Epoch 17, batch 3600, loss[loss=0.116, simple_loss=0.1471, pruned_loss=0.04241, over 14217.00 frames. ], tot_loss[loss=0.1061, simple_loss=0.1419, pruned_loss=0.03515, over 1947041.91 frames. ], batch size: 94, lr: 4.63e-03, grad_scale: 8.0 2022-12-08 11:19:34,072 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5542, 3.7373, 3.3912, 3.8433, 2.7704, 3.8080, 3.6184, 2.0552], device='cuda:3'), covar=tensor([0.1288, 0.0605, 0.0939, 0.0509, 0.0857, 0.0343, 0.0768, 0.1901], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0090, 0.0071, 0.0076, 0.0101, 0.0090, 0.0102, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 11:19:40,803 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.418e+02 1.900e+02 2.307e+02 2.789e+02 7.825e+02, threshold=4.613e+02, percent-clipped=1.0 2022-12-08 11:19:53,720 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124628.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:20:02,893 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124638.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:20:24,745 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124663.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:20:50,175 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124692.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:20:50,939 INFO [train.py:873] (3/4) Epoch 17, batch 3700, loss[loss=0.1185, simple_loss=0.1422, pruned_loss=0.04743, over 3903.00 frames. ], tot_loss[loss=0.1064, simple_loss=0.1422, pruned_loss=0.03532, over 1973293.75 frames. ], batch size: 100, lr: 4.63e-03, grad_scale: 8.0 2022-12-08 11:20:52,232 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2022-12-08 11:20:56,597 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124699.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:21:00,482 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124704.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 11:21:07,859 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.074e+02 2.026e+02 2.516e+02 3.153e+02 6.578e+02, threshold=5.031e+02, percent-clipped=7.0 2022-12-08 11:21:12,576 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124718.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:21:18,151 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124724.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:21:31,752 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=124740.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:21:49,542 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2022-12-08 11:21:54,353 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=124766.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:22:13,601 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124788.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:22:17,561 INFO [train.py:873] (3/4) Epoch 17, batch 3800, loss[loss=0.108, simple_loss=0.1442, pruned_loss=0.03587, over 14222.00 frames. ], tot_loss[loss=0.1069, simple_loss=0.1425, pruned_loss=0.0356, over 1975756.41 frames. ], batch size: 94, lr: 4.62e-03, grad_scale: 8.0 2022-12-08 11:22:35,562 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.197e+02 2.150e+02 2.611e+02 3.247e+02 6.025e+02, threshold=5.221e+02, percent-clipped=4.0 2022-12-08 11:22:41,971 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124820.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:23:10,629 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.6977, 2.1701, 2.2561, 2.4332, 2.2273, 1.3582, 2.4285, 2.4907], device='cuda:3'), covar=tensor([0.0609, 0.0802, 0.1134, 0.0722, 0.0835, 0.0967, 0.0639, 0.1330], device='cuda:3'), in_proj_covar=tensor([0.0035, 0.0034, 0.0039, 0.0032, 0.0034, 0.0046, 0.0035, 0.0037], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 11:23:17,705 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.3610, 1.7054, 1.8794, 1.8665, 1.7576, 1.8530, 1.5403, 1.2316], device='cuda:3'), covar=tensor([0.1228, 0.1186, 0.0667, 0.0744, 0.1224, 0.1025, 0.2122, 0.1938], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0090, 0.0071, 0.0076, 0.0100, 0.0090, 0.0102, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 11:23:35,470 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124881.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:23:45,832 INFO [train.py:873] (3/4) Epoch 17, batch 3900, loss[loss=0.1121, simple_loss=0.1486, pruned_loss=0.03781, over 14254.00 frames. ], tot_loss[loss=0.1062, simple_loss=0.1417, pruned_loss=0.03532, over 1965481.33 frames. ], batch size: 57, lr: 4.62e-03, grad_scale: 4.0 2022-12-08 11:23:48,975 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6424, 1.6739, 1.7539, 1.5463, 1.4359, 1.3715, 1.2940, 1.0963], device='cuda:3'), covar=tensor([0.0181, 0.0231, 0.0195, 0.0220, 0.0238, 0.0341, 0.0242, 0.0398], device='cuda:3'), in_proj_covar=tensor([0.0022, 0.0022, 0.0020, 0.0021, 0.0021, 0.0034, 0.0028, 0.0032], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-08 11:24:04,347 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.083e+02 2.040e+02 2.475e+02 3.105e+02 8.016e+02, threshold=4.950e+02, percent-clipped=1.0 2022-12-08 11:24:17,300 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124928.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:24:32,436 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2022-12-08 11:24:59,257 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=124976.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:25:13,842 INFO [train.py:873] (3/4) Epoch 17, batch 4000, loss[loss=0.1248, simple_loss=0.1285, pruned_loss=0.0605, over 1261.00 frames. ], tot_loss[loss=0.1061, simple_loss=0.1416, pruned_loss=0.03524, over 1878605.84 frames. ], batch size: 100, lr: 4.62e-03, grad_scale: 8.0 2022-12-08 11:25:14,807 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124994.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:25:27,972 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125004.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 11:25:36,307 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.370e+02 2.040e+02 2.496e+02 2.950e+02 4.930e+02, threshold=4.992e+02, percent-clipped=0.0 2022-12-08 11:25:40,973 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125019.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:25:47,746 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1790, 2.0376, 2.1988, 2.3182, 1.9328, 1.8992, 2.1164, 2.1689], device='cuda:3'), covar=tensor([0.0325, 0.0690, 0.0354, 0.0388, 0.0567, 0.0980, 0.0539, 0.0447], device='cuda:3'), in_proj_covar=tensor([0.0293, 0.0257, 0.0374, 0.0330, 0.0270, 0.0307, 0.0313, 0.0278], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 11:26:09,621 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=125052.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 11:26:22,097 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.59 vs. limit=2.0 2022-12-08 11:26:34,990 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8559, 3.3757, 2.6609, 4.0653, 3.8708, 3.9252, 3.3991, 2.8400], device='cuda:3'), covar=tensor([0.0765, 0.1356, 0.3206, 0.0515, 0.0764, 0.1074, 0.1297, 0.2758], device='cuda:3'), in_proj_covar=tensor([0.0282, 0.0288, 0.0257, 0.0284, 0.0322, 0.0299, 0.0252, 0.0240], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 11:26:40,512 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125088.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:26:44,719 INFO [train.py:873] (3/4) Epoch 17, batch 4100, loss[loss=0.107, simple_loss=0.1452, pruned_loss=0.03436, over 14029.00 frames. ], tot_loss[loss=0.1058, simple_loss=0.1418, pruned_loss=0.03487, over 1956861.27 frames. ], batch size: 22, lr: 4.62e-03, grad_scale: 4.0 2022-12-08 11:27:01,487 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.55 vs. limit=5.0 2022-12-08 11:27:03,766 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.387e+02 2.146e+02 2.604e+02 3.171e+02 6.787e+02, threshold=5.207e+02, percent-clipped=4.0 2022-12-08 11:27:15,880 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125128.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:27:22,528 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=125136.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:27:32,204 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1061, 1.0497, 1.2304, 0.9986, 0.9155, 0.8037, 0.8872, 0.9090], device='cuda:3'), covar=tensor([0.0237, 0.0209, 0.0181, 0.0235, 0.0257, 0.0410, 0.0268, 0.0383], device='cuda:3'), in_proj_covar=tensor([0.0022, 0.0022, 0.0020, 0.0021, 0.0020, 0.0033, 0.0027, 0.0032], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 11:27:57,266 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125176.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:27:59,796 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.02 vs. limit=2.0 2022-12-08 11:28:08,775 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125189.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:28:11,907 INFO [train.py:873] (3/4) Epoch 17, batch 4200, loss[loss=0.103, simple_loss=0.1262, pruned_loss=0.03989, over 2648.00 frames. ], tot_loss[loss=0.1072, simple_loss=0.1427, pruned_loss=0.03582, over 1875908.21 frames. ], batch size: 100, lr: 4.62e-03, grad_scale: 4.0 2022-12-08 11:28:12,984 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8693, 2.2284, 2.8014, 2.9158, 2.8295, 2.1456, 2.8385, 2.4092], device='cuda:3'), covar=tensor([0.0453, 0.1087, 0.0694, 0.0460, 0.0507, 0.1365, 0.0450, 0.0806], device='cuda:3'), in_proj_covar=tensor([0.0290, 0.0254, 0.0371, 0.0326, 0.0267, 0.0304, 0.0308, 0.0274], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 11:28:31,908 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.327e+02 2.000e+02 2.432e+02 3.049e+02 5.744e+02, threshold=4.863e+02, percent-clipped=3.0 2022-12-08 11:29:09,428 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1125, 2.1330, 2.1633, 2.2526, 2.1452, 1.2809, 1.9663, 2.1663], device='cuda:3'), covar=tensor([0.0893, 0.0640, 0.0660, 0.0772, 0.0827, 0.0836, 0.0934, 0.1077], device='cuda:3'), in_proj_covar=tensor([0.0035, 0.0034, 0.0039, 0.0032, 0.0034, 0.0047, 0.0035, 0.0038], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 11:29:40,188 INFO [train.py:873] (3/4) Epoch 17, batch 4300, loss[loss=0.1167, simple_loss=0.1322, pruned_loss=0.05055, over 2661.00 frames. ], tot_loss[loss=0.1082, simple_loss=0.1438, pruned_loss=0.03632, over 1910492.62 frames. ], batch size: 100, lr: 4.62e-03, grad_scale: 4.0 2022-12-08 11:29:41,227 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125294.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:29:50,548 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8067, 0.8354, 0.7212, 0.8734, 0.8544, 0.3541, 0.7488, 0.8611], device='cuda:3'), covar=tensor([0.0438, 0.0488, 0.0496, 0.0494, 0.0392, 0.0452, 0.0983, 0.0851], device='cuda:3'), in_proj_covar=tensor([0.0035, 0.0035, 0.0039, 0.0032, 0.0034, 0.0047, 0.0035, 0.0038], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 11:29:58,839 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.107e+02 2.137e+02 2.626e+02 3.159e+02 7.655e+02, threshold=5.251e+02, percent-clipped=3.0 2022-12-08 11:30:02,807 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125319.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:30:03,617 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.0567, 4.6818, 4.4884, 5.0387, 4.6602, 4.3960, 5.0369, 4.3292], device='cuda:3'), covar=tensor([0.0351, 0.0939, 0.0450, 0.0399, 0.0822, 0.0613, 0.0569, 0.0533], device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0276, 0.0201, 0.0193, 0.0184, 0.0157, 0.0288, 0.0169], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 11:30:23,020 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=125342.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:30:34,871 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125356.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:30:44,401 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=125367.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:30:50,309 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.27 vs. limit=5.0 2022-12-08 11:30:52,475 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125376.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:31:07,130 INFO [train.py:873] (3/4) Epoch 17, batch 4400, loss[loss=0.1535, simple_loss=0.1453, pruned_loss=0.08088, over 1280.00 frames. ], tot_loss[loss=0.1077, simple_loss=0.1437, pruned_loss=0.0359, over 1948354.89 frames. ], batch size: 100, lr: 4.61e-03, grad_scale: 8.0 2022-12-08 11:31:26,864 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.784e+01 2.190e+02 2.587e+02 3.207e+02 7.157e+02, threshold=5.175e+02, percent-clipped=2.0 2022-12-08 11:31:28,836 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125417.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 11:31:32,402 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125421.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:31:38,927 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2022-12-08 11:31:46,339 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125437.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:32:07,440 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.81 vs. limit=5.0 2022-12-08 11:32:20,293 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125476.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:32:22,795 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9152, 2.7030, 2.4606, 2.6564, 2.8350, 2.8328, 2.8575, 2.8726], device='cuda:3'), covar=tensor([0.0982, 0.0724, 0.2417, 0.2617, 0.0972, 0.1086, 0.1357, 0.0919], device='cuda:3'), in_proj_covar=tensor([0.0394, 0.0274, 0.0457, 0.0569, 0.0352, 0.0449, 0.0394, 0.0402], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 11:32:25,269 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125482.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:32:26,966 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125484.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:32:35,102 INFO [train.py:873] (3/4) Epoch 17, batch 4500, loss[loss=0.08787, simple_loss=0.1377, pruned_loss=0.01901, over 14294.00 frames. ], tot_loss[loss=0.108, simple_loss=0.1437, pruned_loss=0.03614, over 1932898.58 frames. ], batch size: 31, lr: 4.61e-03, grad_scale: 4.0 2022-12-08 11:32:54,649 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.256e+02 2.002e+02 2.492e+02 3.086e+02 6.444e+02, threshold=4.984e+02, percent-clipped=2.0 2022-12-08 11:33:01,068 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.93 vs. limit=5.0 2022-12-08 11:33:01,488 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=125524.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:34:01,916 INFO [train.py:873] (3/4) Epoch 17, batch 4600, loss[loss=0.1008, simple_loss=0.1439, pruned_loss=0.02888, over 14278.00 frames. ], tot_loss[loss=0.1081, simple_loss=0.1439, pruned_loss=0.03615, over 1988899.07 frames. ], batch size: 76, lr: 4.61e-03, grad_scale: 4.0 2022-12-08 11:34:11,408 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125603.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:34:22,483 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.136e+02 2.137e+02 2.692e+02 3.129e+02 4.972e+02, threshold=5.384e+02, percent-clipped=0.0 2022-12-08 11:35:05,309 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125664.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:35:30,461 INFO [train.py:873] (3/4) Epoch 17, batch 4700, loss[loss=0.1153, simple_loss=0.1472, pruned_loss=0.04167, over 9501.00 frames. ], tot_loss[loss=0.1075, simple_loss=0.1432, pruned_loss=0.03591, over 1985478.83 frames. ], batch size: 100, lr: 4.61e-03, grad_scale: 4.0 2022-12-08 11:35:47,159 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125712.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 11:35:48,944 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1185, 2.4457, 2.2749, 2.2971, 2.2723, 1.3302, 1.9356, 2.4831], device='cuda:3'), covar=tensor([0.1357, 0.0552, 0.0839, 0.1501, 0.1360, 0.0945, 0.1832, 0.0925], device='cuda:3'), in_proj_covar=tensor([0.0035, 0.0034, 0.0039, 0.0032, 0.0034, 0.0047, 0.0035, 0.0038], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 11:35:50,738 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.171e+02 2.065e+02 2.672e+02 3.737e+02 1.296e+03, threshold=5.344e+02, percent-clipped=9.0 2022-12-08 11:35:57,942 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125725.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:36:04,513 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125732.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:36:44,477 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125777.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:36:50,792 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125784.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:36:52,581 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125786.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:36:58,260 INFO [train.py:873] (3/4) Epoch 17, batch 4800, loss[loss=0.09231, simple_loss=0.1348, pruned_loss=0.0249, over 14643.00 frames. ], tot_loss[loss=0.1072, simple_loss=0.1427, pruned_loss=0.03591, over 1981543.94 frames. ], batch size: 23, lr: 4.61e-03, grad_scale: 8.0 2022-12-08 11:37:19,643 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.022e+02 1.900e+02 2.405e+02 3.076e+02 6.689e+02, threshold=4.810e+02, percent-clipped=2.0 2022-12-08 11:37:32,556 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=125832.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:37:58,773 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.74 vs. limit=2.0 2022-12-08 11:38:26,385 INFO [train.py:873] (3/4) Epoch 17, batch 4900, loss[loss=0.09304, simple_loss=0.1363, pruned_loss=0.02489, over 14355.00 frames. ], tot_loss[loss=0.1077, simple_loss=0.1433, pruned_loss=0.03601, over 2021653.17 frames. ], batch size: 28, lr: 4.60e-03, grad_scale: 4.0 2022-12-08 11:38:28,228 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4366, 1.5456, 2.5280, 1.3761, 2.4853, 2.5038, 2.0238, 2.6564], device='cuda:3'), covar=tensor([0.0301, 0.2429, 0.0496, 0.1954, 0.0525, 0.0556, 0.1105, 0.0339], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0156, 0.0159, 0.0166, 0.0166, 0.0179, 0.0131, 0.0151], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 11:38:47,414 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.413e+02 2.229e+02 2.578e+02 3.069e+02 1.052e+03, threshold=5.155e+02, percent-clipped=5.0 2022-12-08 11:39:24,581 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125959.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:39:54,284 INFO [train.py:873] (3/4) Epoch 17, batch 5000, loss[loss=0.1024, simple_loss=0.1363, pruned_loss=0.03424, over 14156.00 frames. ], tot_loss[loss=0.1086, simple_loss=0.1443, pruned_loss=0.03649, over 2024021.74 frames. ], batch size: 99, lr: 4.60e-03, grad_scale: 4.0 2022-12-08 11:39:59,124 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125998.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:40:08,841 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=7.18 vs. limit=5.0 2022-12-08 11:40:11,462 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126012.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 11:40:15,916 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.327e+02 2.233e+02 2.614e+02 3.180e+02 5.706e+02, threshold=5.227e+02, percent-clipped=5.0 2022-12-08 11:40:19,916 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2022-12-08 11:40:23,465 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.13 vs. limit=5.0 2022-12-08 11:40:28,999 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126032.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:40:52,733 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126059.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:40:53,520 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=126060.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:41:08,234 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126077.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:41:08,321 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9677, 2.0603, 2.8692, 2.3114, 2.8807, 2.8016, 2.6416, 2.5029], device='cuda:3'), covar=tensor([0.0880, 0.2796, 0.1073, 0.1738, 0.0717, 0.1128, 0.1121, 0.1381], device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0310, 0.0393, 0.0299, 0.0363, 0.0323, 0.0360, 0.0296], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 11:41:11,124 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=126080.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:41:12,048 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126081.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:41:22,062 INFO [train.py:873] (3/4) Epoch 17, batch 5100, loss[loss=0.1218, simple_loss=0.1207, pruned_loss=0.06141, over 1256.00 frames. ], tot_loss[loss=0.1077, simple_loss=0.1431, pruned_loss=0.03619, over 1942229.61 frames. ], batch size: 100, lr: 4.60e-03, grad_scale: 4.0 2022-12-08 11:41:34,110 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5689, 3.8499, 3.6391, 3.5504, 2.9249, 3.8633, 3.6717, 2.1214], device='cuda:3'), covar=tensor([0.1432, 0.1079, 0.1028, 0.0822, 0.0767, 0.0387, 0.0811, 0.1743], device='cuda:3'), in_proj_covar=tensor([0.0138, 0.0089, 0.0070, 0.0076, 0.0099, 0.0090, 0.0101, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 11:41:40,590 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.1025, 2.6868, 5.1497, 3.4165, 4.8133, 2.4054, 3.7326, 4.7467], device='cuda:3'), covar=tensor([0.0492, 0.3654, 0.0310, 0.6432, 0.0648, 0.3247, 0.1329, 0.0689], device='cuda:3'), in_proj_covar=tensor([0.0255, 0.0202, 0.0221, 0.0274, 0.0239, 0.0206, 0.0204, 0.0221], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0005], device='cuda:3') 2022-12-08 11:41:42,935 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.364e+02 2.125e+02 2.682e+02 3.243e+02 7.763e+02, threshold=5.364e+02, percent-clipped=5.0 2022-12-08 11:41:50,068 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=126125.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:41:54,943 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9741, 3.5425, 2.7453, 4.0557, 3.9086, 3.9608, 3.4591, 2.8913], device='cuda:3'), covar=tensor([0.0690, 0.1163, 0.2933, 0.0671, 0.0861, 0.1161, 0.1139, 0.2550], device='cuda:3'), in_proj_covar=tensor([0.0287, 0.0293, 0.0264, 0.0291, 0.0326, 0.0306, 0.0257, 0.0246], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 11:42:29,384 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.6191, 1.4988, 2.7610, 1.4653, 2.7406, 2.7360, 2.0746, 2.8879], device='cuda:3'), covar=tensor([0.0278, 0.2539, 0.0408, 0.1914, 0.0448, 0.0478, 0.1009, 0.0280], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0155, 0.0158, 0.0165, 0.0166, 0.0178, 0.0131, 0.0149], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 11:42:31,691 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.19 vs. limit=5.0 2022-12-08 11:42:36,641 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1641, 1.1316, 1.0670, 1.2776, 1.2896, 0.8342, 0.9528, 1.1123], device='cuda:3'), covar=tensor([0.0647, 0.0842, 0.0785, 0.0590, 0.0574, 0.0784, 0.1191, 0.0790], device='cuda:3'), in_proj_covar=tensor([0.0036, 0.0035, 0.0040, 0.0033, 0.0035, 0.0048, 0.0036, 0.0039], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 11:42:49,534 INFO [train.py:873] (3/4) Epoch 17, batch 5200, loss[loss=0.09652, simple_loss=0.1365, pruned_loss=0.02827, over 14411.00 frames. ], tot_loss[loss=0.108, simple_loss=0.1435, pruned_loss=0.0362, over 1963101.18 frames. ], batch size: 53, lr: 4.60e-03, grad_scale: 8.0 2022-12-08 11:42:50,502 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.7984, 0.8186, 0.7039, 0.8817, 0.8957, 0.4441, 0.7901, 0.8524], device='cuda:3'), covar=tensor([0.0351, 0.0538, 0.0567, 0.0493, 0.0401, 0.0381, 0.0871, 0.0716], device='cuda:3'), in_proj_covar=tensor([0.0036, 0.0035, 0.0040, 0.0033, 0.0035, 0.0048, 0.0036, 0.0039], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 11:43:10,801 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.250e+02 2.118e+02 2.718e+02 3.349e+02 1.527e+03, threshold=5.437e+02, percent-clipped=4.0 2022-12-08 11:43:48,291 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126259.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:44:17,890 INFO [train.py:873] (3/4) Epoch 17, batch 5300, loss[loss=0.1728, simple_loss=0.1513, pruned_loss=0.09711, over 1297.00 frames. ], tot_loss[loss=0.1075, simple_loss=0.143, pruned_loss=0.03596, over 1970669.45 frames. ], batch size: 100, lr: 4.60e-03, grad_scale: 8.0 2022-12-08 11:44:20,955 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4093, 1.4745, 2.5504, 1.5565, 2.4749, 2.4951, 1.9253, 2.6574], device='cuda:3'), covar=tensor([0.0261, 0.2295, 0.0438, 0.1615, 0.0532, 0.0598, 0.1109, 0.0325], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0156, 0.0159, 0.0166, 0.0166, 0.0179, 0.0132, 0.0150], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 11:44:30,681 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=126307.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:44:39,315 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.051e+02 2.097e+02 2.584e+02 3.197e+02 6.466e+02, threshold=5.169e+02, percent-clipped=5.0 2022-12-08 11:45:12,377 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126354.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:45:13,912 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2022-12-08 11:45:35,442 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126380.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:45:36,313 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126381.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:45:46,520 INFO [train.py:873] (3/4) Epoch 17, batch 5400, loss[loss=0.1042, simple_loss=0.1466, pruned_loss=0.03088, over 14204.00 frames. ], tot_loss[loss=0.107, simple_loss=0.143, pruned_loss=0.03549, over 1987118.39 frames. ], batch size: 25, lr: 4.60e-03, grad_scale: 8.0 2022-12-08 11:45:48,108 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2022-12-08 11:46:08,053 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.423e+02 2.193e+02 2.737e+02 3.423e+02 8.531e+02, threshold=5.474e+02, percent-clipped=2.0 2022-12-08 11:46:19,126 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=126429.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:46:29,949 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126441.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:46:37,543 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126450.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:46:57,723 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.25 vs. limit=5.0 2022-12-08 11:47:14,892 INFO [train.py:873] (3/4) Epoch 17, batch 5500, loss[loss=0.1171, simple_loss=0.1326, pruned_loss=0.05079, over 3824.00 frames. ], tot_loss[loss=0.1061, simple_loss=0.1423, pruned_loss=0.03499, over 1956310.62 frames. ], batch size: 100, lr: 4.59e-03, grad_scale: 8.0 2022-12-08 11:47:31,203 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126511.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:47:36,284 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.274e+02 2.086e+02 2.565e+02 2.972e+02 5.597e+02, threshold=5.130e+02, percent-clipped=1.0 2022-12-08 11:47:45,226 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4118, 2.7353, 4.2238, 3.0946, 4.2230, 4.1265, 4.0519, 3.6414], device='cuda:3'), covar=tensor([0.0773, 0.3002, 0.0953, 0.1706, 0.0790, 0.0964, 0.1389, 0.1513], device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0310, 0.0393, 0.0298, 0.0364, 0.0323, 0.0360, 0.0296], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 11:48:42,925 INFO [train.py:873] (3/4) Epoch 17, batch 5600, loss[loss=0.163, simple_loss=0.1511, pruned_loss=0.08743, over 1312.00 frames. ], tot_loss[loss=0.1072, simple_loss=0.1431, pruned_loss=0.03568, over 2011179.89 frames. ], batch size: 100, lr: 4.59e-03, grad_scale: 8.0 2022-12-08 11:49:04,887 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.254e+02 1.973e+02 2.464e+02 3.253e+02 5.344e+02, threshold=4.927e+02, percent-clipped=2.0 2022-12-08 11:49:36,643 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126654.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:49:38,897 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.7012, 5.1731, 5.0785, 5.6476, 5.1542, 4.6862, 5.5518, 4.6461], device='cuda:3'), covar=tensor([0.0292, 0.0893, 0.0333, 0.0330, 0.0822, 0.0420, 0.0505, 0.0477], device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0275, 0.0200, 0.0194, 0.0185, 0.0158, 0.0287, 0.0169], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 11:49:51,686 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.10 vs. limit=2.0 2022-12-08 11:50:10,197 INFO [train.py:873] (3/4) Epoch 17, batch 5700, loss[loss=0.1239, simple_loss=0.1547, pruned_loss=0.04653, over 11977.00 frames. ], tot_loss[loss=0.1073, simple_loss=0.1434, pruned_loss=0.03562, over 2026479.90 frames. ], batch size: 100, lr: 4.59e-03, grad_scale: 8.0 2022-12-08 11:50:18,643 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=126702.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:50:32,277 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.662e+01 2.224e+02 2.546e+02 3.152e+02 6.937e+02, threshold=5.092e+02, percent-clipped=3.0 2022-12-08 11:50:48,056 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126736.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:51:26,989 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5290, 1.3845, 3.5096, 1.4929, 3.4013, 3.5905, 2.5888, 3.8821], device='cuda:3'), covar=tensor([0.0263, 0.3254, 0.0426, 0.2443, 0.0735, 0.0420, 0.0931, 0.0188], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0154, 0.0157, 0.0165, 0.0165, 0.0177, 0.0130, 0.0150], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 11:51:30,234 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8180, 1.8648, 2.0147, 1.8338, 1.7699, 1.7026, 1.3768, 1.2374], device='cuda:3'), covar=tensor([0.0213, 0.0228, 0.0195, 0.0212, 0.0227, 0.0314, 0.0304, 0.0430], device='cuda:3'), in_proj_covar=tensor([0.0022, 0.0022, 0.0020, 0.0021, 0.0021, 0.0033, 0.0027, 0.0032], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 11:51:37,726 INFO [train.py:873] (3/4) Epoch 17, batch 5800, loss[loss=0.08656, simple_loss=0.1339, pruned_loss=0.01959, over 14289.00 frames. ], tot_loss[loss=0.1073, simple_loss=0.1433, pruned_loss=0.03562, over 2031402.54 frames. ], batch size: 25, lr: 4.59e-03, grad_scale: 4.0 2022-12-08 11:51:49,359 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126806.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:51:57,973 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126816.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:52:00,003 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.29 vs. limit=5.0 2022-12-08 11:52:01,107 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.070e+02 2.153e+02 2.501e+02 2.856e+02 4.579e+02, threshold=5.001e+02, percent-clipped=0.0 2022-12-08 11:52:29,891 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.56 vs. limit=5.0 2022-12-08 11:52:51,097 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126877.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:53:04,763 INFO [train.py:873] (3/4) Epoch 17, batch 5900, loss[loss=0.1243, simple_loss=0.121, pruned_loss=0.0638, over 1372.00 frames. ], tot_loss[loss=0.1067, simple_loss=0.1428, pruned_loss=0.03532, over 2029221.74 frames. ], batch size: 100, lr: 4.59e-03, grad_scale: 2.0 2022-12-08 11:53:29,087 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.268e+02 2.272e+02 2.688e+02 3.245e+02 5.334e+02, threshold=5.377e+02, percent-clipped=3.0 2022-12-08 11:54:32,808 INFO [train.py:873] (3/4) Epoch 17, batch 6000, loss[loss=0.1312, simple_loss=0.166, pruned_loss=0.04824, over 14298.00 frames. ], tot_loss[loss=0.106, simple_loss=0.1424, pruned_loss=0.03485, over 2074843.81 frames. ], batch size: 39, lr: 4.58e-03, grad_scale: 4.0 2022-12-08 11:54:32,808 INFO [train.py:896] (3/4) Computing validation loss 2022-12-08 11:54:41,594 INFO [train.py:905] (3/4) Epoch 17, validation: loss=0.1381, simple_loss=0.176, pruned_loss=0.05009, over 857387.00 frames. 2022-12-08 11:54:41,595 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-08 11:54:57,629 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.39 vs. limit=5.0 2022-12-08 11:55:05,926 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.176e+02 1.999e+02 2.582e+02 3.270e+02 6.669e+02, threshold=5.164e+02, percent-clipped=4.0 2022-12-08 11:55:20,265 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=127036.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:56:01,545 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=127084.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:56:02,989 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2022-12-08 11:56:09,390 INFO [train.py:873] (3/4) Epoch 17, batch 6100, loss[loss=0.1327, simple_loss=0.1451, pruned_loss=0.06017, over 3867.00 frames. ], tot_loss[loss=0.106, simple_loss=0.1425, pruned_loss=0.03482, over 2080382.01 frames. ], batch size: 100, lr: 4.58e-03, grad_scale: 4.0 2022-12-08 11:56:13,183 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7572, 3.8729, 4.0550, 3.6089, 3.9114, 3.9527, 1.5985, 3.6989], device='cuda:3'), covar=tensor([0.0363, 0.0317, 0.0347, 0.0549, 0.0338, 0.0362, 0.3099, 0.0297], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0175, 0.0146, 0.0148, 0.0207, 0.0142, 0.0157, 0.0194], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 11:56:21,098 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=127106.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:56:28,879 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2022-12-08 11:56:33,264 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.291e+02 2.062e+02 2.398e+02 3.031e+02 5.398e+02, threshold=4.796e+02, percent-clipped=2.0 2022-12-08 11:57:03,158 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=127154.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:57:18,835 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127172.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:57:37,194 INFO [train.py:873] (3/4) Epoch 17, batch 6200, loss[loss=0.1067, simple_loss=0.138, pruned_loss=0.03769, over 9478.00 frames. ], tot_loss[loss=0.1064, simple_loss=0.1425, pruned_loss=0.03509, over 2054063.39 frames. ], batch size: 100, lr: 4.58e-03, grad_scale: 4.0 2022-12-08 11:57:39,467 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1437, 2.0613, 4.3758, 4.0654, 4.0507, 4.4963, 3.9798, 4.5317], device='cuda:3'), covar=tensor([0.1513, 0.1442, 0.0119, 0.0242, 0.0241, 0.0143, 0.0206, 0.0121], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0159, 0.0130, 0.0169, 0.0147, 0.0143, 0.0126, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 11:58:01,694 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.348e+02 2.117e+02 2.626e+02 3.289e+02 8.585e+02, threshold=5.253e+02, percent-clipped=4.0 2022-12-08 11:58:26,209 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.84 vs. limit=5.0 2022-12-08 11:58:27,551 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=127249.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:59:05,810 INFO [train.py:873] (3/4) Epoch 17, batch 6300, loss[loss=0.09974, simple_loss=0.1412, pruned_loss=0.02913, over 14453.00 frames. ], tot_loss[loss=0.1058, simple_loss=0.1423, pruned_loss=0.03469, over 1984211.34 frames. ], batch size: 51, lr: 4.58e-03, grad_scale: 4.0 2022-12-08 11:59:21,126 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127310.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 11:59:29,649 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.358e+02 2.074e+02 2.479e+02 3.103e+02 6.398e+02, threshold=4.958e+02, percent-clipped=2.0 2022-12-08 12:00:33,563 INFO [train.py:873] (3/4) Epoch 17, batch 6400, loss[loss=0.09839, simple_loss=0.1326, pruned_loss=0.03211, over 6909.00 frames. ], tot_loss[loss=0.1056, simple_loss=0.1422, pruned_loss=0.0345, over 1949017.62 frames. ], batch size: 100, lr: 4.58e-03, grad_scale: 8.0 2022-12-08 12:00:58,035 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.677e+01 2.186e+02 2.674e+02 3.273e+02 5.591e+02, threshold=5.347e+02, percent-clipped=3.0 2022-12-08 12:01:04,706 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=127427.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 12:01:24,623 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.3289, 1.6641, 1.8499, 1.8770, 1.6689, 1.8333, 1.5776, 1.2336], device='cuda:3'), covar=tensor([0.1076, 0.1555, 0.0757, 0.0708, 0.1335, 0.1006, 0.1848, 0.2500], device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0089, 0.0070, 0.0075, 0.0098, 0.0089, 0.0100, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 12:01:28,837 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3365, 1.4529, 2.4886, 1.4032, 2.4465, 2.4543, 1.9775, 2.6428], device='cuda:3'), covar=tensor([0.0289, 0.2359, 0.0464, 0.1874, 0.0530, 0.0641, 0.1196, 0.0355], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0152, 0.0157, 0.0164, 0.0165, 0.0177, 0.0130, 0.0149], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 12:01:43,652 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=127472.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:01:57,872 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127488.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 12:02:01,914 INFO [train.py:873] (3/4) Epoch 17, batch 6500, loss[loss=0.1018, simple_loss=0.1376, pruned_loss=0.03305, over 14252.00 frames. ], tot_loss[loss=0.1069, simple_loss=0.1428, pruned_loss=0.03544, over 1896600.96 frames. ], batch size: 63, lr: 4.58e-03, grad_scale: 8.0 2022-12-08 12:02:25,882 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.047e+02 2.139e+02 2.589e+02 3.452e+02 6.629e+02, threshold=5.179e+02, percent-clipped=2.0 2022-12-08 12:02:25,992 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=127520.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:02:34,174 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=127529.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:02:40,095 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4307, 1.3987, 1.2892, 1.5478, 1.5380, 0.9756, 1.1751, 1.3129], device='cuda:3'), covar=tensor([0.0591, 0.0715, 0.0670, 0.0829, 0.0488, 0.0806, 0.1054, 0.0680], device='cuda:3'), in_proj_covar=tensor([0.0036, 0.0035, 0.0039, 0.0032, 0.0034, 0.0048, 0.0036, 0.0038], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 12:02:56,082 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3204, 4.0689, 3.9196, 4.3832, 4.0529, 3.9117, 4.4011, 3.6820], device='cuda:3'), covar=tensor([0.0490, 0.1012, 0.0445, 0.0444, 0.0866, 0.0948, 0.0520, 0.0579], device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0274, 0.0199, 0.0193, 0.0185, 0.0156, 0.0284, 0.0168], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 12:03:20,558 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8749, 3.9163, 3.9901, 3.6651, 3.9526, 4.1238, 1.4790, 3.6933], device='cuda:3'), covar=tensor([0.0391, 0.0406, 0.0546, 0.0654, 0.0433, 0.0320, 0.3713, 0.0415], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0175, 0.0147, 0.0148, 0.0207, 0.0142, 0.0157, 0.0194], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 12:03:27,654 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127590.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:03:29,962 INFO [train.py:873] (3/4) Epoch 17, batch 6600, loss[loss=0.1262, simple_loss=0.1552, pruned_loss=0.04856, over 14567.00 frames. ], tot_loss[loss=0.1068, simple_loss=0.1427, pruned_loss=0.03547, over 1946470.66 frames. ], batch size: 22, lr: 4.57e-03, grad_scale: 8.0 2022-12-08 12:03:36,461 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.3769, 1.0617, 1.2123, 0.8535, 1.1338, 1.3512, 1.0810, 1.1120], device='cuda:3'), covar=tensor([0.0524, 0.0957, 0.0810, 0.0604, 0.1075, 0.1009, 0.0659, 0.1392], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0172, 0.0141, 0.0126, 0.0145, 0.0157, 0.0136, 0.0143], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0007, 0.0006, 0.0006], device='cuda:3') 2022-12-08 12:03:40,641 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127605.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:03:54,225 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.226e+02 2.073e+02 2.578e+02 3.202e+02 5.753e+02, threshold=5.156e+02, percent-clipped=2.0 2022-12-08 12:03:55,245 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1190, 3.1953, 3.1869, 3.1802, 2.5266, 3.3614, 3.1674, 1.9073], device='cuda:3'), covar=tensor([0.1001, 0.1062, 0.0926, 0.0699, 0.0950, 0.0444, 0.0892, 0.1861], device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0089, 0.0069, 0.0075, 0.0098, 0.0089, 0.0100, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 12:04:30,719 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=127662.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:04:47,348 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0250, 1.9983, 2.0830, 2.0734, 1.9818, 1.6399, 1.3233, 1.7989], device='cuda:3'), covar=tensor([0.0706, 0.0661, 0.0501, 0.0419, 0.0544, 0.1555, 0.2510, 0.0551], device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0176, 0.0148, 0.0149, 0.0209, 0.0143, 0.0158, 0.0195], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 12:04:57,440 INFO [train.py:873] (3/4) Epoch 17, batch 6700, loss[loss=0.09564, simple_loss=0.1351, pruned_loss=0.02808, over 13881.00 frames. ], tot_loss[loss=0.1072, simple_loss=0.1428, pruned_loss=0.0358, over 1901686.34 frames. ], batch size: 23, lr: 4.57e-03, grad_scale: 4.0 2022-12-08 12:05:21,963 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.328e+02 2.055e+02 2.568e+02 3.169e+02 6.326e+02, threshold=5.135e+02, percent-clipped=1.0 2022-12-08 12:05:22,530 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.45 vs. limit=5.0 2022-12-08 12:05:23,957 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127723.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:05:59,340 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2022-12-08 12:06:05,557 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.69 vs. limit=2.0 2022-12-08 12:06:07,690 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8257, 1.8587, 1.9076, 1.9244, 1.8752, 1.7228, 1.4490, 1.3145], device='cuda:3'), covar=tensor([0.0259, 0.0501, 0.0228, 0.0234, 0.0269, 0.0278, 0.0306, 0.0469], device='cuda:3'), in_proj_covar=tensor([0.0022, 0.0022, 0.0020, 0.0021, 0.0021, 0.0033, 0.0028, 0.0032], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-08 12:06:13,185 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=127779.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:06:16,519 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127783.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 12:06:25,291 INFO [train.py:873] (3/4) Epoch 17, batch 6800, loss[loss=0.1156, simple_loss=0.1291, pruned_loss=0.05106, over 2634.00 frames. ], tot_loss[loss=0.1071, simple_loss=0.1428, pruned_loss=0.03566, over 1974241.47 frames. ], batch size: 100, lr: 4.57e-03, grad_scale: 8.0 2022-12-08 12:06:50,326 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.244e+02 2.046e+02 2.510e+02 3.119e+02 5.930e+02, threshold=5.019e+02, percent-clipped=2.0 2022-12-08 12:06:55,529 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.3875, 1.0299, 1.2021, 0.7794, 1.1075, 1.3777, 1.0098, 1.0658], device='cuda:3'), covar=tensor([0.0596, 0.0901, 0.0856, 0.0539, 0.1055, 0.0755, 0.0602, 0.1481], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0172, 0.0141, 0.0126, 0.0144, 0.0157, 0.0136, 0.0143], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0007, 0.0006, 0.0006], device='cuda:3') 2022-12-08 12:07:07,188 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127840.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:07:16,130 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=127850.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:07:17,182 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.29 vs. limit=5.0 2022-12-08 12:07:35,252 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=127872.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:07:46,387 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127885.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:07:53,423 INFO [train.py:873] (3/4) Epoch 17, batch 6900, loss[loss=0.1028, simple_loss=0.1285, pruned_loss=0.03857, over 3881.00 frames. ], tot_loss[loss=0.1066, simple_loss=0.1427, pruned_loss=0.03532, over 2023295.53 frames. ], batch size: 100, lr: 4.57e-03, grad_scale: 8.0 2022-12-08 12:08:03,979 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=127905.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:08:09,348 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127911.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:08:17,922 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.924e+01 2.068e+02 2.418e+02 3.139e+02 1.064e+03, threshold=4.836e+02, percent-clipped=7.0 2022-12-08 12:08:28,437 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127933.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:08:46,270 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=127953.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:08:47,192 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1241, 1.4575, 3.9869, 1.9660, 4.0085, 4.1457, 3.2785, 4.5010], device='cuda:3'), covar=tensor([0.0219, 0.3212, 0.0432, 0.2127, 0.0406, 0.0363, 0.0658, 0.0166], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0154, 0.0158, 0.0166, 0.0166, 0.0178, 0.0132, 0.0151], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 12:09:20,964 INFO [train.py:873] (3/4) Epoch 17, batch 7000, loss[loss=0.125, simple_loss=0.1572, pruned_loss=0.04643, over 12753.00 frames. ], tot_loss[loss=0.1069, simple_loss=0.1426, pruned_loss=0.03564, over 1969343.04 frames. ], batch size: 100, lr: 4.57e-03, grad_scale: 8.0 2022-12-08 12:09:43,966 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128018.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:09:47,265 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.330e+02 2.259e+02 2.605e+02 3.114e+02 4.629e+02, threshold=5.210e+02, percent-clipped=0.0 2022-12-08 12:09:59,622 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5805, 3.2715, 3.1460, 2.3452, 2.9188, 3.2462, 3.5148, 2.8427], device='cuda:3'), covar=tensor([0.0542, 0.0751, 0.0755, 0.1120, 0.0850, 0.0642, 0.0606, 0.1040], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0170, 0.0140, 0.0125, 0.0144, 0.0156, 0.0136, 0.0142], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0007, 0.0006, 0.0006], device='cuda:3') 2022-12-08 12:10:41,188 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128083.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 12:10:49,319 INFO [train.py:873] (3/4) Epoch 17, batch 7100, loss[loss=0.1203, simple_loss=0.158, pruned_loss=0.04131, over 14386.00 frames. ], tot_loss[loss=0.1063, simple_loss=0.1421, pruned_loss=0.03528, over 1965545.38 frames. ], batch size: 55, lr: 4.56e-03, grad_scale: 4.0 2022-12-08 12:10:57,102 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.7809, 1.8857, 4.5144, 2.3118, 4.3477, 4.6603, 4.3059, 5.1384], device='cuda:3'), covar=tensor([0.0198, 0.2913, 0.0363, 0.2030, 0.0299, 0.0340, 0.0263, 0.0142], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0156, 0.0159, 0.0168, 0.0167, 0.0179, 0.0133, 0.0151], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 12:11:14,254 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.191e+02 2.219e+02 2.800e+02 3.533e+02 6.463e+02, threshold=5.601e+02, percent-clipped=4.0 2022-12-08 12:11:22,289 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=128131.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 12:11:25,704 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128135.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:12:09,581 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128185.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:12:16,539 INFO [train.py:873] (3/4) Epoch 17, batch 7200, loss[loss=0.1012, simple_loss=0.1386, pruned_loss=0.03192, over 14163.00 frames. ], tot_loss[loss=0.1068, simple_loss=0.1427, pruned_loss=0.03551, over 2003757.87 frames. ], batch size: 99, lr: 4.56e-03, grad_scale: 8.0 2022-12-08 12:12:24,173 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7007, 3.7972, 3.9653, 3.5236, 3.8292, 3.8681, 1.5357, 3.6665], device='cuda:3'), covar=tensor([0.0356, 0.0338, 0.0331, 0.0518, 0.0320, 0.0390, 0.3157, 0.0288], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0174, 0.0146, 0.0147, 0.0205, 0.0141, 0.0156, 0.0192], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 12:12:26,314 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.62 vs. limit=5.0 2022-12-08 12:12:28,389 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128206.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:12:42,993 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.381e+02 2.064e+02 2.461e+02 3.111e+02 6.198e+02, threshold=4.923e+02, percent-clipped=2.0 2022-12-08 12:12:47,578 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128228.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:12:52,162 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=128233.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:13:44,323 INFO [train.py:873] (3/4) Epoch 17, batch 7300, loss[loss=0.1119, simple_loss=0.1393, pruned_loss=0.04218, over 4969.00 frames. ], tot_loss[loss=0.106, simple_loss=0.1421, pruned_loss=0.0349, over 2010474.41 frames. ], batch size: 100, lr: 4.56e-03, grad_scale: 4.0 2022-12-08 12:14:06,391 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128318.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:14:08,075 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128320.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:14:10,504 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.359e+02 2.143e+02 2.591e+02 3.352e+02 1.032e+03, threshold=5.183e+02, percent-clipped=4.0 2022-12-08 12:14:32,002 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2022-12-08 12:14:47,737 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=128366.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:15:00,862 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128381.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:15:11,145 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2022-12-08 12:15:11,328 INFO [train.py:873] (3/4) Epoch 17, batch 7400, loss[loss=0.1096, simple_loss=0.1391, pruned_loss=0.04001, over 5996.00 frames. ], tot_loss[loss=0.1077, simple_loss=0.1429, pruned_loss=0.03627, over 1925303.27 frames. ], batch size: 100, lr: 4.56e-03, grad_scale: 4.0 2022-12-08 12:15:29,214 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128413.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:15:29,983 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.3847, 1.0593, 1.1913, 0.8102, 1.1414, 1.4021, 1.0695, 1.0724], device='cuda:3'), covar=tensor([0.0540, 0.1030, 0.0875, 0.0543, 0.1209, 0.0772, 0.0808, 0.1387], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0171, 0.0141, 0.0126, 0.0145, 0.0158, 0.0138, 0.0143], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0007, 0.0006, 0.0006], device='cuda:3') 2022-12-08 12:15:37,701 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.490e+02 2.104e+02 2.629e+02 3.153e+02 1.005e+03, threshold=5.259e+02, percent-clipped=2.0 2022-12-08 12:15:48,025 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5410, 4.2798, 4.0491, 4.2119, 4.3845, 4.4560, 4.5381, 4.5412], device='cuda:3'), covar=tensor([0.0709, 0.0458, 0.1922, 0.2276, 0.0681, 0.0706, 0.0728, 0.0650], device='cuda:3'), in_proj_covar=tensor([0.0393, 0.0276, 0.0455, 0.0569, 0.0354, 0.0456, 0.0395, 0.0400], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 12:15:48,074 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128435.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:15:53,179 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.65 vs. limit=2.0 2022-12-08 12:16:05,403 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9673, 2.0429, 2.1953, 1.4977, 1.6370, 1.9740, 1.2837, 1.9982], device='cuda:3'), covar=tensor([0.1269, 0.1551, 0.0840, 0.2077, 0.2335, 0.1140, 0.3168, 0.1002], device='cuda:3'), in_proj_covar=tensor([0.0086, 0.0102, 0.0095, 0.0100, 0.0114, 0.0091, 0.0117, 0.0093], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 12:16:22,016 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128474.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:16:29,534 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=128483.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:16:38,248 INFO [train.py:873] (3/4) Epoch 17, batch 7500, loss[loss=0.09773, simple_loss=0.1401, pruned_loss=0.02765, over 14082.00 frames. ], tot_loss[loss=0.1062, simple_loss=0.1426, pruned_loss=0.03488, over 2027487.30 frames. ], batch size: 29, lr: 4.56e-03, grad_scale: 4.0 2022-12-08 12:16:49,227 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128506.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:17:04,009 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.021e+02 2.073e+02 2.518e+02 3.139e+02 6.346e+02, threshold=5.035e+02, percent-clipped=2.0 2022-12-08 12:17:08,366 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128528.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:17:09,111 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0119, 2.0261, 2.0843, 2.0949, 2.0411, 1.6428, 1.4636, 1.8375], device='cuda:3'), covar=tensor([0.0692, 0.0536, 0.0510, 0.0369, 0.0503, 0.1467, 0.2272, 0.0509], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0175, 0.0146, 0.0147, 0.0207, 0.0142, 0.0157, 0.0192], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 12:18:06,085 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=128554.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:18:06,902 INFO [train.py:873] (3/4) Epoch 18, batch 0, loss[loss=0.1317, simple_loss=0.1683, pruned_loss=0.04751, over 14231.00 frames. ], tot_loss[loss=0.1317, simple_loss=0.1683, pruned_loss=0.04751, over 14231.00 frames. ], batch size: 60, lr: 4.43e-03, grad_scale: 8.0 2022-12-08 12:18:06,902 INFO [train.py:896] (3/4) Computing validation loss 2022-12-08 12:18:14,462 INFO [train.py:905] (3/4) Epoch 18, validation: loss=0.1457, simple_loss=0.1856, pruned_loss=0.05295, over 857387.00 frames. 2022-12-08 12:18:14,463 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-08 12:18:29,088 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8184, 2.5144, 3.6703, 2.9176, 3.6582, 3.5522, 3.5253, 3.1411], device='cuda:3'), covar=tensor([0.1078, 0.2716, 0.1129, 0.1687, 0.0951, 0.1079, 0.1327, 0.1652], device='cuda:3'), in_proj_covar=tensor([0.0354, 0.0311, 0.0391, 0.0298, 0.0364, 0.0322, 0.0361, 0.0297], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 12:18:33,519 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=128576.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:19:02,438 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8522, 1.7484, 4.6816, 2.3140, 4.3824, 4.8274, 4.4364, 5.2695], device='cuda:3'), covar=tensor([0.0193, 0.3059, 0.0303, 0.2024, 0.0301, 0.0344, 0.0282, 0.0162], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0155, 0.0158, 0.0167, 0.0167, 0.0178, 0.0132, 0.0151], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 12:19:15,798 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2650, 2.8478, 2.9411, 1.9426, 2.6842, 2.9765, 3.2640, 2.5575], device='cuda:3'), covar=tensor([0.0645, 0.0907, 0.0943, 0.1422, 0.1164, 0.0801, 0.0573, 0.1266], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0172, 0.0141, 0.0126, 0.0146, 0.0157, 0.0138, 0.0144], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0007, 0.0006, 0.0006], device='cuda:3') 2022-12-08 12:19:16,502 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 5.261e+01 1.855e+02 2.639e+02 3.708e+02 1.096e+03, threshold=5.279e+02, percent-clipped=10.0 2022-12-08 12:19:43,879 INFO [train.py:873] (3/4) Epoch 18, batch 100, loss[loss=0.1196, simple_loss=0.1466, pruned_loss=0.04629, over 6001.00 frames. ], tot_loss[loss=0.105, simple_loss=0.1425, pruned_loss=0.03372, over 825016.74 frames. ], batch size: 100, lr: 4.42e-03, grad_scale: 4.0 2022-12-08 12:19:46,784 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7570, 2.7422, 2.6082, 2.8651, 2.4737, 2.5452, 2.8269, 2.7449], device='cuda:3'), covar=tensor([0.0782, 0.1014, 0.1094, 0.0689, 0.1115, 0.0792, 0.0823, 0.0905], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0144, 0.0148, 0.0162, 0.0149, 0.0125, 0.0170, 0.0151], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 12:20:01,887 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128676.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:20:42,730 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.284e+02 2.050e+02 2.470e+02 3.144e+02 6.150e+02, threshold=4.940e+02, percent-clipped=2.0 2022-12-08 12:21:09,631 INFO [train.py:873] (3/4) Epoch 18, batch 200, loss[loss=0.1456, simple_loss=0.1404, pruned_loss=0.07546, over 1253.00 frames. ], tot_loss[loss=0.1069, simple_loss=0.1433, pruned_loss=0.03528, over 1263285.22 frames. ], batch size: 100, lr: 4.42e-03, grad_scale: 4.0 2022-12-08 12:21:22,390 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128769.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:21:32,280 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128781.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:21:58,289 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0137, 2.0297, 2.0739, 2.0865, 2.0401, 1.6527, 1.3261, 1.8674], device='cuda:3'), covar=tensor([0.0744, 0.0596, 0.0538, 0.0370, 0.0537, 0.1493, 0.2358, 0.0518], device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0177, 0.0147, 0.0149, 0.0208, 0.0144, 0.0159, 0.0195], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 12:22:09,532 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.012e+02 1.948e+02 2.392e+02 2.813e+02 5.621e+02, threshold=4.784e+02, percent-clipped=1.0 2022-12-08 12:22:25,113 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 2022-12-08 12:22:25,636 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128842.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:22:31,463 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1303, 3.9351, 3.7738, 4.2115, 3.9176, 3.6893, 4.2584, 3.5118], device='cuda:3'), covar=tensor([0.0588, 0.1000, 0.0504, 0.0447, 0.0777, 0.1213, 0.0542, 0.0617], device='cuda:3'), in_proj_covar=tensor([0.0179, 0.0274, 0.0200, 0.0193, 0.0183, 0.0157, 0.0285, 0.0169], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 12:22:37,493 INFO [train.py:873] (3/4) Epoch 18, batch 300, loss[loss=0.08145, simple_loss=0.1281, pruned_loss=0.01741, over 14288.00 frames. ], tot_loss[loss=0.1071, simple_loss=0.1431, pruned_loss=0.03555, over 1519832.81 frames. ], batch size: 63, lr: 4.42e-03, grad_scale: 4.0 2022-12-08 12:22:50,348 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 2022-12-08 12:23:00,009 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8349, 2.0976, 2.8494, 2.9108, 2.8384, 2.1468, 2.8047, 2.3759], device='cuda:3'), covar=tensor([0.0509, 0.1222, 0.0756, 0.0528, 0.0570, 0.1505, 0.0464, 0.0919], device='cuda:3'), in_proj_covar=tensor([0.0297, 0.0263, 0.0379, 0.0335, 0.0274, 0.0309, 0.0314, 0.0280], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 12:23:08,224 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128890.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 12:23:25,308 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.12 vs. limit=2.0 2022-12-08 12:23:31,131 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9879, 4.6882, 4.3790, 4.6087, 4.6799, 4.8930, 4.9578, 4.9609], device='cuda:3'), covar=tensor([0.0627, 0.0405, 0.1831, 0.2160, 0.0678, 0.0696, 0.0633, 0.0641], device='cuda:3'), in_proj_covar=tensor([0.0393, 0.0279, 0.0454, 0.0570, 0.0353, 0.0456, 0.0394, 0.0404], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 12:23:37,893 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.345e+02 2.277e+02 2.735e+02 3.513e+02 6.767e+02, threshold=5.469e+02, percent-clipped=7.0 2022-12-08 12:23:43,248 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9028, 1.4333, 3.9027, 1.7038, 3.8829, 4.0077, 2.9829, 4.2952], device='cuda:3'), covar=tensor([0.0232, 0.3161, 0.0360, 0.2280, 0.0391, 0.0330, 0.0744, 0.0169], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0157, 0.0160, 0.0169, 0.0169, 0.0180, 0.0133, 0.0153], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 12:24:02,423 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128951.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 12:24:05,632 INFO [train.py:873] (3/4) Epoch 18, batch 400, loss[loss=0.09988, simple_loss=0.1409, pruned_loss=0.02941, over 14206.00 frames. ], tot_loss[loss=0.1056, simple_loss=0.1423, pruned_loss=0.03445, over 1755705.88 frames. ], batch size: 60, lr: 4.42e-03, grad_scale: 8.0 2022-12-08 12:24:24,467 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128976.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:24:54,670 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6363, 3.3731, 3.1209, 3.3437, 3.5172, 3.5466, 3.5944, 3.5885], device='cuda:3'), covar=tensor([0.0753, 0.0606, 0.1961, 0.2172, 0.0781, 0.0942, 0.0889, 0.0838], device='cuda:3'), in_proj_covar=tensor([0.0393, 0.0279, 0.0456, 0.0573, 0.0353, 0.0458, 0.0395, 0.0404], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 12:25:06,979 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.216e+02 2.168e+02 2.627e+02 3.251e+02 6.252e+02, threshold=5.254e+02, percent-clipped=2.0 2022-12-08 12:25:07,075 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=129024.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:25:34,887 INFO [train.py:873] (3/4) Epoch 18, batch 500, loss[loss=0.1163, simple_loss=0.1566, pruned_loss=0.03801, over 13810.00 frames. ], tot_loss[loss=0.1072, simple_loss=0.1431, pruned_loss=0.03568, over 1796599.85 frames. ], batch size: 23, lr: 4.42e-03, grad_scale: 8.0 2022-12-08 12:25:47,415 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=129069.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:26:11,888 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.5704, 1.5162, 1.4117, 1.5538, 1.6782, 1.0642, 1.4045, 1.4001], device='cuda:3'), covar=tensor([0.0635, 0.0631, 0.0852, 0.0587, 0.0568, 0.0926, 0.0835, 0.0706], device='cuda:3'), in_proj_covar=tensor([0.0037, 0.0035, 0.0040, 0.0033, 0.0034, 0.0048, 0.0037, 0.0038], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 12:26:29,818 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=129117.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:26:35,673 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.427e+02 2.085e+02 2.575e+02 3.232e+02 8.773e+02, threshold=5.149e+02, percent-clipped=4.0 2022-12-08 12:26:47,846 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=129137.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:27:03,409 INFO [train.py:873] (3/4) Epoch 18, batch 600, loss[loss=0.1147, simple_loss=0.1491, pruned_loss=0.0402, over 14332.00 frames. ], tot_loss[loss=0.1071, simple_loss=0.1427, pruned_loss=0.03578, over 1834639.85 frames. ], batch size: 73, lr: 4.42e-03, grad_scale: 8.0 2022-12-08 12:28:04,833 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.209e+01 2.130e+02 2.657e+02 3.284e+02 6.073e+02, threshold=5.313e+02, percent-clipped=6.0 2022-12-08 12:28:08,584 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1154, 2.7202, 3.4511, 2.2867, 2.1992, 2.8340, 1.6852, 2.9266], device='cuda:3'), covar=tensor([0.0616, 0.1030, 0.0580, 0.2586, 0.1926, 0.0942, 0.2831, 0.1022], device='cuda:3'), in_proj_covar=tensor([0.0086, 0.0102, 0.0096, 0.0100, 0.0115, 0.0091, 0.0118, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 12:28:24,376 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=129246.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 12:28:32,551 INFO [train.py:873] (3/4) Epoch 18, batch 700, loss[loss=0.09761, simple_loss=0.1405, pruned_loss=0.02737, over 14420.00 frames. ], tot_loss[loss=0.1064, simple_loss=0.1422, pruned_loss=0.03531, over 1859541.46 frames. ], batch size: 53, lr: 4.41e-03, grad_scale: 4.0 2022-12-08 12:28:35,298 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3198, 4.2367, 3.9489, 4.3980, 3.9591, 3.8799, 4.3769, 4.1955], device='cuda:3'), covar=tensor([0.0594, 0.0853, 0.0868, 0.0519, 0.0824, 0.0600, 0.0620, 0.0662], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0148, 0.0150, 0.0166, 0.0151, 0.0126, 0.0174, 0.0153], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 12:29:33,262 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.087e+02 1.934e+02 2.414e+02 2.791e+02 7.591e+02, threshold=4.829e+02, percent-clipped=2.0 2022-12-08 12:29:45,544 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.51 vs. limit=5.0 2022-12-08 12:29:59,463 INFO [train.py:873] (3/4) Epoch 18, batch 800, loss[loss=0.1019, simple_loss=0.1426, pruned_loss=0.0306, over 14233.00 frames. ], tot_loss[loss=0.1074, simple_loss=0.1426, pruned_loss=0.03616, over 1845153.51 frames. ], batch size: 80, lr: 4.41e-03, grad_scale: 8.0 2022-12-08 12:30:05,756 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7967, 1.5325, 1.8283, 1.9812, 1.5605, 1.6817, 1.6043, 1.8748], device='cuda:3'), covar=tensor([0.0240, 0.0363, 0.0215, 0.0203, 0.0335, 0.0416, 0.0315, 0.0207], device='cuda:3'), in_proj_covar=tensor([0.0295, 0.0260, 0.0376, 0.0332, 0.0271, 0.0308, 0.0312, 0.0279], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 12:30:51,428 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=129413.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 12:31:01,566 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.411e+02 2.286e+02 2.776e+02 3.751e+02 1.143e+03, threshold=5.551e+02, percent-clipped=10.0 2022-12-08 12:31:11,912 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=129437.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:31:20,332 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.9724, 5.4122, 5.4204, 5.9036, 5.5317, 4.7971, 5.8647, 4.7720], device='cuda:3'), covar=tensor([0.0265, 0.0761, 0.0317, 0.0352, 0.0657, 0.0300, 0.0432, 0.0495], device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0276, 0.0201, 0.0194, 0.0185, 0.0158, 0.0285, 0.0169], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 12:31:27,671 INFO [train.py:873] (3/4) Epoch 18, batch 900, loss[loss=0.1705, simple_loss=0.1581, pruned_loss=0.09146, over 1247.00 frames. ], tot_loss[loss=0.1064, simple_loss=0.1419, pruned_loss=0.03547, over 1896197.74 frames. ], batch size: 100, lr: 4.41e-03, grad_scale: 8.0 2022-12-08 12:31:37,851 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4935, 3.2350, 2.4921, 3.5488, 3.4381, 3.4618, 2.9998, 2.5486], device='cuda:3'), covar=tensor([0.0844, 0.1248, 0.3115, 0.0668, 0.0888, 0.1007, 0.1245, 0.2814], device='cuda:3'), in_proj_covar=tensor([0.0280, 0.0288, 0.0258, 0.0288, 0.0321, 0.0302, 0.0252, 0.0241], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 12:31:40,222 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9097, 1.4863, 3.7035, 3.4063, 3.5086, 3.7875, 3.0487, 3.7549], device='cuda:3'), covar=tensor([0.1682, 0.1868, 0.0166, 0.0335, 0.0319, 0.0210, 0.0342, 0.0175], device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0156, 0.0130, 0.0167, 0.0147, 0.0142, 0.0126, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 12:31:44,563 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=129474.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 12:31:53,764 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=129485.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:32:28,737 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.182e+02 2.020e+02 2.482e+02 3.043e+02 9.483e+02, threshold=4.965e+02, percent-clipped=2.0 2022-12-08 12:32:48,043 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=129546.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 12:32:55,376 INFO [train.py:873] (3/4) Epoch 18, batch 1000, loss[loss=0.1169, simple_loss=0.1553, pruned_loss=0.0392, over 14067.00 frames. ], tot_loss[loss=0.1054, simple_loss=0.1418, pruned_loss=0.03454, over 1985156.41 frames. ], batch size: 22, lr: 4.41e-03, grad_scale: 8.0 2022-12-08 12:33:29,307 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=129594.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 12:33:45,365 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=129612.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:33:56,802 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.083e+02 2.171e+02 2.641e+02 3.416e+02 7.976e+02, threshold=5.283e+02, percent-clipped=6.0 2022-12-08 12:34:03,145 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.61 vs. limit=5.0 2022-12-08 12:34:23,015 INFO [train.py:873] (3/4) Epoch 18, batch 1100, loss[loss=0.1452, simple_loss=0.1651, pruned_loss=0.06266, over 7765.00 frames. ], tot_loss[loss=0.1059, simple_loss=0.142, pruned_loss=0.03487, over 1953331.01 frames. ], batch size: 100, lr: 4.41e-03, grad_scale: 8.0 2022-12-08 12:34:39,292 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=129673.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:35:23,666 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.177e+02 2.155e+02 2.731e+02 3.358e+02 7.442e+02, threshold=5.463e+02, percent-clipped=7.0 2022-12-08 12:35:36,968 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7571, 1.5324, 1.8097, 1.9278, 1.4933, 1.6391, 1.7014, 1.8158], device='cuda:3'), covar=tensor([0.0250, 0.0456, 0.0243, 0.0229, 0.0394, 0.0560, 0.0320, 0.0267], device='cuda:3'), in_proj_covar=tensor([0.0297, 0.0263, 0.0381, 0.0336, 0.0274, 0.0312, 0.0315, 0.0282], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 12:35:41,081 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2022-12-08 12:35:50,417 INFO [train.py:873] (3/4) Epoch 18, batch 1200, loss[loss=0.1372, simple_loss=0.1651, pruned_loss=0.05463, over 14361.00 frames. ], tot_loss[loss=0.1075, simple_loss=0.1431, pruned_loss=0.03594, over 1955490.04 frames. ], batch size: 55, lr: 4.41e-03, grad_scale: 8.0 2022-12-08 12:36:02,171 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=129769.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 12:36:10,972 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0452, 1.8388, 1.9616, 2.0136, 1.7644, 1.6894, 1.7149, 1.1602], device='cuda:3'), covar=tensor([0.0218, 0.0370, 0.0267, 0.0224, 0.0273, 0.0315, 0.0235, 0.0450], device='cuda:3'), in_proj_covar=tensor([0.0023, 0.0023, 0.0020, 0.0022, 0.0021, 0.0034, 0.0028, 0.0033], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-08 12:36:12,744 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1390, 2.1566, 3.0617, 2.3929, 3.0952, 2.9958, 2.8973, 2.6195], device='cuda:3'), covar=tensor([0.0753, 0.2681, 0.0898, 0.1796, 0.0829, 0.1061, 0.1174, 0.1719], device='cuda:3'), in_proj_covar=tensor([0.0357, 0.0314, 0.0396, 0.0300, 0.0368, 0.0326, 0.0361, 0.0300], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 12:36:21,545 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=129791.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:36:51,110 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.275e+02 2.078e+02 2.653e+02 3.269e+02 7.697e+02, threshold=5.307e+02, percent-clipped=3.0 2022-12-08 12:36:51,462 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.6452, 2.9031, 4.6080, 4.7892, 4.4995, 2.7948, 4.7872, 3.8295], device='cuda:3'), covar=tensor([0.0383, 0.1071, 0.0931, 0.0399, 0.0459, 0.1664, 0.0306, 0.0761], device='cuda:3'), in_proj_covar=tensor([0.0295, 0.0261, 0.0378, 0.0333, 0.0273, 0.0309, 0.0312, 0.0281], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 12:36:55,628 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5454, 3.2594, 3.0807, 2.2136, 2.9866, 3.2442, 3.5114, 2.8534], device='cuda:3'), covar=tensor([0.0602, 0.0870, 0.0886, 0.1235, 0.0910, 0.0675, 0.0626, 0.1225], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0170, 0.0141, 0.0124, 0.0142, 0.0154, 0.0135, 0.0142], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0007, 0.0006, 0.0006], device='cuda:3') 2022-12-08 12:37:14,929 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=129852.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:37:17,148 INFO [train.py:873] (3/4) Epoch 18, batch 1300, loss[loss=0.09825, simple_loss=0.1459, pruned_loss=0.02528, over 14543.00 frames. ], tot_loss[loss=0.1077, simple_loss=0.1431, pruned_loss=0.03618, over 1975828.51 frames. ], batch size: 43, lr: 4.40e-03, grad_scale: 8.0 2022-12-08 12:38:18,569 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.477e+02 2.020e+02 2.586e+02 3.074e+02 6.619e+02, threshold=5.173e+02, percent-clipped=1.0 2022-12-08 12:38:44,885 INFO [train.py:873] (3/4) Epoch 18, batch 1400, loss[loss=0.0835, simple_loss=0.136, pruned_loss=0.0155, over 14380.00 frames. ], tot_loss[loss=0.1067, simple_loss=0.1426, pruned_loss=0.03536, over 1941566.04 frames. ], batch size: 41, lr: 4.40e-03, grad_scale: 8.0 2022-12-08 12:38:55,882 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=129968.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:39:28,268 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130000.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:39:30,082 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130002.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:39:49,579 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9520, 2.2520, 2.5820, 1.5502, 2.5043, 2.6373, 2.9780, 2.2706], device='cuda:3'), covar=tensor([0.0667, 0.1083, 0.0967, 0.1704, 0.1167, 0.0915, 0.0716, 0.1411], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0171, 0.0141, 0.0125, 0.0143, 0.0155, 0.0135, 0.0142], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0007, 0.0006, 0.0006], device='cuda:3') 2022-12-08 12:39:51,054 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.109e+02 2.021e+02 2.494e+02 2.988e+02 5.995e+02, threshold=4.988e+02, percent-clipped=3.0 2022-12-08 12:40:15,733 INFO [train.py:873] (3/4) Epoch 18, batch 1500, loss[loss=0.1304, simple_loss=0.137, pruned_loss=0.06189, over 2650.00 frames. ], tot_loss[loss=0.1058, simple_loss=0.1413, pruned_loss=0.03512, over 1877501.33 frames. ], batch size: 100, lr: 4.40e-03, grad_scale: 4.0 2022-12-08 12:40:21,344 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130061.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:40:23,096 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130063.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 12:40:28,187 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9045, 1.6566, 3.1027, 2.8213, 2.9541, 3.1300, 2.3744, 3.1281], device='cuda:3'), covar=tensor([0.1350, 0.1441, 0.0193, 0.0400, 0.0389, 0.0217, 0.0516, 0.0200], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0159, 0.0132, 0.0169, 0.0148, 0.0144, 0.0127, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 12:40:28,193 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130069.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 12:41:09,681 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=130117.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 12:41:17,027 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.082e+02 2.103e+02 2.601e+02 2.993e+02 6.866e+02, threshold=5.203e+02, percent-clipped=5.0 2022-12-08 12:41:35,528 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130147.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:41:42,392 INFO [train.py:873] (3/4) Epoch 18, batch 1600, loss[loss=0.1013, simple_loss=0.1423, pruned_loss=0.03014, over 14276.00 frames. ], tot_loss[loss=0.1068, simple_loss=0.1418, pruned_loss=0.03587, over 1788841.55 frames. ], batch size: 76, lr: 4.40e-03, grad_scale: 8.0 2022-12-08 12:41:46,417 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 2022-12-08 12:42:03,146 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.54 vs. limit=5.0 2022-12-08 12:42:44,998 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.342e+02 2.040e+02 2.513e+02 3.158e+02 8.589e+02, threshold=5.027e+02, percent-clipped=3.0 2022-12-08 12:43:10,263 INFO [train.py:873] (3/4) Epoch 18, batch 1700, loss[loss=0.1219, simple_loss=0.1516, pruned_loss=0.04608, over 4985.00 frames. ], tot_loss[loss=0.1063, simple_loss=0.1416, pruned_loss=0.03552, over 1824005.36 frames. ], batch size: 100, lr: 4.40e-03, grad_scale: 8.0 2022-12-08 12:43:14,111 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4106, 3.2094, 3.9517, 2.9533, 2.4927, 3.4420, 1.8914, 3.5108], device='cuda:3'), covar=tensor([0.1043, 0.1013, 0.0522, 0.1623, 0.1967, 0.0907, 0.2917, 0.0758], device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0103, 0.0095, 0.0101, 0.0115, 0.0092, 0.0118, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 12:43:22,178 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130268.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:43:41,352 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2022-12-08 12:43:56,494 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.5668, 1.8296, 2.0110, 1.9784, 1.8116, 1.9821, 1.7558, 1.3786], device='cuda:3'), covar=tensor([0.1092, 0.1409, 0.0679, 0.0806, 0.1164, 0.0762, 0.1762, 0.1832], device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0090, 0.0070, 0.0076, 0.0100, 0.0090, 0.0100, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 12:44:04,004 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=130316.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:44:12,322 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.118e+02 2.048e+02 2.469e+02 3.168e+02 7.190e+02, threshold=4.938e+02, percent-clipped=8.0 2022-12-08 12:44:37,343 INFO [train.py:873] (3/4) Epoch 18, batch 1800, loss[loss=0.1452, simple_loss=0.1419, pruned_loss=0.07421, over 1207.00 frames. ], tot_loss[loss=0.1063, simple_loss=0.1415, pruned_loss=0.03553, over 1814561.57 frames. ], batch size: 100, lr: 4.40e-03, grad_scale: 8.0 2022-12-08 12:44:38,350 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130356.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:44:40,100 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130358.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 12:44:47,521 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2022-12-08 12:44:49,316 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.23 vs. limit=5.0 2022-12-08 12:44:58,543 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.56 vs. limit=5.0 2022-12-08 12:45:03,097 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130384.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:45:37,267 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2022-12-08 12:45:37,687 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7212, 2.2847, 2.4862, 1.4969, 2.3284, 2.5771, 2.7277, 2.2226], device='cuda:3'), covar=tensor([0.0798, 0.0849, 0.1112, 0.1828, 0.1007, 0.0832, 0.0719, 0.1382], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0169, 0.0139, 0.0124, 0.0141, 0.0153, 0.0134, 0.0140], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 12:45:39,183 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.710e+01 2.001e+02 2.479e+02 3.133e+02 6.689e+02, threshold=4.957e+02, percent-clipped=4.0 2022-12-08 12:45:44,507 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130432.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:45:49,332 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130438.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:45:55,538 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130445.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:45:57,222 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130447.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:46:04,199 INFO [train.py:873] (3/4) Epoch 18, batch 1900, loss[loss=0.1143, simple_loss=0.1466, pruned_loss=0.04099, over 5960.00 frames. ], tot_loss[loss=0.1069, simple_loss=0.1423, pruned_loss=0.03577, over 1878499.44 frames. ], batch size: 100, lr: 4.39e-03, grad_scale: 8.0 2022-12-08 12:46:37,641 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130493.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:46:39,393 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=130495.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:46:41,361 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4813, 1.6225, 2.6674, 2.1742, 2.5080, 1.6738, 2.2152, 2.4563], device='cuda:3'), covar=tensor([0.1794, 0.4029, 0.0850, 0.3087, 0.1495, 0.3287, 0.1274, 0.1273], device='cuda:3'), in_proj_covar=tensor([0.0258, 0.0201, 0.0223, 0.0274, 0.0242, 0.0207, 0.0205, 0.0224], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0005], device='cuda:3') 2022-12-08 12:46:42,989 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130499.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:46:50,133 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6045, 3.7745, 3.9469, 3.5231, 3.8136, 3.7801, 1.5971, 3.6250], device='cuda:3'), covar=tensor([0.0437, 0.0401, 0.0366, 0.0563, 0.0340, 0.0386, 0.3206, 0.0305], device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0176, 0.0146, 0.0149, 0.0208, 0.0143, 0.0158, 0.0195], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 12:47:06,079 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.096e+02 2.089e+02 2.666e+02 3.206e+02 5.999e+02, threshold=5.333e+02, percent-clipped=5.0 2022-12-08 12:47:25,687 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130548.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:47:31,377 INFO [train.py:873] (3/4) Epoch 18, batch 2000, loss[loss=0.13, simple_loss=0.1593, pruned_loss=0.05041, over 11174.00 frames. ], tot_loss[loss=0.1075, simple_loss=0.1432, pruned_loss=0.03594, over 1909034.39 frames. ], batch size: 100, lr: 4.39e-03, grad_scale: 8.0 2022-12-08 12:47:40,379 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130565.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:47:41,942 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2843, 3.3925, 3.5530, 3.2635, 3.4827, 3.1629, 1.5084, 3.2508], device='cuda:3'), covar=tensor([0.0487, 0.0433, 0.0354, 0.0546, 0.0344, 0.0665, 0.3241, 0.0330], device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0176, 0.0146, 0.0149, 0.0207, 0.0143, 0.0158, 0.0195], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 12:48:00,011 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.26 vs. limit=5.0 2022-12-08 12:48:03,046 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1361, 2.9000, 3.8526, 2.7833, 2.3221, 3.1385, 1.9414, 3.3615], device='cuda:3'), covar=tensor([0.1004, 0.1398, 0.0401, 0.1716, 0.1842, 0.0814, 0.2744, 0.0837], device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0103, 0.0096, 0.0101, 0.0116, 0.0092, 0.0118, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 12:48:13,566 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6485, 1.7436, 1.8320, 1.2988, 1.3109, 1.5881, 1.2293, 1.6763], device='cuda:3'), covar=tensor([0.1498, 0.2237, 0.0991, 0.2609, 0.3097, 0.1289, 0.2677, 0.1206], device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0103, 0.0095, 0.0101, 0.0116, 0.0092, 0.0118, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 12:48:18,852 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130609.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 12:48:34,463 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130626.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:48:35,244 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.254e+02 2.007e+02 2.517e+02 3.001e+02 4.503e+02, threshold=5.034e+02, percent-clipped=0.0 2022-12-08 12:48:59,206 INFO [train.py:873] (3/4) Epoch 18, batch 2100, loss[loss=0.1221, simple_loss=0.1248, pruned_loss=0.05967, over 1255.00 frames. ], tot_loss[loss=0.1056, simple_loss=0.1419, pruned_loss=0.03468, over 1941387.70 frames. ], batch size: 100, lr: 4.39e-03, grad_scale: 4.0 2022-12-08 12:49:00,458 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130656.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:49:02,111 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130658.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 12:49:41,923 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=130704.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:49:43,637 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=130706.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:50:01,668 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.093e+02 2.249e+02 2.636e+02 3.244e+02 1.045e+03, threshold=5.271e+02, percent-clipped=2.0 2022-12-08 12:50:13,029 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130740.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:50:26,108 INFO [train.py:873] (3/4) Epoch 18, batch 2200, loss[loss=0.09727, simple_loss=0.122, pruned_loss=0.03628, over 2626.00 frames. ], tot_loss[loss=0.1064, simple_loss=0.1424, pruned_loss=0.03516, over 1929630.86 frames. ], batch size: 100, lr: 4.39e-03, grad_scale: 4.0 2022-12-08 12:50:53,268 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2022-12-08 12:50:55,150 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130788.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:50:58,996 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.73 vs. limit=2.0 2022-12-08 12:51:00,288 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130794.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:51:29,862 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.247e+02 2.113e+02 2.522e+02 3.160e+02 5.458e+02, threshold=5.044e+02, percent-clipped=1.0 2022-12-08 12:51:53,885 INFO [train.py:873] (3/4) Epoch 18, batch 2300, loss[loss=0.0926, simple_loss=0.1368, pruned_loss=0.0242, over 14287.00 frames. ], tot_loss[loss=0.1065, simple_loss=0.1425, pruned_loss=0.03527, over 1942364.64 frames. ], batch size: 25, lr: 4.39e-03, grad_scale: 4.0 2022-12-08 12:52:37,093 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130904.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 12:52:51,127 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2932, 2.6236, 2.6652, 2.6149, 2.1807, 2.6884, 2.5297, 1.4227], device='cuda:3'), covar=tensor([0.0978, 0.0881, 0.0875, 0.0818, 0.1026, 0.0680, 0.1008, 0.2089], device='cuda:3'), in_proj_covar=tensor([0.0138, 0.0091, 0.0071, 0.0077, 0.0100, 0.0091, 0.0101, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 12:52:51,912 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130921.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:52:56,900 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.171e+02 2.034e+02 2.487e+02 3.081e+02 6.323e+02, threshold=4.975e+02, percent-clipped=2.0 2022-12-08 12:53:22,075 INFO [train.py:873] (3/4) Epoch 18, batch 2400, loss[loss=0.1301, simple_loss=0.1595, pruned_loss=0.05037, over 14291.00 frames. ], tot_loss[loss=0.1066, simple_loss=0.1427, pruned_loss=0.03523, over 1963731.99 frames. ], batch size: 76, lr: 4.39e-03, grad_scale: 8.0 2022-12-08 12:54:12,710 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 2022-12-08 12:54:25,831 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.164e+02 2.032e+02 2.544e+02 3.351e+02 8.852e+02, threshold=5.088e+02, percent-clipped=2.0 2022-12-08 12:54:37,359 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131040.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:54:50,163 INFO [train.py:873] (3/4) Epoch 18, batch 2500, loss[loss=0.1048, simple_loss=0.1391, pruned_loss=0.0352, over 13542.00 frames. ], tot_loss[loss=0.1064, simple_loss=0.1426, pruned_loss=0.03509, over 2005089.18 frames. ], batch size: 100, lr: 4.38e-03, grad_scale: 8.0 2022-12-08 12:55:01,572 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7616, 1.6300, 1.8869, 1.7531, 1.6621, 1.5579, 1.7650, 1.2690], device='cuda:3'), covar=tensor([0.0237, 0.0425, 0.0248, 0.0286, 0.0257, 0.0311, 0.0299, 0.0520], device='cuda:3'), in_proj_covar=tensor([0.0023, 0.0022, 0.0020, 0.0022, 0.0021, 0.0034, 0.0028, 0.0033], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-08 12:55:17,352 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8608, 1.4896, 3.8146, 1.8968, 3.8299, 3.9978, 2.9068, 4.2683], device='cuda:3'), covar=tensor([0.0249, 0.3217, 0.0421, 0.2073, 0.0482, 0.0369, 0.0736, 0.0174], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0156, 0.0160, 0.0167, 0.0167, 0.0179, 0.0134, 0.0153], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 12:55:19,766 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=131088.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:55:19,871 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131088.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:55:22,500 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8607, 1.7969, 1.5890, 1.9108, 1.8084, 1.8852, 1.7960, 1.7087], device='cuda:3'), covar=tensor([0.1047, 0.0967, 0.1948, 0.1021, 0.1212, 0.0631, 0.1455, 0.0969], device='cuda:3'), in_proj_covar=tensor([0.0280, 0.0287, 0.0256, 0.0288, 0.0320, 0.0300, 0.0251, 0.0242], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 12:55:25,080 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131094.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:55:33,160 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4367, 2.1653, 3.3153, 3.5299, 3.3322, 2.2449, 3.4411, 2.6087], device='cuda:3'), covar=tensor([0.0517, 0.1420, 0.0962, 0.0571, 0.0672, 0.1954, 0.0528, 0.1181], device='cuda:3'), in_proj_covar=tensor([0.0295, 0.0262, 0.0378, 0.0333, 0.0273, 0.0310, 0.0313, 0.0279], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 12:55:54,048 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.009e+02 2.192e+02 2.699e+02 3.352e+02 8.933e+02, threshold=5.398e+02, percent-clipped=5.0 2022-12-08 12:56:02,099 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=131136.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:56:02,802 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.68 vs. limit=2.0 2022-12-08 12:56:07,953 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=131142.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:56:19,379 INFO [train.py:873] (3/4) Epoch 18, batch 2600, loss[loss=0.149, simple_loss=0.1422, pruned_loss=0.0779, over 1219.00 frames. ], tot_loss[loss=0.1066, simple_loss=0.1427, pruned_loss=0.03522, over 1947207.68 frames. ], batch size: 100, lr: 4.38e-03, grad_scale: 8.0 2022-12-08 12:56:35,672 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131174.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:57:02,642 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131204.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 12:57:17,411 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131221.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:57:22,470 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.383e+02 2.143e+02 2.636e+02 3.180e+02 5.935e+02, threshold=5.272e+02, percent-clipped=1.0 2022-12-08 12:57:23,763 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2022-12-08 12:57:29,992 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131235.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 12:57:35,586 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.9389, 5.7521, 5.3174, 5.9834, 5.4353, 5.2560, 5.9655, 5.7632], device='cuda:3'), covar=tensor([0.0511, 0.0552, 0.0771, 0.0310, 0.0683, 0.0584, 0.0474, 0.0443], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0149, 0.0151, 0.0166, 0.0152, 0.0127, 0.0174, 0.0154], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 12:57:44,012 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=131252.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:57:46,567 INFO [train.py:873] (3/4) Epoch 18, batch 2700, loss[loss=0.101, simple_loss=0.1424, pruned_loss=0.0298, over 14295.00 frames. ], tot_loss[loss=0.1068, simple_loss=0.1429, pruned_loss=0.0354, over 1944652.46 frames. ], batch size: 25, lr: 4.38e-03, grad_scale: 4.0 2022-12-08 12:57:59,031 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=131269.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 12:58:50,253 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.346e+01 2.091e+02 2.717e+02 3.519e+02 6.308e+02, threshold=5.434e+02, percent-clipped=4.0 2022-12-08 12:59:14,300 INFO [train.py:873] (3/4) Epoch 18, batch 2800, loss[loss=0.1205, simple_loss=0.1563, pruned_loss=0.04232, over 14065.00 frames. ], tot_loss[loss=0.1064, simple_loss=0.1425, pruned_loss=0.03515, over 1984364.65 frames. ], batch size: 29, lr: 4.38e-03, grad_scale: 8.0 2022-12-08 12:59:41,297 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3357, 2.3146, 1.9446, 2.4506, 2.2553, 2.3505, 2.1575, 2.0309], device='cuda:3'), covar=tensor([0.1145, 0.1046, 0.2148, 0.1004, 0.1234, 0.0898, 0.1560, 0.1768], device='cuda:3'), in_proj_covar=tensor([0.0281, 0.0289, 0.0257, 0.0291, 0.0321, 0.0301, 0.0251, 0.0241], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 12:59:44,647 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8673, 1.8232, 1.9663, 1.7903, 1.8955, 1.7100, 1.6755, 1.2625], device='cuda:3'), covar=tensor([0.0202, 0.0307, 0.0182, 0.0220, 0.0193, 0.0286, 0.0262, 0.0395], device='cuda:3'), in_proj_covar=tensor([0.0023, 0.0022, 0.0020, 0.0022, 0.0021, 0.0034, 0.0028, 0.0033], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-08 12:59:48,853 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2022-12-08 13:00:02,563 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131409.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:00:08,495 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131416.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:00:18,981 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.304e+02 1.962e+02 2.375e+02 2.873e+02 4.439e+02, threshold=4.749e+02, percent-clipped=0.0 2022-12-08 13:00:23,398 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9088, 1.5284, 3.3202, 3.0376, 3.1621, 3.3518, 2.7622, 3.3254], device='cuda:3'), covar=tensor([0.1575, 0.1698, 0.0159, 0.0337, 0.0301, 0.0191, 0.0396, 0.0181], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0159, 0.0132, 0.0169, 0.0149, 0.0145, 0.0127, 0.0125], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 13:00:42,311 INFO [train.py:873] (3/4) Epoch 18, batch 2900, loss[loss=0.1127, simple_loss=0.1475, pruned_loss=0.03899, over 12791.00 frames. ], tot_loss[loss=0.1062, simple_loss=0.1425, pruned_loss=0.03498, over 1964685.95 frames. ], batch size: 100, lr: 4.38e-03, grad_scale: 8.0 2022-12-08 13:00:55,824 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131470.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:01:02,122 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131477.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:01:15,255 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131492.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:01:46,459 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.231e+02 2.257e+02 2.729e+02 3.510e+02 6.783e+02, threshold=5.458e+02, percent-clipped=7.0 2022-12-08 13:01:47,602 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3834, 3.0822, 3.8599, 2.7057, 2.4673, 3.4092, 1.9510, 3.4329], device='cuda:3'), covar=tensor([0.0889, 0.1233, 0.0550, 0.1883, 0.2019, 0.0863, 0.2868, 0.0980], device='cuda:3'), in_proj_covar=tensor([0.0086, 0.0103, 0.0095, 0.0101, 0.0116, 0.0092, 0.0117, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 13:01:48,363 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=131530.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 13:02:08,737 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131553.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:02:10,261 INFO [train.py:873] (3/4) Epoch 18, batch 3000, loss[loss=0.1115, simple_loss=0.1489, pruned_loss=0.03706, over 14219.00 frames. ], tot_loss[loss=0.1064, simple_loss=0.1426, pruned_loss=0.03504, over 1940483.93 frames. ], batch size: 89, lr: 4.38e-03, grad_scale: 8.0 2022-12-08 13:02:10,261 INFO [train.py:896] (3/4) Computing validation loss 2022-12-08 13:02:18,732 INFO [train.py:905] (3/4) Epoch 18, validation: loss=0.1388, simple_loss=0.176, pruned_loss=0.05082, over 857387.00 frames. 2022-12-08 13:02:18,733 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-08 13:02:44,188 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3561, 3.5103, 3.6650, 3.3425, 3.5050, 3.3194, 1.4873, 3.3469], device='cuda:3'), covar=tensor([0.0453, 0.0378, 0.0332, 0.0483, 0.0342, 0.0562, 0.3092, 0.0300], device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0177, 0.0147, 0.0150, 0.0209, 0.0143, 0.0159, 0.0196], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 13:03:22,346 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.140e+02 2.067e+02 2.535e+02 3.249e+02 7.779e+02, threshold=5.070e+02, percent-clipped=4.0 2022-12-08 13:03:44,987 INFO [train.py:873] (3/4) Epoch 18, batch 3100, loss[loss=0.09366, simple_loss=0.1358, pruned_loss=0.02576, over 14016.00 frames. ], tot_loss[loss=0.1072, simple_loss=0.143, pruned_loss=0.03571, over 1935568.17 frames. ], batch size: 29, lr: 4.37e-03, grad_scale: 8.0 2022-12-08 13:04:22,726 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9464, 1.4420, 3.1029, 1.7568, 3.1338, 3.0407, 2.1581, 3.2311], device='cuda:3'), covar=tensor([0.0265, 0.2738, 0.0424, 0.1768, 0.0369, 0.0438, 0.1007, 0.0260], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0156, 0.0160, 0.0167, 0.0168, 0.0178, 0.0133, 0.0153], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 13:04:49,758 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.195e+02 2.137e+02 2.592e+02 3.013e+02 4.856e+02, threshold=5.183e+02, percent-clipped=0.0 2022-12-08 13:04:54,592 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=7.14 vs. limit=5.0 2022-12-08 13:05:12,290 INFO [train.py:873] (3/4) Epoch 18, batch 3200, loss[loss=0.09932, simple_loss=0.135, pruned_loss=0.03184, over 5956.00 frames. ], tot_loss[loss=0.1068, simple_loss=0.1428, pruned_loss=0.0354, over 1957018.64 frames. ], batch size: 100, lr: 4.37e-03, grad_scale: 8.0 2022-12-08 13:05:20,961 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=131765.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:05:21,001 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.0846, 1.3522, 1.3380, 0.9976, 0.8066, 1.1179, 0.9359, 1.2130], device='cuda:3'), covar=tensor([0.2085, 0.2461, 0.1236, 0.2490, 0.3717, 0.1541, 0.1689, 0.1404], device='cuda:3'), in_proj_covar=tensor([0.0085, 0.0103, 0.0095, 0.0100, 0.0116, 0.0091, 0.0117, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 13:05:27,314 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=131772.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:05:44,667 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.7534, 1.7727, 4.5961, 2.3343, 4.2269, 4.7795, 4.3938, 5.1490], device='cuda:3'), covar=tensor([0.0227, 0.3063, 0.0416, 0.1980, 0.0333, 0.0356, 0.0317, 0.0170], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0154, 0.0159, 0.0166, 0.0168, 0.0177, 0.0132, 0.0152], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 13:06:17,105 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.234e+02 2.113e+02 2.491e+02 2.980e+02 5.007e+02, threshold=4.982e+02, percent-clipped=0.0 2022-12-08 13:06:18,141 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131830.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:06:34,216 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=131848.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:06:40,048 INFO [train.py:873] (3/4) Epoch 18, batch 3300, loss[loss=0.1068, simple_loss=0.1376, pruned_loss=0.03796, over 4957.00 frames. ], tot_loss[loss=0.1059, simple_loss=0.1419, pruned_loss=0.03495, over 1971267.42 frames. ], batch size: 100, lr: 4.37e-03, grad_scale: 8.0 2022-12-08 13:06:41,841 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131857.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:06:53,659 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.75 vs. limit=2.0 2022-12-08 13:07:00,062 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=131878.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:07:07,168 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.6556, 2.4502, 2.2508, 2.4150, 2.5468, 2.5822, 2.6198, 2.6002], device='cuda:3'), covar=tensor([0.1070, 0.0873, 0.2284, 0.2281, 0.1133, 0.1078, 0.1171, 0.0932], device='cuda:3'), in_proj_covar=tensor([0.0396, 0.0278, 0.0450, 0.0570, 0.0352, 0.0453, 0.0396, 0.0402], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 13:07:29,504 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9502, 1.4306, 2.9932, 1.7140, 3.1485, 2.9912, 2.2662, 3.2161], device='cuda:3'), covar=tensor([0.0272, 0.3052, 0.0449, 0.1904, 0.0393, 0.0495, 0.0976, 0.0285], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0154, 0.0159, 0.0166, 0.0168, 0.0177, 0.0133, 0.0153], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 13:07:34,751 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131918.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:07:44,549 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.221e+02 2.033e+02 2.408e+02 3.060e+02 5.712e+02, threshold=4.815e+02, percent-clipped=3.0 2022-12-08 13:08:03,357 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.1978, 4.9379, 4.6183, 4.8629, 4.8957, 5.1280, 5.2090, 5.2315], device='cuda:3'), covar=tensor([0.0927, 0.0408, 0.1994, 0.2547, 0.0730, 0.0723, 0.0719, 0.0665], device='cuda:3'), in_proj_covar=tensor([0.0396, 0.0278, 0.0450, 0.0570, 0.0352, 0.0453, 0.0394, 0.0401], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 13:08:06,597 INFO [train.py:873] (3/4) Epoch 18, batch 3400, loss[loss=0.1357, simple_loss=0.1399, pruned_loss=0.06571, over 1168.00 frames. ], tot_loss[loss=0.1056, simple_loss=0.1416, pruned_loss=0.03482, over 1957469.38 frames. ], batch size: 100, lr: 4.37e-03, grad_scale: 8.0 2022-12-08 13:08:30,407 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4755, 3.6838, 4.1757, 3.3335, 2.5157, 3.7205, 1.9492, 3.6238], device='cuda:3'), covar=tensor([0.1113, 0.0696, 0.0732, 0.1644, 0.2171, 0.0704, 0.3511, 0.1440], device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0104, 0.0096, 0.0102, 0.0117, 0.0093, 0.0119, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 13:08:51,105 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9120, 1.5470, 3.3944, 3.0860, 3.2682, 3.4226, 2.7299, 3.3886], device='cuda:3'), covar=tensor([0.1649, 0.1799, 0.0151, 0.0345, 0.0311, 0.0178, 0.0382, 0.0179], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0158, 0.0132, 0.0169, 0.0148, 0.0144, 0.0127, 0.0125], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 13:09:10,665 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.372e+02 2.058e+02 2.441e+02 2.998e+02 5.403e+02, threshold=4.883e+02, percent-clipped=4.0 2022-12-08 13:09:12,606 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132031.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:09:13,450 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8333, 1.4920, 3.3837, 3.0820, 3.2223, 3.4277, 2.7692, 3.3895], device='cuda:3'), covar=tensor([0.1611, 0.1719, 0.0161, 0.0325, 0.0304, 0.0178, 0.0338, 0.0176], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0158, 0.0132, 0.0170, 0.0148, 0.0144, 0.0127, 0.0125], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 13:09:32,499 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132054.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:09:33,284 INFO [train.py:873] (3/4) Epoch 18, batch 3500, loss[loss=0.1554, simple_loss=0.1414, pruned_loss=0.08471, over 1229.00 frames. ], tot_loss[loss=0.1058, simple_loss=0.1418, pruned_loss=0.03486, over 1989936.89 frames. ], batch size: 100, lr: 4.37e-03, grad_scale: 8.0 2022-12-08 13:09:42,104 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132065.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:09:47,865 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132072.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:10:05,413 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132092.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:10:15,984 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.2096, 1.4719, 1.7284, 1.6614, 1.6379, 1.6342, 1.4045, 1.2282], device='cuda:3'), covar=tensor([0.1242, 0.1364, 0.0490, 0.0614, 0.1172, 0.1002, 0.1846, 0.1657], device='cuda:3'), in_proj_covar=tensor([0.0138, 0.0091, 0.0071, 0.0077, 0.0100, 0.0091, 0.0102, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 13:10:23,479 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=132113.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:10:25,343 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132115.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 13:10:29,241 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=132120.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:10:37,536 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.237e+02 2.039e+02 2.680e+02 3.337e+02 6.531e+02, threshold=5.359e+02, percent-clipped=3.0 2022-12-08 13:10:54,100 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132148.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:10:59,846 INFO [train.py:873] (3/4) Epoch 18, batch 3600, loss[loss=0.1063, simple_loss=0.1446, pruned_loss=0.03396, over 14458.00 frames. ], tot_loss[loss=0.1052, simple_loss=0.1413, pruned_loss=0.0345, over 2011982.72 frames. ], batch size: 51, lr: 4.37e-03, grad_scale: 8.0 2022-12-08 13:11:35,841 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=132196.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:11:51,476 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=132213.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:12:05,413 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.022e+02 2.046e+02 2.523e+02 3.076e+02 9.449e+02, threshold=5.046e+02, percent-clipped=2.0 2022-12-08 13:12:28,036 INFO [train.py:873] (3/4) Epoch 18, batch 3700, loss[loss=0.1047, simple_loss=0.1396, pruned_loss=0.0349, over 10373.00 frames. ], tot_loss[loss=0.1053, simple_loss=0.1416, pruned_loss=0.03447, over 2062889.79 frames. ], batch size: 100, lr: 4.36e-03, grad_scale: 8.0 2022-12-08 13:13:18,124 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.41 vs. limit=5.0 2022-12-08 13:13:31,933 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2022-12-08 13:13:32,947 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.341e+02 2.108e+02 2.462e+02 3.226e+02 6.885e+02, threshold=4.924e+02, percent-clipped=3.0 2022-12-08 13:13:42,386 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5975, 1.6053, 2.8052, 2.1373, 2.6092, 1.7022, 2.2336, 2.5656], device='cuda:3'), covar=tensor([0.1674, 0.3933, 0.0936, 0.3084, 0.1479, 0.3195, 0.1336, 0.1255], device='cuda:3'), in_proj_covar=tensor([0.0257, 0.0202, 0.0222, 0.0273, 0.0240, 0.0204, 0.0203, 0.0224], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0005], device='cuda:3') 2022-12-08 13:13:43,539 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2022-12-08 13:13:54,143 INFO [train.py:873] (3/4) Epoch 18, batch 3800, loss[loss=0.1084, simple_loss=0.1269, pruned_loss=0.04498, over 3883.00 frames. ], tot_loss[loss=0.1067, simple_loss=0.1422, pruned_loss=0.03558, over 1972345.04 frames. ], batch size: 100, lr: 4.36e-03, grad_scale: 4.0 2022-12-08 13:14:16,386 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0655, 2.8087, 2.2132, 3.1519, 2.9663, 2.9920, 2.6366, 2.2683], device='cuda:3'), covar=tensor([0.0856, 0.1488, 0.3178, 0.0764, 0.1131, 0.1173, 0.1436, 0.2914], device='cuda:3'), in_proj_covar=tensor([0.0283, 0.0290, 0.0259, 0.0292, 0.0324, 0.0304, 0.0252, 0.0242], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 13:14:22,271 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=132387.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:14:34,939 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0 2022-12-08 13:14:42,727 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=132410.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 13:14:59,235 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.308e+02 2.016e+02 2.539e+02 3.051e+02 5.858e+02, threshold=5.078e+02, percent-clipped=3.0 2022-12-08 13:15:20,091 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2022-12-08 13:15:21,339 INFO [train.py:873] (3/4) Epoch 18, batch 3900, loss[loss=0.09519, simple_loss=0.14, pruned_loss=0.02517, over 14601.00 frames. ], tot_loss[loss=0.1055, simple_loss=0.1417, pruned_loss=0.03469, over 2023712.88 frames. ], batch size: 22, lr: 4.36e-03, grad_scale: 4.0 2022-12-08 13:16:12,131 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132513.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:16:26,818 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.188e+02 2.073e+02 2.605e+02 3.188e+02 7.778e+02, threshold=5.211e+02, percent-clipped=4.0 2022-12-08 13:16:48,239 INFO [train.py:873] (3/4) Epoch 18, batch 4000, loss[loss=0.1144, simple_loss=0.152, pruned_loss=0.03842, over 14261.00 frames. ], tot_loss[loss=0.1054, simple_loss=0.1412, pruned_loss=0.03485, over 1938273.15 frames. ], batch size: 39, lr: 4.36e-03, grad_scale: 8.0 2022-12-08 13:16:53,536 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=132561.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:17:25,999 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2022-12-08 13:17:35,983 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.63 vs. limit=2.0 2022-12-08 13:17:54,707 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.290e+02 2.137e+02 2.506e+02 3.319e+02 7.426e+02, threshold=5.012e+02, percent-clipped=3.0 2022-12-08 13:18:02,728 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8453, 1.9953, 2.7606, 2.1871, 2.6879, 2.5482, 2.5068, 2.3880], device='cuda:3'), covar=tensor([0.0883, 0.2802, 0.0869, 0.1689, 0.0583, 0.1091, 0.1114, 0.1501], device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0307, 0.0389, 0.0295, 0.0363, 0.0316, 0.0358, 0.0296], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 13:18:03,375 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9129, 4.7888, 4.4814, 4.8479, 4.4544, 4.1552, 4.9301, 4.6824], device='cuda:3'), covar=tensor([0.0500, 0.0591, 0.0710, 0.0531, 0.0632, 0.0669, 0.0491, 0.0636], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0145, 0.0148, 0.0163, 0.0148, 0.0124, 0.0170, 0.0151], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 13:18:16,275 INFO [train.py:873] (3/4) Epoch 18, batch 4100, loss[loss=0.1426, simple_loss=0.1609, pruned_loss=0.06209, over 5936.00 frames. ], tot_loss[loss=0.1054, simple_loss=0.141, pruned_loss=0.03488, over 1880788.11 frames. ], batch size: 100, lr: 4.36e-03, grad_scale: 8.0 2022-12-08 13:18:20,592 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132660.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:18:44,590 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132687.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:18:58,824 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4714, 3.3738, 2.8451, 3.0119, 3.3831, 3.4537, 3.5643, 3.4654], device='cuda:3'), covar=tensor([0.1203, 0.0816, 0.3158, 0.3360, 0.1287, 0.1341, 0.1107, 0.1400], device='cuda:3'), in_proj_covar=tensor([0.0393, 0.0275, 0.0450, 0.0570, 0.0350, 0.0451, 0.0390, 0.0399], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 13:19:05,270 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132710.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 13:19:14,359 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132721.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:19:22,530 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.248e+02 2.213e+02 2.659e+02 3.609e+02 5.635e+02, threshold=5.318e+02, percent-clipped=5.0 2022-12-08 13:19:27,157 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=132735.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:19:36,736 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132746.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:19:44,648 INFO [train.py:873] (3/4) Epoch 18, batch 4200, loss[loss=0.07527, simple_loss=0.1176, pruned_loss=0.01646, over 13449.00 frames. ], tot_loss[loss=0.1039, simple_loss=0.1401, pruned_loss=0.03388, over 1928170.10 frames. ], batch size: 17, lr: 4.36e-03, grad_scale: 8.0 2022-12-08 13:19:47,323 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=132758.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:20:30,868 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132807.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:20:48,436 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7906, 3.5612, 3.4296, 3.8177, 3.5502, 3.7586, 3.8793, 3.1618], device='cuda:3'), covar=tensor([0.0569, 0.1047, 0.0551, 0.0464, 0.0878, 0.0414, 0.0543, 0.0631], device='cuda:3'), in_proj_covar=tensor([0.0179, 0.0280, 0.0204, 0.0197, 0.0186, 0.0162, 0.0291, 0.0173], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 13:20:50,970 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.381e+02 2.134e+02 2.571e+02 3.230e+02 5.703e+02, threshold=5.142e+02, percent-clipped=2.0 2022-12-08 13:20:54,494 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3185, 2.5517, 2.6169, 2.6635, 2.2337, 2.6696, 2.5492, 1.5348], device='cuda:3'), covar=tensor([0.1081, 0.0842, 0.0721, 0.0606, 0.0879, 0.0613, 0.0889, 0.1821], device='cuda:3'), in_proj_covar=tensor([0.0140, 0.0092, 0.0071, 0.0078, 0.0101, 0.0092, 0.0103, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 13:20:57,489 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.00 vs. limit=2.0 2022-12-08 13:21:13,201 INFO [train.py:873] (3/4) Epoch 18, batch 4300, loss[loss=0.1075, simple_loss=0.1451, pruned_loss=0.03494, over 11178.00 frames. ], tot_loss[loss=0.1053, simple_loss=0.1411, pruned_loss=0.03472, over 1899417.56 frames. ], batch size: 100, lr: 4.35e-03, grad_scale: 4.0 2022-12-08 13:22:15,909 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.04 vs. limit=2.0 2022-12-08 13:22:20,489 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.080e+02 2.027e+02 2.464e+02 3.102e+02 5.439e+02, threshold=4.929e+02, percent-clipped=1.0 2022-12-08 13:22:25,680 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2022-12-08 13:22:41,308 INFO [train.py:873] (3/4) Epoch 18, batch 4400, loss[loss=0.1224, simple_loss=0.1549, pruned_loss=0.04495, over 8575.00 frames. ], tot_loss[loss=0.1053, simple_loss=0.1414, pruned_loss=0.03459, over 1926898.41 frames. ], batch size: 100, lr: 4.35e-03, grad_scale: 8.0 2022-12-08 13:23:01,054 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7616, 1.3140, 1.6979, 1.2255, 1.5345, 1.8589, 1.5129, 1.5557], device='cuda:3'), covar=tensor([0.0649, 0.0844, 0.0670, 0.0881, 0.1328, 0.0868, 0.0758, 0.1425], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0171, 0.0141, 0.0126, 0.0144, 0.0155, 0.0136, 0.0142], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0007, 0.0006, 0.0006], device='cuda:3') 2022-12-08 13:23:12,528 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8033, 1.6698, 1.9242, 1.6304, 1.9355, 1.7431, 1.5861, 1.7747], device='cuda:3'), covar=tensor([0.0826, 0.1910, 0.0556, 0.0584, 0.0626, 0.1124, 0.0465, 0.0509], device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0307, 0.0388, 0.0294, 0.0364, 0.0316, 0.0357, 0.0296], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 13:23:35,944 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=133016.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:23:48,921 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.037e+02 1.881e+02 2.190e+02 2.815e+02 5.316e+02, threshold=4.380e+02, percent-clipped=1.0 2022-12-08 13:24:10,913 INFO [train.py:873] (3/4) Epoch 18, batch 4500, loss[loss=0.08545, simple_loss=0.1311, pruned_loss=0.01991, over 14279.00 frames. ], tot_loss[loss=0.1034, simple_loss=0.1406, pruned_loss=0.03311, over 1992849.76 frames. ], batch size: 44, lr: 4.35e-03, grad_scale: 8.0 2022-12-08 13:24:52,393 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=133102.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:25:01,280 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2706, 2.2434, 1.7993, 2.3440, 2.2021, 2.2076, 2.1298, 1.9576], device='cuda:3'), covar=tensor([0.1098, 0.1100, 0.2009, 0.0922, 0.1220, 0.0859, 0.1564, 0.1541], device='cuda:3'), in_proj_covar=tensor([0.0281, 0.0288, 0.0258, 0.0290, 0.0321, 0.0303, 0.0253, 0.0241], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 13:25:17,476 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.181e+02 2.087e+02 2.404e+02 3.093e+02 4.922e+02, threshold=4.807e+02, percent-clipped=5.0 2022-12-08 13:25:38,356 INFO [train.py:873] (3/4) Epoch 18, batch 4600, loss[loss=0.1352, simple_loss=0.1569, pruned_loss=0.0567, over 7803.00 frames. ], tot_loss[loss=0.1055, simple_loss=0.1418, pruned_loss=0.0346, over 1962714.86 frames. ], batch size: 100, lr: 4.35e-03, grad_scale: 8.0 2022-12-08 13:25:38,468 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6224, 1.7702, 1.8129, 1.3262, 1.2356, 1.6522, 1.1947, 1.6455], device='cuda:3'), covar=tensor([0.1525, 0.1978, 0.0975, 0.2347, 0.2868, 0.1178, 0.2548, 0.1223], device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0103, 0.0096, 0.0101, 0.0116, 0.0092, 0.0118, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 13:25:50,876 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9041, 3.6910, 3.4403, 3.6303, 3.8335, 3.8393, 3.9154, 3.9030], device='cuda:3'), covar=tensor([0.0850, 0.0524, 0.2119, 0.2284, 0.0724, 0.0864, 0.0834, 0.0769], device='cuda:3'), in_proj_covar=tensor([0.0391, 0.0275, 0.0450, 0.0571, 0.0349, 0.0454, 0.0390, 0.0398], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 13:26:13,932 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1479, 1.1506, 1.2337, 1.0257, 1.0380, 0.9316, 1.1561, 1.0045], device='cuda:3'), covar=tensor([0.0270, 0.0246, 0.0205, 0.0284, 0.0280, 0.0392, 0.0294, 0.0373], device='cuda:3'), in_proj_covar=tensor([0.0023, 0.0023, 0.0021, 0.0022, 0.0022, 0.0034, 0.0029, 0.0033], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-08 13:26:44,620 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.241e+02 2.089e+02 2.539e+02 2.980e+02 5.650e+02, threshold=5.079e+02, percent-clipped=2.0 2022-12-08 13:26:50,247 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3253, 3.7649, 3.5393, 3.5015, 2.6974, 3.6390, 3.5082, 2.1354], device='cuda:3'), covar=tensor([0.1268, 0.0419, 0.0510, 0.0582, 0.0869, 0.0566, 0.0699, 0.1630], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0091, 0.0071, 0.0077, 0.0100, 0.0092, 0.0103, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 13:27:05,958 INFO [train.py:873] (3/4) Epoch 18, batch 4700, loss[loss=0.09826, simple_loss=0.144, pruned_loss=0.02628, over 14242.00 frames. ], tot_loss[loss=0.1049, simple_loss=0.1416, pruned_loss=0.03415, over 1975892.79 frames. ], batch size: 35, lr: 4.35e-03, grad_scale: 8.0 2022-12-08 13:27:06,884 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.5819, 5.1482, 5.0832, 5.5860, 5.3183, 4.7506, 5.5516, 4.6069], device='cuda:3'), covar=tensor([0.0311, 0.1138, 0.0396, 0.0397, 0.0666, 0.0453, 0.0523, 0.0502], device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0276, 0.0201, 0.0195, 0.0183, 0.0159, 0.0289, 0.0170], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 13:27:38,844 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.7318, 3.9817, 2.9294, 5.0517, 4.4608, 4.6496, 3.8707, 3.2478], device='cuda:3'), covar=tensor([0.0491, 0.1121, 0.3455, 0.0285, 0.0910, 0.1031, 0.1323, 0.2999], device='cuda:3'), in_proj_covar=tensor([0.0282, 0.0289, 0.0259, 0.0291, 0.0321, 0.0303, 0.0253, 0.0242], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 13:27:59,033 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=133316.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:28:12,073 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.321e+02 2.140e+02 2.537e+02 3.389e+02 1.353e+03, threshold=5.074e+02, percent-clipped=8.0 2022-12-08 13:28:32,821 INFO [train.py:873] (3/4) Epoch 18, batch 4800, loss[loss=0.1108, simple_loss=0.15, pruned_loss=0.0358, over 14362.00 frames. ], tot_loss[loss=0.105, simple_loss=0.1417, pruned_loss=0.03415, over 2044800.35 frames. ], batch size: 66, lr: 4.35e-03, grad_scale: 8.0 2022-12-08 13:28:40,336 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=133364.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:28:49,133 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=133374.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 13:28:51,176 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=133376.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:29:13,844 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=133402.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:29:38,999 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.315e+02 1.939e+02 2.430e+02 2.952e+02 7.648e+02, threshold=4.861e+02, percent-clipped=3.0 2022-12-08 13:29:42,987 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=133435.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 13:29:44,622 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=133437.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:29:55,367 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=133450.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:29:59,645 INFO [train.py:873] (3/4) Epoch 18, batch 4900, loss[loss=0.1548, simple_loss=0.1727, pruned_loss=0.06848, over 9516.00 frames. ], tot_loss[loss=0.1057, simple_loss=0.1423, pruned_loss=0.03457, over 2037857.88 frames. ], batch size: 100, lr: 4.34e-03, grad_scale: 8.0 2022-12-08 13:30:00,570 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3393, 4.0771, 3.9406, 4.3843, 4.1097, 3.9244, 4.4085, 3.7323], device='cuda:3'), covar=tensor([0.0433, 0.0838, 0.0442, 0.0406, 0.0687, 0.0928, 0.0522, 0.0499], device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0278, 0.0201, 0.0196, 0.0184, 0.0159, 0.0291, 0.0171], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 13:30:06,843 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 2022-12-08 13:30:16,311 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=133474.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:31:05,365 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.072e+02 2.069e+02 2.523e+02 3.509e+02 9.363e+02, threshold=5.045e+02, percent-clipped=9.0 2022-12-08 13:31:09,304 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=133535.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:31:26,065 INFO [train.py:873] (3/4) Epoch 18, batch 5000, loss[loss=0.1477, simple_loss=0.1383, pruned_loss=0.0785, over 1265.00 frames. ], tot_loss[loss=0.1055, simple_loss=0.1419, pruned_loss=0.0345, over 2038760.51 frames. ], batch size: 100, lr: 4.34e-03, grad_scale: 4.0 2022-12-08 13:31:36,565 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.50 vs. limit=2.0 2022-12-08 13:32:05,253 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1591, 1.1114, 1.0198, 1.1582, 1.2378, 0.7952, 1.0260, 1.1593], device='cuda:3'), covar=tensor([0.0588, 0.0765, 0.0573, 0.0563, 0.0525, 0.0680, 0.1006, 0.0758], device='cuda:3'), in_proj_covar=tensor([0.0036, 0.0035, 0.0040, 0.0034, 0.0034, 0.0048, 0.0036, 0.0038], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 13:32:27,246 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7749, 1.5999, 1.8285, 1.9020, 1.4286, 1.6567, 1.6549, 1.7662], device='cuda:3'), covar=tensor([0.0263, 0.0437, 0.0224, 0.0194, 0.0450, 0.0458, 0.0296, 0.0241], device='cuda:3'), in_proj_covar=tensor([0.0292, 0.0259, 0.0374, 0.0331, 0.0271, 0.0307, 0.0310, 0.0275], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 13:32:32,830 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.157e+02 2.004e+02 2.618e+02 3.128e+02 4.580e+02, threshold=5.235e+02, percent-clipped=0.0 2022-12-08 13:32:34,167 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.5883, 5.0775, 5.0512, 5.5161, 5.2250, 4.6788, 5.4677, 4.6321], device='cuda:3'), covar=tensor([0.0275, 0.0970, 0.0319, 0.0371, 0.0625, 0.0431, 0.0459, 0.0438], device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0275, 0.0200, 0.0194, 0.0183, 0.0158, 0.0289, 0.0170], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 13:32:50,783 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=133652.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:32:51,746 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7844, 1.6078, 1.8533, 1.9151, 1.4793, 1.6235, 1.7176, 1.7863], device='cuda:3'), covar=tensor([0.0284, 0.0435, 0.0261, 0.0239, 0.0463, 0.0564, 0.0338, 0.0267], device='cuda:3'), in_proj_covar=tensor([0.0292, 0.0259, 0.0373, 0.0330, 0.0271, 0.0306, 0.0310, 0.0275], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 13:32:53,177 INFO [train.py:873] (3/4) Epoch 18, batch 5100, loss[loss=0.09524, simple_loss=0.1389, pruned_loss=0.02578, over 14243.00 frames. ], tot_loss[loss=0.1044, simple_loss=0.1411, pruned_loss=0.03387, over 1979624.41 frames. ], batch size: 35, lr: 4.34e-03, grad_scale: 4.0 2022-12-08 13:33:39,822 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.2839, 2.1308, 4.2081, 3.0504, 4.1428, 1.9353, 3.2329, 4.0965], device='cuda:3'), covar=tensor([0.0590, 0.3976, 0.0523, 0.5614, 0.0668, 0.3388, 0.1412, 0.0587], device='cuda:3'), in_proj_covar=tensor([0.0255, 0.0199, 0.0221, 0.0269, 0.0239, 0.0202, 0.0202, 0.0220], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0005], device='cuda:3') 2022-12-08 13:33:44,286 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=133713.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:33:58,835 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=133730.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 13:34:00,754 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.056e+02 2.175e+02 2.651e+02 3.112e+02 1.000e+03, threshold=5.302e+02, percent-clipped=1.0 2022-12-08 13:34:00,907 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=133732.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:34:21,231 INFO [train.py:873] (3/4) Epoch 18, batch 5200, loss[loss=0.0924, simple_loss=0.134, pruned_loss=0.0254, over 14411.00 frames. ], tot_loss[loss=0.1048, simple_loss=0.1412, pruned_loss=0.03419, over 1930943.47 frames. ], batch size: 53, lr: 4.34e-03, grad_scale: 8.0 2022-12-08 13:35:05,969 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1650, 3.4903, 3.1919, 3.3160, 2.4744, 3.3645, 3.3573, 1.8706], device='cuda:3'), covar=tensor([0.1318, 0.0819, 0.1593, 0.0577, 0.1065, 0.0679, 0.0853, 0.1864], device='cuda:3'), in_proj_covar=tensor([0.0138, 0.0091, 0.0071, 0.0077, 0.0100, 0.0092, 0.0102, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 13:35:27,051 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=133830.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:35:29,271 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.226e+02 1.917e+02 2.403e+02 3.012e+02 6.543e+02, threshold=4.806e+02, percent-clipped=3.0 2022-12-08 13:35:48,360 INFO [train.py:873] (3/4) Epoch 18, batch 5300, loss[loss=0.09788, simple_loss=0.1368, pruned_loss=0.02946, over 14226.00 frames. ], tot_loss[loss=0.1047, simple_loss=0.1408, pruned_loss=0.03428, over 1912481.41 frames. ], batch size: 89, lr: 4.34e-03, grad_scale: 4.0 2022-12-08 13:36:21,457 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1010, 2.3839, 2.4367, 2.4525, 2.0836, 2.5252, 2.3947, 1.4809], device='cuda:3'), covar=tensor([0.0996, 0.0976, 0.0742, 0.0692, 0.0994, 0.0672, 0.0887, 0.1854], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0092, 0.0071, 0.0078, 0.0101, 0.0092, 0.0103, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 13:36:55,585 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.086e+02 2.076e+02 2.589e+02 3.060e+02 6.026e+02, threshold=5.178e+02, percent-clipped=1.0 2022-12-08 13:37:14,385 INFO [train.py:873] (3/4) Epoch 18, batch 5400, loss[loss=0.1229, simple_loss=0.124, pruned_loss=0.06095, over 1311.00 frames. ], tot_loss[loss=0.104, simple_loss=0.1406, pruned_loss=0.03371, over 1926010.44 frames. ], batch size: 100, lr: 4.34e-03, grad_scale: 4.0 2022-12-08 13:37:50,997 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=133997.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:38:00,648 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=134008.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:38:14,149 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2022-12-08 13:38:20,316 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134030.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 13:38:21,995 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134032.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:38:22,722 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.308e+02 1.858e+02 2.390e+02 3.308e+02 6.761e+02, threshold=4.781e+02, percent-clipped=5.0 2022-12-08 13:38:25,962 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2022-12-08 13:38:42,050 INFO [train.py:873] (3/4) Epoch 18, batch 5500, loss[loss=0.08931, simple_loss=0.1378, pruned_loss=0.02039, over 14179.00 frames. ], tot_loss[loss=0.1032, simple_loss=0.1403, pruned_loss=0.0331, over 1922360.39 frames. ], batch size: 35, lr: 4.33e-03, grad_scale: 4.0 2022-12-08 13:38:45,164 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134058.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:39:02,243 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=134078.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 13:39:03,903 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=134080.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:39:32,521 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0127, 2.0330, 2.0894, 2.0805, 1.9936, 1.5836, 1.3631, 1.8395], device='cuda:3'), covar=tensor([0.0754, 0.0605, 0.0475, 0.0444, 0.0527, 0.1513, 0.2293, 0.0503], device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0178, 0.0149, 0.0150, 0.0211, 0.0143, 0.0159, 0.0197], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 13:39:36,736 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9283, 1.6596, 3.7739, 3.4620, 3.5851, 3.8045, 3.2398, 3.8283], device='cuda:3'), covar=tensor([0.1749, 0.1689, 0.0142, 0.0291, 0.0281, 0.0176, 0.0270, 0.0146], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0157, 0.0130, 0.0168, 0.0147, 0.0143, 0.0125, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 13:39:47,435 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134130.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:39:50,149 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.270e+02 2.212e+02 2.722e+02 3.305e+02 6.043e+02, threshold=5.445e+02, percent-clipped=3.0 2022-12-08 13:40:09,464 INFO [train.py:873] (3/4) Epoch 18, batch 5600, loss[loss=0.0951, simple_loss=0.1322, pruned_loss=0.02901, over 11164.00 frames. ], tot_loss[loss=0.1038, simple_loss=0.1405, pruned_loss=0.03355, over 1950163.86 frames. ], batch size: 100, lr: 4.33e-03, grad_scale: 8.0 2022-12-08 13:40:19,466 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2022-12-08 13:40:29,749 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=134178.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:40:29,912 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0544, 1.8715, 2.1100, 1.9387, 2.1219, 1.7421, 1.6080, 1.3166], device='cuda:3'), covar=tensor([0.0253, 0.0388, 0.0257, 0.0291, 0.0184, 0.0298, 0.0279, 0.0474], device='cuda:3'), in_proj_covar=tensor([0.0023, 0.0023, 0.0021, 0.0022, 0.0021, 0.0034, 0.0028, 0.0033], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-08 13:40:35,058 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8748, 3.7326, 3.6620, 3.9715, 3.5632, 3.4594, 3.9112, 3.7850], device='cuda:3'), covar=tensor([0.0707, 0.0963, 0.0922, 0.0618, 0.0995, 0.0781, 0.0677, 0.0816], device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0148, 0.0152, 0.0166, 0.0151, 0.0127, 0.0173, 0.0153], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 13:41:18,750 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.315e+02 1.953e+02 2.384e+02 3.110e+02 5.351e+02, threshold=4.768e+02, percent-clipped=0.0 2022-12-08 13:41:36,149 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8620, 0.8551, 0.7292, 0.9009, 0.8687, 0.4622, 0.7971, 0.8993], device='cuda:3'), covar=tensor([0.0434, 0.0432, 0.0631, 0.0383, 0.0324, 0.0315, 0.0847, 0.0628], device='cuda:3'), in_proj_covar=tensor([0.0037, 0.0035, 0.0040, 0.0034, 0.0034, 0.0049, 0.0036, 0.0039], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2022-12-08 13:41:38,336 INFO [train.py:873] (3/4) Epoch 18, batch 5700, loss[loss=0.1615, simple_loss=0.1495, pruned_loss=0.08678, over 1226.00 frames. ], tot_loss[loss=0.1037, simple_loss=0.1406, pruned_loss=0.03341, over 1955238.43 frames. ], batch size: 100, lr: 4.33e-03, grad_scale: 8.0 2022-12-08 13:41:42,299 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.39 vs. limit=5.0 2022-12-08 13:42:21,773 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=134304.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:42:25,205 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134308.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:42:26,034 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=134309.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:42:27,658 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0666, 4.1903, 4.4033, 3.8728, 4.2294, 4.4446, 1.5237, 3.9913], device='cuda:3'), covar=tensor([0.0382, 0.0361, 0.0320, 0.0436, 0.0331, 0.0237, 0.3335, 0.0303], device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0178, 0.0149, 0.0151, 0.0211, 0.0143, 0.0159, 0.0198], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 13:42:35,863 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.67 vs. limit=5.0 2022-12-08 13:42:47,525 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.380e+02 2.136e+02 2.600e+02 3.296e+02 5.182e+02, threshold=5.200e+02, percent-clipped=3.0 2022-12-08 13:42:57,212 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.80 vs. limit=5.0 2022-12-08 13:43:05,204 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=134353.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:43:06,869 INFO [train.py:873] (3/4) Epoch 18, batch 5800, loss[loss=0.1026, simple_loss=0.1427, pruned_loss=0.03123, over 14363.00 frames. ], tot_loss[loss=0.1044, simple_loss=0.1411, pruned_loss=0.03383, over 1980639.43 frames. ], batch size: 41, lr: 4.33e-03, grad_scale: 8.0 2022-12-08 13:43:07,820 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=134356.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:43:16,168 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134365.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:43:20,531 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134370.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:44:17,489 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.465e+02 2.136e+02 2.508e+02 3.176e+02 6.308e+02, threshold=5.017e+02, percent-clipped=2.0 2022-12-08 13:44:37,063 INFO [train.py:873] (3/4) Epoch 18, batch 5900, loss[loss=0.1023, simple_loss=0.1446, pruned_loss=0.02996, over 14097.00 frames. ], tot_loss[loss=0.104, simple_loss=0.1412, pruned_loss=0.03345, over 2015103.22 frames. ], batch size: 29, lr: 4.33e-03, grad_scale: 8.0 2022-12-08 13:44:47,051 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.50 vs. limit=2.0 2022-12-08 13:45:24,082 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9180, 1.8247, 2.0467, 1.9591, 1.7935, 1.6714, 1.3190, 1.2428], device='cuda:3'), covar=tensor([0.0225, 0.0381, 0.0157, 0.0163, 0.0208, 0.0294, 0.0296, 0.0371], device='cuda:3'), in_proj_covar=tensor([0.0023, 0.0023, 0.0021, 0.0022, 0.0022, 0.0034, 0.0028, 0.0033], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-08 13:45:43,314 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9201, 1.6307, 3.4873, 3.2079, 3.3128, 3.5440, 2.7524, 3.5374], device='cuda:3'), covar=tensor([0.1613, 0.1654, 0.0142, 0.0339, 0.0311, 0.0174, 0.0465, 0.0163], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0158, 0.0131, 0.0169, 0.0148, 0.0144, 0.0127, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 13:45:45,582 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.224e+02 2.002e+02 2.502e+02 3.052e+02 5.386e+02, threshold=5.003e+02, percent-clipped=2.0 2022-12-08 13:46:04,786 INFO [train.py:873] (3/4) Epoch 18, batch 6000, loss[loss=0.1066, simple_loss=0.1484, pruned_loss=0.03245, over 14296.00 frames. ], tot_loss[loss=0.1033, simple_loss=0.1409, pruned_loss=0.03287, over 2081201.10 frames. ], batch size: 25, lr: 4.33e-03, grad_scale: 8.0 2022-12-08 13:46:04,786 INFO [train.py:896] (3/4) Computing validation loss 2022-12-08 13:46:17,880 INFO [train.py:905] (3/4) Epoch 18, validation: loss=0.1412, simple_loss=0.1783, pruned_loss=0.05203, over 857387.00 frames. 2022-12-08 13:46:17,881 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-08 13:47:26,072 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.309e+02 2.089e+02 2.569e+02 3.442e+02 1.078e+03, threshold=5.138e+02, percent-clipped=4.0 2022-12-08 13:47:41,492 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4758, 1.7187, 4.3250, 1.9392, 4.3414, 4.5874, 3.9431, 4.8665], device='cuda:3'), covar=tensor([0.0204, 0.3058, 0.0363, 0.2088, 0.0295, 0.0334, 0.0386, 0.0147], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0155, 0.0162, 0.0169, 0.0169, 0.0180, 0.0132, 0.0153], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 13:47:44,086 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134653.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:47:45,554 INFO [train.py:873] (3/4) Epoch 18, batch 6100, loss[loss=0.09665, simple_loss=0.143, pruned_loss=0.02517, over 14382.00 frames. ], tot_loss[loss=0.1033, simple_loss=0.1407, pruned_loss=0.03299, over 2032422.62 frames. ], batch size: 55, lr: 4.33e-03, grad_scale: 8.0 2022-12-08 13:47:49,952 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=134660.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:47:54,184 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=134665.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:48:25,738 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=134701.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:48:53,587 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.283e+02 2.174e+02 2.543e+02 3.201e+02 1.223e+03, threshold=5.086e+02, percent-clipped=2.0 2022-12-08 13:49:12,510 INFO [train.py:873] (3/4) Epoch 18, batch 6200, loss[loss=0.09968, simple_loss=0.1278, pruned_loss=0.03576, over 4968.00 frames. ], tot_loss[loss=0.1048, simple_loss=0.1416, pruned_loss=0.03403, over 2007398.03 frames. ], batch size: 100, lr: 4.32e-03, grad_scale: 8.0 2022-12-08 13:49:39,988 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.20 vs. limit=5.0 2022-12-08 13:50:17,461 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2022-12-08 13:50:20,437 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.216e+02 2.042e+02 2.445e+02 3.055e+02 6.497e+02, threshold=4.889e+02, percent-clipped=2.0 2022-12-08 13:50:40,211 INFO [train.py:873] (3/4) Epoch 18, batch 6300, loss[loss=0.1209, simple_loss=0.1577, pruned_loss=0.04206, over 13524.00 frames. ], tot_loss[loss=0.1053, simple_loss=0.1416, pruned_loss=0.03444, over 1962100.33 frames. ], batch size: 100, lr: 4.32e-03, grad_scale: 8.0 2022-12-08 13:50:46,260 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=134862.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:50:50,495 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5820, 2.3141, 2.8845, 1.7560, 1.9264, 2.5930, 1.4457, 2.6822], device='cuda:3'), covar=tensor([0.1024, 0.1552, 0.0685, 0.2398, 0.2233, 0.0894, 0.3078, 0.0804], device='cuda:3'), in_proj_covar=tensor([0.0085, 0.0101, 0.0094, 0.0099, 0.0114, 0.0091, 0.0115, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 13:51:25,959 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1499, 2.0075, 1.8194, 1.9351, 2.1055, 2.1348, 2.1265, 2.0753], device='cuda:3'), covar=tensor([0.1400, 0.1125, 0.2863, 0.2603, 0.1308, 0.1153, 0.1532, 0.1340], device='cuda:3'), in_proj_covar=tensor([0.0394, 0.0277, 0.0452, 0.0574, 0.0352, 0.0455, 0.0387, 0.0401], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 13:51:29,460 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9147, 3.3834, 3.0717, 3.1790, 2.4074, 3.3200, 3.0996, 1.8581], device='cuda:3'), covar=tensor([0.1152, 0.0429, 0.0845, 0.0525, 0.0981, 0.0481, 0.0815, 0.1772], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0091, 0.0071, 0.0078, 0.0102, 0.0092, 0.0104, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0006, 0.0007, 0.0006, 0.0007, 0.0006], device='cuda:3') 2022-12-08 13:51:39,597 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134923.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:51:47,984 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.134e+02 2.107e+02 2.599e+02 3.257e+02 7.432e+02, threshold=5.197e+02, percent-clipped=1.0 2022-12-08 13:51:57,908 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.03 vs. limit=5.0 2022-12-08 13:52:05,348 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2465, 1.2996, 3.3823, 1.5233, 3.2763, 3.4637, 2.5842, 3.6741], device='cuda:3'), covar=tensor([0.0412, 0.4579, 0.0693, 0.3093, 0.1011, 0.0575, 0.1044, 0.0393], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0156, 0.0163, 0.0169, 0.0170, 0.0181, 0.0134, 0.0154], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 13:52:07,217 INFO [train.py:873] (3/4) Epoch 18, batch 6400, loss[loss=0.1103, simple_loss=0.1128, pruned_loss=0.05392, over 1249.00 frames. ], tot_loss[loss=0.1046, simple_loss=0.1412, pruned_loss=0.03397, over 1967574.94 frames. ], batch size: 100, lr: 4.32e-03, grad_scale: 8.0 2022-12-08 13:52:12,155 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134960.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:52:16,348 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134965.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:52:19,275 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.52 vs. limit=5.0 2022-12-08 13:52:26,478 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8363, 3.4752, 4.1112, 3.0540, 2.5810, 3.5682, 2.0108, 3.7986], device='cuda:3'), covar=tensor([0.0544, 0.0707, 0.0437, 0.1270, 0.1669, 0.0831, 0.2534, 0.0553], device='cuda:3'), in_proj_covar=tensor([0.0086, 0.0102, 0.0095, 0.0099, 0.0115, 0.0092, 0.0116, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 13:52:57,659 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=135008.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:53:01,780 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=135013.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:53:18,809 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.337e+02 2.100e+02 2.565e+02 3.370e+02 5.862e+02, threshold=5.131e+02, percent-clipped=3.0 2022-12-08 13:53:38,284 INFO [train.py:873] (3/4) Epoch 18, batch 6500, loss[loss=0.09373, simple_loss=0.1332, pruned_loss=0.02713, over 13941.00 frames. ], tot_loss[loss=0.105, simple_loss=0.1416, pruned_loss=0.03419, over 1962452.17 frames. ], batch size: 23, lr: 4.32e-03, grad_scale: 8.0 2022-12-08 13:53:45,374 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9975, 1.6907, 1.9605, 1.8262, 1.8691, 1.6502, 1.6626, 1.1737], device='cuda:3'), covar=tensor([0.0234, 0.0426, 0.0284, 0.0284, 0.0253, 0.0319, 0.0270, 0.0465], device='cuda:3'), in_proj_covar=tensor([0.0024, 0.0023, 0.0021, 0.0023, 0.0022, 0.0035, 0.0029, 0.0033], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-08 13:54:15,937 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.30 vs. limit=5.0 2022-12-08 13:54:31,066 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0239, 1.9427, 2.1590, 1.3550, 1.5659, 2.0200, 1.2984, 2.0766], device='cuda:3'), covar=tensor([0.1016, 0.1814, 0.0849, 0.2558, 0.2778, 0.0989, 0.3027, 0.0924], device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0102, 0.0096, 0.0100, 0.0116, 0.0092, 0.0117, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 13:54:46,810 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.352e+02 2.038e+02 2.650e+02 3.158e+02 7.242e+02, threshold=5.300e+02, percent-clipped=3.0 2022-12-08 13:55:05,523 INFO [train.py:873] (3/4) Epoch 18, batch 6600, loss[loss=0.0999, simple_loss=0.1386, pruned_loss=0.03058, over 6932.00 frames. ], tot_loss[loss=0.1045, simple_loss=0.1412, pruned_loss=0.03389, over 1999414.09 frames. ], batch size: 100, lr: 4.32e-03, grad_scale: 8.0 2022-12-08 13:55:20,648 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=135171.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:55:39,070 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1081, 1.4837, 4.0647, 1.8167, 4.0513, 4.2571, 3.4786, 4.5196], device='cuda:3'), covar=tensor([0.0225, 0.3198, 0.0354, 0.2222, 0.0413, 0.0370, 0.0634, 0.0172], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0156, 0.0162, 0.0168, 0.0170, 0.0181, 0.0133, 0.0155], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 13:56:01,510 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=135218.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:56:13,693 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=135232.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 13:56:14,313 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.473e+02 2.150e+02 2.615e+02 3.167e+02 5.492e+02, threshold=5.230e+02, percent-clipped=1.0 2022-12-08 13:56:32,961 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7870, 2.8018, 2.6782, 2.8621, 2.5167, 2.7056, 2.8585, 2.7378], device='cuda:3'), covar=tensor([0.0876, 0.1180, 0.1066, 0.0793, 0.1270, 0.0716, 0.0902, 0.0961], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0149, 0.0152, 0.0167, 0.0152, 0.0128, 0.0174, 0.0154], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 13:56:34,472 INFO [train.py:873] (3/4) Epoch 18, batch 6700, loss[loss=0.1034, simple_loss=0.1139, pruned_loss=0.04646, over 2617.00 frames. ], tot_loss[loss=0.1044, simple_loss=0.1413, pruned_loss=0.03375, over 2005848.74 frames. ], batch size: 100, lr: 4.32e-03, grad_scale: 8.0 2022-12-08 13:57:07,627 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9200, 1.4757, 3.0257, 1.5177, 3.1494, 3.0452, 2.3092, 3.2363], device='cuda:3'), covar=tensor([0.0299, 0.2942, 0.0492, 0.2182, 0.0418, 0.0498, 0.0963, 0.0281], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0155, 0.0162, 0.0168, 0.0170, 0.0180, 0.0133, 0.0154], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 13:57:14,348 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2022-12-08 13:57:24,062 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1147, 1.9734, 4.6626, 4.2496, 4.1403, 4.7545, 4.1923, 4.8133], device='cuda:3'), covar=tensor([0.1567, 0.1536, 0.0099, 0.0221, 0.0264, 0.0136, 0.0176, 0.0106], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0157, 0.0130, 0.0168, 0.0147, 0.0142, 0.0126, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 13:57:28,960 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.0769, 4.8593, 4.5064, 4.7102, 4.7385, 4.9573, 5.0681, 5.0779], device='cuda:3'), covar=tensor([0.0697, 0.0465, 0.2177, 0.2443, 0.0709, 0.0830, 0.0711, 0.0691], device='cuda:3'), in_proj_covar=tensor([0.0392, 0.0277, 0.0451, 0.0575, 0.0353, 0.0455, 0.0386, 0.0400], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 13:57:41,809 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.031e+02 2.097e+02 2.445e+02 2.909e+02 5.288e+02, threshold=4.890e+02, percent-clipped=1.0 2022-12-08 13:58:00,101 INFO [train.py:873] (3/4) Epoch 18, batch 6800, loss[loss=0.0965, simple_loss=0.123, pruned_loss=0.03502, over 3927.00 frames. ], tot_loss[loss=0.1042, simple_loss=0.1412, pruned_loss=0.03363, over 2020448.92 frames. ], batch size: 100, lr: 4.31e-03, grad_scale: 8.0 2022-12-08 13:58:18,097 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.6587, 2.2285, 2.4975, 1.5054, 2.2840, 2.5238, 2.6474, 2.1618], device='cuda:3'), covar=tensor([0.0718, 0.0749, 0.0893, 0.1554, 0.1177, 0.0849, 0.0778, 0.1309], device='cuda:3'), in_proj_covar=tensor([0.0154, 0.0171, 0.0139, 0.0126, 0.0147, 0.0156, 0.0137, 0.0142], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0007, 0.0006, 0.0006], device='cuda:3') 2022-12-08 13:58:27,152 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6983, 1.6002, 1.6631, 1.5486, 1.6348, 1.5455, 1.3539, 1.2184], device='cuda:3'), covar=tensor([0.0191, 0.0269, 0.0228, 0.0303, 0.0187, 0.0321, 0.0218, 0.0385], device='cuda:3'), in_proj_covar=tensor([0.0024, 0.0024, 0.0021, 0.0023, 0.0022, 0.0035, 0.0029, 0.0033], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-08 13:58:37,991 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.7113, 1.8227, 4.6188, 2.4044, 4.3769, 4.8014, 4.4951, 5.2027], device='cuda:3'), covar=tensor([0.0212, 0.2918, 0.0400, 0.1780, 0.0328, 0.0361, 0.0263, 0.0137], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0155, 0.0162, 0.0167, 0.0170, 0.0180, 0.0133, 0.0154], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 13:59:01,599 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.10 vs. limit=2.0 2022-12-08 13:59:07,784 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.077e+02 2.213e+02 2.677e+02 3.514e+02 1.198e+03, threshold=5.353e+02, percent-clipped=7.0 2022-12-08 13:59:27,059 INFO [train.py:873] (3/4) Epoch 18, batch 6900, loss[loss=0.1138, simple_loss=0.149, pruned_loss=0.0393, over 14403.00 frames. ], tot_loss[loss=0.1056, simple_loss=0.1418, pruned_loss=0.0347, over 2012946.34 frames. ], batch size: 53, lr: 4.31e-03, grad_scale: 8.0 2022-12-08 13:59:59,184 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8138, 1.7985, 4.6538, 2.2534, 4.3285, 4.8143, 4.5947, 5.2058], device='cuda:3'), covar=tensor([0.0230, 0.3163, 0.0461, 0.2279, 0.0442, 0.0369, 0.0294, 0.0220], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0154, 0.0161, 0.0167, 0.0169, 0.0179, 0.0132, 0.0153], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 14:00:21,774 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=135518.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:00:29,530 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=135527.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:00:35,294 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.613e+02 2.207e+02 2.585e+02 3.410e+02 8.303e+02, threshold=5.169e+02, percent-clipped=6.0 2022-12-08 14:00:53,013 INFO [train.py:873] (3/4) Epoch 18, batch 7000, loss[loss=0.1005, simple_loss=0.1432, pruned_loss=0.02888, over 14355.00 frames. ], tot_loss[loss=0.1055, simple_loss=0.1416, pruned_loss=0.03467, over 1955339.07 frames. ], batch size: 73, lr: 4.31e-03, grad_scale: 4.0 2022-12-08 14:01:03,628 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=135566.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:01:10,192 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.72 vs. limit=2.0 2022-12-08 14:01:59,476 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=135630.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:02:02,085 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2997, 2.0165, 2.4432, 2.5280, 2.2222, 1.9420, 2.5170, 2.0126], device='cuda:3'), covar=tensor([0.0470, 0.1311, 0.0527, 0.0488, 0.0673, 0.1602, 0.0508, 0.0785], device='cuda:3'), in_proj_covar=tensor([0.0295, 0.0261, 0.0377, 0.0334, 0.0274, 0.0309, 0.0311, 0.0280], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 14:02:02,411 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.259e+02 2.039e+02 2.453e+02 3.005e+02 1.669e+03, threshold=4.906e+02, percent-clipped=2.0 2022-12-08 14:02:04,568 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7189, 3.5293, 3.3644, 2.5123, 3.1707, 3.4430, 3.6793, 3.0566], device='cuda:3'), covar=tensor([0.0577, 0.0920, 0.0732, 0.1144, 0.0863, 0.0644, 0.0802, 0.0894], device='cuda:3'), in_proj_covar=tensor([0.0154, 0.0170, 0.0140, 0.0126, 0.0146, 0.0156, 0.0138, 0.0142], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0007, 0.0006, 0.0006], device='cuda:3') 2022-12-08 14:02:17,558 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4256, 4.0340, 3.0595, 4.7969, 4.2784, 4.5849, 4.1008, 3.4266], device='cuda:3'), covar=tensor([0.0808, 0.1021, 0.3099, 0.0692, 0.0802, 0.2049, 0.0941, 0.2580], device='cuda:3'), in_proj_covar=tensor([0.0283, 0.0287, 0.0260, 0.0292, 0.0322, 0.0303, 0.0253, 0.0242], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 14:02:21,585 INFO [train.py:873] (3/4) Epoch 18, batch 7100, loss[loss=0.1025, simple_loss=0.1456, pruned_loss=0.02976, over 14234.00 frames. ], tot_loss[loss=0.1055, simple_loss=0.1415, pruned_loss=0.03475, over 1910110.52 frames. ], batch size: 60, lr: 4.31e-03, grad_scale: 4.0 2022-12-08 14:02:50,991 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2022-12-08 14:02:53,356 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=135691.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:03:06,775 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0714, 2.8824, 2.6212, 2.7980, 3.0003, 3.0487, 3.0410, 3.0589], device='cuda:3'), covar=tensor([0.0999, 0.0734, 0.2319, 0.2372, 0.0891, 0.0985, 0.1106, 0.0929], device='cuda:3'), in_proj_covar=tensor([0.0394, 0.0274, 0.0451, 0.0571, 0.0351, 0.0454, 0.0384, 0.0398], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 14:03:08,691 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.6822, 2.0895, 2.6240, 2.7591, 2.6353, 2.0749, 2.7459, 2.2461], device='cuda:3'), covar=tensor([0.0522, 0.1285, 0.0704, 0.0550, 0.0726, 0.1749, 0.0536, 0.0960], device='cuda:3'), in_proj_covar=tensor([0.0297, 0.0263, 0.0378, 0.0336, 0.0276, 0.0311, 0.0313, 0.0282], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 14:03:30,725 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.202e+02 1.933e+02 2.466e+02 3.185e+02 5.052e+02, threshold=4.932e+02, percent-clipped=2.0 2022-12-08 14:03:48,867 INFO [train.py:873] (3/4) Epoch 18, batch 7200, loss[loss=0.1394, simple_loss=0.1393, pruned_loss=0.06978, over 1228.00 frames. ], tot_loss[loss=0.105, simple_loss=0.1411, pruned_loss=0.0344, over 1867195.69 frames. ], batch size: 100, lr: 4.31e-03, grad_scale: 8.0 2022-12-08 14:04:02,741 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5672, 2.4158, 2.9512, 1.8578, 1.9703, 2.5999, 1.5021, 2.5747], device='cuda:3'), covar=tensor([0.0904, 0.1385, 0.0646, 0.2309, 0.2380, 0.1159, 0.3306, 0.0996], device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0103, 0.0096, 0.0100, 0.0116, 0.0092, 0.0117, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 14:04:08,114 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9249, 2.3818, 3.7960, 2.7698, 3.7741, 3.6813, 3.5705, 3.0641], device='cuda:3'), covar=tensor([0.0931, 0.3065, 0.1140, 0.1911, 0.0963, 0.1029, 0.1549, 0.1887], device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0308, 0.0391, 0.0298, 0.0364, 0.0322, 0.0361, 0.0295], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 14:04:50,232 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3344, 1.5224, 2.5173, 1.4791, 2.4946, 2.4930, 1.9812, 2.6543], device='cuda:3'), covar=tensor([0.0290, 0.2276, 0.0459, 0.1823, 0.0491, 0.0589, 0.1285, 0.0318], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0155, 0.0162, 0.0168, 0.0169, 0.0179, 0.0133, 0.0153], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 14:04:53,019 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=135827.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:04:58,958 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.262e+02 2.075e+02 2.523e+02 3.199e+02 6.288e+02, threshold=5.045e+02, percent-clipped=2.0 2022-12-08 14:05:18,067 INFO [train.py:873] (3/4) Epoch 18, batch 7300, loss[loss=0.1248, simple_loss=0.1375, pruned_loss=0.05604, over 2658.00 frames. ], tot_loss[loss=0.1045, simple_loss=0.1408, pruned_loss=0.03409, over 1960180.96 frames. ], batch size: 100, lr: 4.31e-03, grad_scale: 8.0 2022-12-08 14:05:31,345 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0679, 1.7323, 4.3444, 4.0451, 3.9556, 4.4586, 3.9133, 4.4262], device='cuda:3'), covar=tensor([0.1530, 0.1609, 0.0127, 0.0245, 0.0249, 0.0128, 0.0164, 0.0129], device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0155, 0.0130, 0.0167, 0.0146, 0.0142, 0.0125, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 14:05:35,237 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=135875.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:06:12,578 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0799, 4.1681, 4.4341, 3.6581, 4.3040, 4.4322, 1.6197, 3.9756], device='cuda:3'), covar=tensor([0.0384, 0.0403, 0.0371, 0.0750, 0.0325, 0.0280, 0.3164, 0.0322], device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0178, 0.0149, 0.0151, 0.0209, 0.0142, 0.0159, 0.0198], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 14:06:16,028 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2022-12-08 14:06:27,399 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.731e+01 2.137e+02 2.530e+02 3.235e+02 6.685e+02, threshold=5.060e+02, percent-clipped=2.0 2022-12-08 14:06:45,612 INFO [train.py:873] (3/4) Epoch 18, batch 7400, loss[loss=0.132, simple_loss=0.161, pruned_loss=0.05148, over 10360.00 frames. ], tot_loss[loss=0.1054, simple_loss=0.1413, pruned_loss=0.03477, over 1941910.38 frames. ], batch size: 100, lr: 4.30e-03, grad_scale: 8.0 2022-12-08 14:07:09,854 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=135982.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:07:13,198 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=135986.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:07:20,228 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=135994.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:07:41,671 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.7451, 1.0683, 1.2265, 1.1675, 0.9413, 1.2337, 1.0269, 0.7602], device='cuda:3'), covar=tensor([0.2259, 0.1163, 0.0643, 0.0745, 0.2083, 0.1339, 0.2119, 0.1643], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0091, 0.0071, 0.0078, 0.0102, 0.0093, 0.0103, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0006, 0.0007, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 14:07:45,985 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0272, 1.8183, 1.9477, 2.0468, 1.9300, 1.3097, 1.7195, 1.8368], device='cuda:3'), covar=tensor([0.0728, 0.0755, 0.0737, 0.0924, 0.0858, 0.0825, 0.0677, 0.0929], device='cuda:3'), in_proj_covar=tensor([0.0038, 0.0036, 0.0041, 0.0034, 0.0036, 0.0050, 0.0038, 0.0040], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-08 14:07:48,550 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1265, 4.2010, 4.4336, 3.8326, 4.3035, 4.4795, 1.5029, 4.0491], device='cuda:3'), covar=tensor([0.0287, 0.0336, 0.0315, 0.0561, 0.0285, 0.0194, 0.3168, 0.0278], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0177, 0.0148, 0.0150, 0.0208, 0.0141, 0.0158, 0.0197], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 14:07:55,412 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.302e+02 2.076e+02 2.483e+02 2.988e+02 6.746e+02, threshold=4.966e+02, percent-clipped=1.0 2022-12-08 14:08:04,434 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136043.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:08:14,774 INFO [train.py:873] (3/4) Epoch 18, batch 7500, loss[loss=0.08305, simple_loss=0.127, pruned_loss=0.01954, over 14286.00 frames. ], tot_loss[loss=0.1048, simple_loss=0.1413, pruned_loss=0.03418, over 1959989.54 frames. ], batch size: 31, lr: 4.30e-03, grad_scale: 8.0 2022-12-08 14:08:14,927 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136055.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:08:22,574 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1212, 3.5161, 3.2226, 3.3739, 2.6045, 3.5033, 3.3297, 1.8182], device='cuda:3'), covar=tensor([0.1080, 0.0488, 0.0844, 0.0547, 0.0828, 0.0469, 0.0746, 0.1883], device='cuda:3'), in_proj_covar=tensor([0.0138, 0.0090, 0.0071, 0.0077, 0.0102, 0.0092, 0.0102, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0006, 0.0007, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 14:08:47,500 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9058, 2.0999, 2.8188, 2.2879, 2.8480, 2.8150, 2.7232, 2.5322], device='cuda:3'), covar=tensor([0.0719, 0.2817, 0.0918, 0.1616, 0.0709, 0.0991, 0.1176, 0.1362], device='cuda:3'), in_proj_covar=tensor([0.0348, 0.0305, 0.0388, 0.0295, 0.0360, 0.0319, 0.0358, 0.0292], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 14:08:49,114 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.7400, 1.0547, 1.2595, 1.1766, 1.0164, 1.2607, 1.1237, 0.7599], device='cuda:3'), covar=tensor([0.2389, 0.1367, 0.0484, 0.0681, 0.1824, 0.1141, 0.1802, 0.1703], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0091, 0.0071, 0.0078, 0.0102, 0.0092, 0.0102, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0006, 0.0007, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 14:09:45,356 INFO [train.py:873] (3/4) Epoch 19, batch 0, loss[loss=0.1165, simple_loss=0.1577, pruned_loss=0.03766, over 14522.00 frames. ], tot_loss[loss=0.1165, simple_loss=0.1577, pruned_loss=0.03766, over 14522.00 frames. ], batch size: 51, lr: 4.19e-03, grad_scale: 8.0 2022-12-08 14:09:45,357 INFO [train.py:896] (3/4) Computing validation loss 2022-12-08 14:09:53,121 INFO [train.py:905] (3/4) Epoch 19, validation: loss=0.1445, simple_loss=0.1825, pruned_loss=0.05324, over 857387.00 frames. 2022-12-08 14:09:53,122 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-08 14:09:58,716 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 2022-12-08 14:10:08,830 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.063e+01 1.495e+02 2.536e+02 3.389e+02 8.096e+02, threshold=5.072e+02, percent-clipped=9.0 2022-12-08 14:10:16,258 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=136142.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:10:56,857 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1322, 4.0090, 3.8128, 4.1466, 3.7652, 3.5465, 4.2125, 3.9989], device='cuda:3'), covar=tensor([0.0626, 0.0836, 0.0860, 0.0643, 0.0920, 0.0757, 0.0641, 0.0838], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0148, 0.0151, 0.0167, 0.0152, 0.0127, 0.0175, 0.0155], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 14:11:10,077 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136203.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:11:22,947 INFO [train.py:873] (3/4) Epoch 19, batch 100, loss[loss=0.1222, simple_loss=0.1504, pruned_loss=0.04697, over 13528.00 frames. ], tot_loss[loss=0.1035, simple_loss=0.1411, pruned_loss=0.03295, over 869538.86 frames. ], batch size: 100, lr: 4.18e-03, grad_scale: 8.0 2022-12-08 14:11:37,511 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.402e+02 2.463e+02 2.960e+02 3.618e+02 1.073e+03, threshold=5.920e+02, percent-clipped=4.0 2022-12-08 14:12:23,029 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136286.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:12:50,170 INFO [train.py:873] (3/4) Epoch 19, batch 200, loss[loss=0.09273, simple_loss=0.1393, pruned_loss=0.02309, over 14538.00 frames. ], tot_loss[loss=0.1044, simple_loss=0.1411, pruned_loss=0.0338, over 1324351.40 frames. ], batch size: 43, lr: 4.18e-03, grad_scale: 4.0 2022-12-08 14:13:05,477 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=136334.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:13:06,200 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.191e+02 2.129e+02 2.646e+02 3.149e+02 5.242e+02, threshold=5.292e+02, percent-clipped=0.0 2022-12-08 14:13:06,377 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=136335.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:13:09,318 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136338.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:13:19,505 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136350.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:13:59,651 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136396.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:14:18,704 INFO [train.py:873] (3/4) Epoch 19, batch 300, loss[loss=0.1109, simple_loss=0.1512, pruned_loss=0.03526, over 14230.00 frames. ], tot_loss[loss=0.1052, simple_loss=0.1412, pruned_loss=0.03462, over 1561024.14 frames. ], batch size: 76, lr: 4.18e-03, grad_scale: 4.0 2022-12-08 14:14:33,863 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.272e+02 1.988e+02 2.460e+02 2.943e+02 5.876e+02, threshold=4.921e+02, percent-clipped=3.0 2022-12-08 14:14:40,961 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.4140, 1.3887, 1.3845, 1.5737, 1.6027, 1.1531, 1.2776, 1.3050], device='cuda:3'), covar=tensor([0.0720, 0.0630, 0.0615, 0.0701, 0.0448, 0.0914, 0.0829, 0.0779], device='cuda:3'), in_proj_covar=tensor([0.0038, 0.0036, 0.0041, 0.0034, 0.0035, 0.0050, 0.0037, 0.0040], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-08 14:14:59,287 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7776, 1.5077, 2.0024, 1.6003, 1.9214, 1.4448, 1.6293, 1.8703], device='cuda:3'), covar=tensor([0.3226, 0.3251, 0.0594, 0.1948, 0.1463, 0.1528, 0.1479, 0.0949], device='cuda:3'), in_proj_covar=tensor([0.0254, 0.0198, 0.0217, 0.0269, 0.0238, 0.0202, 0.0199, 0.0220], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0005], device='cuda:3') 2022-12-08 14:15:29,539 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136498.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:15:45,665 INFO [train.py:873] (3/4) Epoch 19, batch 400, loss[loss=0.1069, simple_loss=0.1163, pruned_loss=0.04871, over 2562.00 frames. ], tot_loss[loss=0.1046, simple_loss=0.1405, pruned_loss=0.03436, over 1640261.62 frames. ], batch size: 100, lr: 4.18e-03, grad_scale: 8.0 2022-12-08 14:16:01,299 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.382e+02 2.033e+02 2.479e+02 3.164e+02 1.103e+03, threshold=4.959e+02, percent-clipped=2.0 2022-12-08 14:17:14,358 INFO [train.py:873] (3/4) Epoch 19, batch 500, loss[loss=0.1019, simple_loss=0.1367, pruned_loss=0.03352, over 14283.00 frames. ], tot_loss[loss=0.1041, simple_loss=0.1405, pruned_loss=0.03391, over 1770262.46 frames. ], batch size: 63, lr: 4.18e-03, grad_scale: 8.0 2022-12-08 14:17:27,066 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.46 vs. limit=2.0 2022-12-08 14:17:30,014 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.075e+02 1.937e+02 2.457e+02 2.957e+02 6.738e+02, threshold=4.914e+02, percent-clipped=4.0 2022-12-08 14:17:32,428 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136638.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:17:36,842 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=136643.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:17:43,213 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136650.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:17:53,754 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9659, 2.4710, 3.2382, 2.1418, 2.1514, 3.0381, 1.7019, 2.9342], device='cuda:3'), covar=tensor([0.0831, 0.1184, 0.0748, 0.2385, 0.2203, 0.0791, 0.2980, 0.0846], device='cuda:3'), in_proj_covar=tensor([0.0088, 0.0103, 0.0097, 0.0101, 0.0117, 0.0093, 0.0118, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 14:17:54,944 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2311, 2.2570, 3.1340, 3.3372, 3.1943, 2.2277, 3.2206, 2.5415], device='cuda:3'), covar=tensor([0.0507, 0.1273, 0.0874, 0.0549, 0.0565, 0.1788, 0.0496, 0.1092], device='cuda:3'), in_proj_covar=tensor([0.0297, 0.0263, 0.0381, 0.0336, 0.0275, 0.0311, 0.0314, 0.0281], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 14:18:13,560 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=136686.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:18:18,056 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136691.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:18:24,493 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=136698.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:18:29,910 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136704.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:18:40,753 INFO [train.py:873] (3/4) Epoch 19, batch 600, loss[loss=0.08087, simple_loss=0.1232, pruned_loss=0.01925, over 13964.00 frames. ], tot_loss[loss=0.1053, simple_loss=0.1414, pruned_loss=0.03461, over 1844421.09 frames. ], batch size: 20, lr: 4.18e-03, grad_scale: 4.0 2022-12-08 14:18:57,079 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.012e+02 2.028e+02 2.404e+02 2.918e+02 4.938e+02, threshold=4.808e+02, percent-clipped=1.0 2022-12-08 14:19:00,811 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8466, 2.4916, 3.1673, 2.0964, 2.0410, 2.9272, 1.6633, 2.8409], device='cuda:3'), covar=tensor([0.0816, 0.1228, 0.0606, 0.2282, 0.2282, 0.0666, 0.2864, 0.0874], device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0103, 0.0097, 0.0101, 0.0117, 0.0093, 0.0117, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 14:19:40,393 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.2800, 2.7324, 4.2161, 3.0970, 4.1225, 3.9578, 3.9279, 3.5550], device='cuda:3'), covar=tensor([0.0843, 0.2660, 0.0911, 0.1651, 0.0682, 0.0981, 0.1153, 0.1555], device='cuda:3'), in_proj_covar=tensor([0.0348, 0.0305, 0.0386, 0.0294, 0.0358, 0.0318, 0.0356, 0.0292], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 14:19:51,208 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136798.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:19:58,231 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=136806.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:20:08,083 INFO [train.py:873] (3/4) Epoch 19, batch 700, loss[loss=0.1186, simple_loss=0.1367, pruned_loss=0.05026, over 3871.00 frames. ], tot_loss[loss=0.1055, simple_loss=0.1409, pruned_loss=0.03501, over 1809488.42 frames. ], batch size: 100, lr: 4.17e-03, grad_scale: 4.0 2022-12-08 14:20:24,140 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.847e+01 1.962e+02 2.462e+02 3.266e+02 7.518e+02, threshold=4.924e+02, percent-clipped=4.0 2022-12-08 14:20:25,971 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8926, 0.8089, 0.8724, 0.7953, 0.8457, 0.6175, 0.7128, 0.7429], device='cuda:3'), covar=tensor([0.0172, 0.0151, 0.0160, 0.0148, 0.0142, 0.0291, 0.0184, 0.0257], device='cuda:3'), in_proj_covar=tensor([0.0024, 0.0023, 0.0021, 0.0022, 0.0022, 0.0035, 0.0028, 0.0033], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-08 14:20:32,598 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=136846.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:20:51,848 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136867.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:21:35,622 INFO [train.py:873] (3/4) Epoch 19, batch 800, loss[loss=0.0892, simple_loss=0.1383, pruned_loss=0.02007, over 14098.00 frames. ], tot_loss[loss=0.1047, simple_loss=0.141, pruned_loss=0.03425, over 1915662.73 frames. ], batch size: 29, lr: 4.17e-03, grad_scale: 8.0 2022-12-08 14:21:48,426 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.60 vs. limit=2.0 2022-12-08 14:21:52,113 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.236e+02 2.241e+02 2.727e+02 3.173e+02 1.328e+03, threshold=5.453e+02, percent-clipped=3.0 2022-12-08 14:22:09,282 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8781, 1.7709, 1.8850, 1.7556, 1.6673, 1.6708, 1.6992, 1.3045], device='cuda:3'), covar=tensor([0.0146, 0.0208, 0.0156, 0.0157, 0.0185, 0.0249, 0.0191, 0.0335], device='cuda:3'), in_proj_covar=tensor([0.0024, 0.0023, 0.0021, 0.0022, 0.0022, 0.0035, 0.0029, 0.0033], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-08 14:22:40,626 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136991.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:22:47,271 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136999.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:22:48,500 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9995, 1.8453, 4.1991, 3.9226, 3.9233, 4.3287, 3.5988, 4.2977], device='cuda:3'), covar=tensor([0.1546, 0.1520, 0.0119, 0.0289, 0.0236, 0.0136, 0.0286, 0.0129], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0156, 0.0130, 0.0169, 0.0147, 0.0142, 0.0127, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 14:23:03,206 INFO [train.py:873] (3/4) Epoch 19, batch 900, loss[loss=0.1254, simple_loss=0.1545, pruned_loss=0.04813, over 14369.00 frames. ], tot_loss[loss=0.1047, simple_loss=0.1412, pruned_loss=0.03409, over 1945112.02 frames. ], batch size: 73, lr: 4.17e-03, grad_scale: 4.0 2022-12-08 14:23:20,504 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.232e+02 2.006e+02 2.396e+02 3.046e+02 6.535e+02, threshold=4.793e+02, percent-clipped=3.0 2022-12-08 14:23:22,046 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=137039.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:23:29,571 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.62 vs. limit=2.0 2022-12-08 14:23:41,638 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.3397, 4.8992, 4.7896, 5.3061, 5.0020, 4.5899, 5.2626, 4.3773], device='cuda:3'), covar=tensor([0.0339, 0.0934, 0.0366, 0.0420, 0.0754, 0.0445, 0.0510, 0.0504], device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0275, 0.0200, 0.0197, 0.0184, 0.0158, 0.0287, 0.0168], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 14:24:20,379 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5255, 3.3197, 3.2400, 3.5532, 3.3761, 3.5264, 3.5884, 3.0586], device='cuda:3'), covar=tensor([0.0617, 0.1073, 0.0556, 0.0504, 0.0751, 0.0414, 0.0612, 0.0586], device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0275, 0.0200, 0.0197, 0.0183, 0.0158, 0.0288, 0.0168], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 14:24:29,764 INFO [train.py:873] (3/4) Epoch 19, batch 1000, loss[loss=0.1252, simple_loss=0.1473, pruned_loss=0.05151, over 6917.00 frames. ], tot_loss[loss=0.1045, simple_loss=0.1411, pruned_loss=0.0339, over 1991622.31 frames. ], batch size: 100, lr: 4.17e-03, grad_scale: 4.0 2022-12-08 14:24:47,893 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.038e+02 2.060e+02 2.475e+02 3.449e+02 7.129e+02, threshold=4.949e+02, percent-clipped=6.0 2022-12-08 14:25:09,598 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=137162.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:25:47,223 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.6397, 1.6214, 1.6291, 1.3657, 1.4381, 1.2973, 1.4012, 1.1478], device='cuda:3'), covar=tensor([0.0175, 0.0215, 0.0170, 0.0203, 0.0182, 0.0314, 0.0210, 0.0341], device='cuda:3'), in_proj_covar=tensor([0.0024, 0.0023, 0.0021, 0.0022, 0.0022, 0.0034, 0.0029, 0.0033], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-08 14:25:57,947 INFO [train.py:873] (3/4) Epoch 19, batch 1100, loss[loss=0.1244, simple_loss=0.133, pruned_loss=0.05783, over 2602.00 frames. ], tot_loss[loss=0.1043, simple_loss=0.1409, pruned_loss=0.03383, over 2041654.35 frames. ], batch size: 100, lr: 4.17e-03, grad_scale: 4.0 2022-12-08 14:26:07,550 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2022-12-08 14:26:15,916 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.041e+02 1.998e+02 2.599e+02 3.181e+02 8.198e+02, threshold=5.197e+02, percent-clipped=8.0 2022-12-08 14:26:52,667 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2834, 2.5112, 2.5214, 2.5761, 2.1684, 2.5663, 2.4590, 1.5862], device='cuda:3'), covar=tensor([0.1116, 0.0909, 0.0609, 0.0568, 0.0928, 0.0528, 0.0923, 0.1788], device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0091, 0.0071, 0.0077, 0.0101, 0.0092, 0.0102, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 14:27:09,591 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 2022-12-08 14:27:10,336 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=137299.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:27:23,894 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.54 vs. limit=5.0 2022-12-08 14:27:25,931 INFO [train.py:873] (3/4) Epoch 19, batch 1200, loss[loss=0.09149, simple_loss=0.135, pruned_loss=0.024, over 14636.00 frames. ], tot_loss[loss=0.1047, simple_loss=0.141, pruned_loss=0.0342, over 2020365.77 frames. ], batch size: 33, lr: 4.17e-03, grad_scale: 8.0 2022-12-08 14:27:43,343 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.217e+02 2.257e+02 2.745e+02 3.542e+02 9.944e+02, threshold=5.490e+02, percent-clipped=8.0 2022-12-08 14:27:51,875 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=137347.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:28:13,714 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.9265, 5.8121, 5.3137, 5.8819, 5.5396, 5.2273, 5.9908, 5.7323], device='cuda:3'), covar=tensor([0.0532, 0.0608, 0.0798, 0.0523, 0.0683, 0.0478, 0.0461, 0.0581], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0150, 0.0152, 0.0167, 0.0154, 0.0129, 0.0176, 0.0156], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 14:28:54,281 INFO [train.py:873] (3/4) Epoch 19, batch 1300, loss[loss=0.15, simple_loss=0.1384, pruned_loss=0.08075, over 1245.00 frames. ], tot_loss[loss=0.1044, simple_loss=0.1408, pruned_loss=0.03399, over 1955595.06 frames. ], batch size: 100, lr: 4.17e-03, grad_scale: 8.0 2022-12-08 14:29:03,057 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.58 vs. limit=5.0 2022-12-08 14:29:10,398 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.2399, 1.1408, 1.2421, 1.0865, 1.0321, 0.9651, 1.0843, 1.0409], device='cuda:3'), covar=tensor([0.0253, 0.0266, 0.0247, 0.0259, 0.0296, 0.0410, 0.0277, 0.0389], device='cuda:3'), in_proj_covar=tensor([0.0024, 0.0024, 0.0021, 0.0023, 0.0022, 0.0035, 0.0029, 0.0033], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:3') 2022-12-08 14:29:12,653 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.353e+02 1.901e+02 2.314e+02 2.897e+02 7.672e+02, threshold=4.628e+02, percent-clipped=3.0 2022-12-08 14:29:34,052 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=137462.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:29:43,625 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8678, 3.6348, 3.4354, 3.5628, 3.7873, 3.8230, 3.8523, 3.8806], device='cuda:3'), covar=tensor([0.0882, 0.0616, 0.1943, 0.2474, 0.0729, 0.0829, 0.0831, 0.0787], device='cuda:3'), in_proj_covar=tensor([0.0399, 0.0281, 0.0459, 0.0578, 0.0356, 0.0463, 0.0395, 0.0406], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 14:29:47,068 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.5436, 5.1924, 4.8587, 5.1182, 5.0410, 5.4449, 5.5009, 5.5080], device='cuda:3'), covar=tensor([0.0600, 0.0447, 0.2013, 0.2793, 0.0707, 0.0692, 0.0970, 0.0684], device='cuda:3'), in_proj_covar=tensor([0.0399, 0.0281, 0.0459, 0.0578, 0.0356, 0.0463, 0.0395, 0.0407], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004], device='cuda:3') 2022-12-08 14:30:15,565 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=137510.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:30:21,589 INFO [train.py:873] (3/4) Epoch 19, batch 1400, loss[loss=0.1106, simple_loss=0.137, pruned_loss=0.04217, over 6933.00 frames. ], tot_loss[loss=0.1047, simple_loss=0.1411, pruned_loss=0.03413, over 1962073.94 frames. ], batch size: 100, lr: 4.16e-03, grad_scale: 4.0 2022-12-08 14:30:40,257 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.322e+02 2.087e+02 2.410e+02 2.920e+02 5.806e+02, threshold=4.820e+02, percent-clipped=3.0 2022-12-08 14:31:14,010 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.61 vs. limit=5.0 2022-12-08 14:31:30,126 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4717, 1.8645, 2.3951, 2.0681, 2.4222, 2.4502, 2.2627, 2.1933], device='cuda:3'), covar=tensor([0.0689, 0.2767, 0.0734, 0.1444, 0.0553, 0.1179, 0.0944, 0.1306], device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0311, 0.0389, 0.0298, 0.0364, 0.0322, 0.0362, 0.0295], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 14:31:35,634 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 2022-12-08 14:31:46,371 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.2439, 4.3021, 4.5666, 3.9318, 4.4044, 4.5692, 1.7560, 4.0884], device='cuda:3'), covar=tensor([0.0357, 0.0356, 0.0326, 0.0525, 0.0332, 0.0296, 0.3219, 0.0329], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0176, 0.0148, 0.0149, 0.0210, 0.0143, 0.0159, 0.0196], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 14:31:49,007 INFO [train.py:873] (3/4) Epoch 19, batch 1500, loss[loss=0.108, simple_loss=0.145, pruned_loss=0.03549, over 11204.00 frames. ], tot_loss[loss=0.1037, simple_loss=0.1404, pruned_loss=0.03352, over 2006712.58 frames. ], batch size: 100, lr: 4.16e-03, grad_scale: 4.0 2022-12-08 14:32:07,872 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.098e+02 1.988e+02 2.718e+02 3.293e+02 9.468e+02, threshold=5.435e+02, percent-clipped=6.0 2022-12-08 14:32:19,547 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.10 vs. limit=2.0 2022-12-08 14:33:18,037 INFO [train.py:873] (3/4) Epoch 19, batch 1600, loss[loss=0.1094, simple_loss=0.1464, pruned_loss=0.03618, over 14276.00 frames. ], tot_loss[loss=0.1023, simple_loss=0.1394, pruned_loss=0.03264, over 2045906.77 frames. ], batch size: 76, lr: 4.16e-03, grad_scale: 8.0 2022-12-08 14:33:36,639 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.275e+02 2.099e+02 2.503e+02 3.045e+02 5.451e+02, threshold=5.005e+02, percent-clipped=1.0 2022-12-08 14:34:36,885 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0761, 1.8538, 3.1736, 2.2786, 3.0552, 1.9099, 2.3923, 3.0864], device='cuda:3'), covar=tensor([0.0843, 0.3433, 0.0579, 0.4377, 0.0841, 0.2700, 0.1326, 0.0747], device='cuda:3'), in_proj_covar=tensor([0.0253, 0.0196, 0.0218, 0.0268, 0.0237, 0.0201, 0.0200, 0.0218], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0005], device='cuda:3') 2022-12-08 14:34:38,592 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1723, 2.0246, 2.0722, 2.3100, 2.1893, 1.4362, 2.0112, 2.2348], device='cuda:3'), covar=tensor([0.0955, 0.0682, 0.0778, 0.0859, 0.0593, 0.0819, 0.0878, 0.0449], device='cuda:3'), in_proj_covar=tensor([0.0038, 0.0036, 0.0041, 0.0034, 0.0036, 0.0050, 0.0038, 0.0041], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-08 14:34:46,400 INFO [train.py:873] (3/4) Epoch 19, batch 1700, loss[loss=0.1033, simple_loss=0.1395, pruned_loss=0.03355, over 14498.00 frames. ], tot_loss[loss=0.1035, simple_loss=0.1401, pruned_loss=0.03346, over 1979180.31 frames. ], batch size: 51, lr: 4.16e-03, grad_scale: 8.0 2022-12-08 14:35:05,300 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.215e+02 2.077e+02 2.557e+02 3.257e+02 5.458e+02, threshold=5.115e+02, percent-clipped=1.0 2022-12-08 14:35:52,974 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=137892.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:35:55,467 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=137895.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:35:59,497 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2022-12-08 14:36:14,639 INFO [train.py:873] (3/4) Epoch 19, batch 1800, loss[loss=0.0942, simple_loss=0.1407, pruned_loss=0.02383, over 14301.00 frames. ], tot_loss[loss=0.1028, simple_loss=0.1398, pruned_loss=0.03294, over 1974243.98 frames. ], batch size: 31, lr: 4.16e-03, grad_scale: 8.0 2022-12-08 14:36:33,146 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.219e+02 2.138e+02 2.501e+02 2.995e+02 6.998e+02, threshold=5.002e+02, percent-clipped=2.0 2022-12-08 14:36:42,805 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3915, 2.5645, 4.2676, 2.9300, 4.2059, 4.2389, 4.0599, 3.4807], device='cuda:3'), covar=tensor([0.0788, 0.3599, 0.0844, 0.2152, 0.0833, 0.0896, 0.1424, 0.2428], device='cuda:3'), in_proj_covar=tensor([0.0355, 0.0310, 0.0390, 0.0300, 0.0365, 0.0322, 0.0361, 0.0297], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 14:36:46,876 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=137953.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:36:49,273 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=137956.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:36:54,745 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8964, 2.0611, 2.7996, 2.2395, 2.8173, 2.7739, 2.6783, 2.4489], device='cuda:3'), covar=tensor([0.0871, 0.2851, 0.0968, 0.1716, 0.0696, 0.1161, 0.1078, 0.1773], device='cuda:3'), in_proj_covar=tensor([0.0355, 0.0310, 0.0391, 0.0300, 0.0366, 0.0323, 0.0362, 0.0297], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 14:37:19,746 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.33 vs. limit=5.0 2022-12-08 14:37:35,999 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3161, 4.2069, 4.0294, 4.3533, 4.0146, 3.7503, 4.4124, 4.1626], device='cuda:3'), covar=tensor([0.0650, 0.0846, 0.0857, 0.0629, 0.0780, 0.0673, 0.0578, 0.0820], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0148, 0.0151, 0.0166, 0.0151, 0.0128, 0.0174, 0.0153], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 14:37:42,666 INFO [train.py:873] (3/4) Epoch 19, batch 1900, loss[loss=0.1226, simple_loss=0.1433, pruned_loss=0.051, over 8608.00 frames. ], tot_loss[loss=0.1037, simple_loss=0.1401, pruned_loss=0.03364, over 1949290.33 frames. ], batch size: 100, lr: 4.16e-03, grad_scale: 8.0 2022-12-08 14:37:45,673 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1421, 2.4624, 2.4678, 2.5584, 2.1402, 2.5454, 2.4308, 1.4342], device='cuda:3'), covar=tensor([0.0934, 0.0800, 0.0886, 0.0541, 0.0976, 0.0662, 0.0938, 0.1950], device='cuda:3'), in_proj_covar=tensor([0.0137, 0.0091, 0.0071, 0.0078, 0.0101, 0.0093, 0.0102, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 14:38:01,168 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.215e+02 2.140e+02 2.687e+02 3.163e+02 5.638e+02, threshold=5.373e+02, percent-clipped=1.0 2022-12-08 14:38:16,779 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7875, 2.2981, 3.7479, 4.0479, 3.6312, 2.1968, 3.8807, 2.8774], device='cuda:3'), covar=tensor([0.0539, 0.1469, 0.1151, 0.0517, 0.0709, 0.2327, 0.0549, 0.1218], device='cuda:3'), in_proj_covar=tensor([0.0296, 0.0262, 0.0379, 0.0334, 0.0274, 0.0308, 0.0314, 0.0279], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 14:38:32,079 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1451, 2.0452, 2.1023, 2.1641, 2.0800, 2.0767, 2.2011, 1.9421], device='cuda:3'), covar=tensor([0.0842, 0.1194, 0.0696, 0.0792, 0.0968, 0.0706, 0.0862, 0.0659], device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0275, 0.0201, 0.0199, 0.0186, 0.0159, 0.0291, 0.0170], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 14:39:09,901 INFO [train.py:873] (3/4) Epoch 19, batch 2000, loss[loss=0.0964, simple_loss=0.1279, pruned_loss=0.03245, over 5991.00 frames. ], tot_loss[loss=0.1042, simple_loss=0.1409, pruned_loss=0.03374, over 2013170.12 frames. ], batch size: 100, lr: 4.15e-03, grad_scale: 8.0 2022-12-08 14:39:28,120 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.191e+02 1.998e+02 2.546e+02 3.326e+02 7.451e+02, threshold=5.092e+02, percent-clipped=3.0 2022-12-08 14:40:37,637 INFO [train.py:873] (3/4) Epoch 19, batch 2100, loss[loss=0.07931, simple_loss=0.1274, pruned_loss=0.01559, over 14391.00 frames. ], tot_loss[loss=0.1041, simple_loss=0.1407, pruned_loss=0.03371, over 1996104.20 frames. ], batch size: 41, lr: 4.15e-03, grad_scale: 8.0 2022-12-08 14:40:56,379 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.080e+02 1.995e+02 2.366e+02 2.965e+02 6.942e+02, threshold=4.733e+02, percent-clipped=2.0 2022-12-08 14:41:04,846 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=138248.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:41:07,338 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=138251.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:41:24,541 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=138270.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:42:05,014 INFO [train.py:873] (3/4) Epoch 19, batch 2200, loss[loss=0.1086, simple_loss=0.1265, pruned_loss=0.04534, over 3898.00 frames. ], tot_loss[loss=0.104, simple_loss=0.1407, pruned_loss=0.0336, over 1924687.21 frames. ], batch size: 100, lr: 4.15e-03, grad_scale: 8.0 2022-12-08 14:42:09,602 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=138322.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:42:15,477 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.4184, 5.1623, 4.8026, 4.9665, 5.0132, 5.2647, 5.3712, 5.3615], device='cuda:3'), covar=tensor([0.0593, 0.0436, 0.1985, 0.2864, 0.0692, 0.0748, 0.0730, 0.0736], device='cuda:3'), in_proj_covar=tensor([0.0400, 0.0278, 0.0454, 0.0576, 0.0360, 0.0466, 0.0391, 0.0405], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004], device='cuda:3') 2022-12-08 14:42:17,261 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=138331.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:42:22,959 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.147e+02 2.085e+02 2.687e+02 3.331e+02 7.282e+02, threshold=5.373e+02, percent-clipped=5.0 2022-12-08 14:43:01,571 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.23 vs. limit=5.0 2022-12-08 14:43:02,757 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=138383.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:43:25,627 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1348, 2.5151, 3.9686, 3.0372, 3.9753, 3.8153, 3.7322, 3.3246], device='cuda:3'), covar=tensor([0.0746, 0.3146, 0.1011, 0.1660, 0.0769, 0.1031, 0.1433, 0.1810], device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0308, 0.0389, 0.0297, 0.0362, 0.0321, 0.0361, 0.0296], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 14:43:32,133 INFO [train.py:873] (3/4) Epoch 19, batch 2300, loss[loss=0.1069, simple_loss=0.1365, pruned_loss=0.03868, over 6908.00 frames. ], tot_loss[loss=0.1032, simple_loss=0.1401, pruned_loss=0.03316, over 1929314.53 frames. ], batch size: 100, lr: 4.15e-03, grad_scale: 8.0 2022-12-08 14:43:39,057 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0164, 2.0152, 4.0167, 2.8463, 3.8923, 1.9613, 3.0283, 3.9164], device='cuda:3'), covar=tensor([0.0657, 0.4005, 0.0466, 0.4498, 0.0637, 0.3134, 0.1299, 0.0509], device='cuda:3'), in_proj_covar=tensor([0.0254, 0.0197, 0.0218, 0.0265, 0.0237, 0.0201, 0.0200, 0.0219], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0005], device='cuda:3') 2022-12-08 14:43:50,854 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.988e+01 2.098e+02 2.453e+02 3.240e+02 6.762e+02, threshold=4.906e+02, percent-clipped=1.0 2022-12-08 14:44:06,994 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8623, 1.8248, 4.2414, 3.8929, 3.9210, 4.3295, 3.6973, 4.3003], device='cuda:3'), covar=tensor([0.1659, 0.1506, 0.0104, 0.0234, 0.0235, 0.0130, 0.0249, 0.0114], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0155, 0.0130, 0.0169, 0.0147, 0.0142, 0.0125, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 14:45:01,197 INFO [train.py:873] (3/4) Epoch 19, batch 2400, loss[loss=0.1039, simple_loss=0.1116, pruned_loss=0.04813, over 1243.00 frames. ], tot_loss[loss=0.103, simple_loss=0.1403, pruned_loss=0.03284, over 1990451.58 frames. ], batch size: 100, lr: 4.15e-03, grad_scale: 8.0 2022-12-08 14:45:14,345 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.73 vs. limit=5.0 2022-12-08 14:45:18,700 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.048e+02 2.033e+02 2.636e+02 3.505e+02 6.447e+02, threshold=5.271e+02, percent-clipped=6.0 2022-12-08 14:45:28,153 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=138548.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:45:30,909 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=138551.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:46:09,979 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=138596.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:46:12,447 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=138599.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:46:28,222 INFO [train.py:873] (3/4) Epoch 19, batch 2500, loss[loss=0.108, simple_loss=0.1408, pruned_loss=0.03762, over 6009.00 frames. ], tot_loss[loss=0.1034, simple_loss=0.1406, pruned_loss=0.03311, over 1976289.01 frames. ], batch size: 100, lr: 4.15e-03, grad_scale: 4.0 2022-12-08 14:46:36,668 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=138626.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:46:47,813 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.473e+02 2.139e+02 2.554e+02 3.094e+02 5.985e+02, threshold=5.107e+02, percent-clipped=2.0 2022-12-08 14:46:53,466 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9238, 1.4896, 3.3939, 3.0986, 3.1966, 3.4109, 2.8253, 3.4070], device='cuda:3'), covar=tensor([0.1521, 0.1714, 0.0157, 0.0345, 0.0344, 0.0190, 0.0339, 0.0168], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0154, 0.0130, 0.0168, 0.0146, 0.0141, 0.0124, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 14:47:00,702 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8455, 0.8605, 0.7146, 0.8649, 0.8716, 0.3771, 0.8144, 0.8437], device='cuda:3'), covar=tensor([0.0456, 0.0514, 0.0478, 0.0489, 0.0391, 0.0330, 0.0735, 0.0812], device='cuda:3'), in_proj_covar=tensor([0.0038, 0.0036, 0.0041, 0.0034, 0.0036, 0.0050, 0.0038, 0.0040], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-08 14:47:21,907 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=138678.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:47:24,879 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=138681.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:47:29,049 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=138686.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:47:38,363 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=138697.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:47:55,443 INFO [train.py:873] (3/4) Epoch 19, batch 2600, loss[loss=0.103, simple_loss=0.1433, pruned_loss=0.03134, over 13927.00 frames. ], tot_loss[loss=0.1035, simple_loss=0.1404, pruned_loss=0.03329, over 2021056.61 frames. ], batch size: 23, lr: 4.15e-03, grad_scale: 2.0 2022-12-08 14:48:15,369 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.052e+02 2.043e+02 2.512e+02 3.155e+02 8.164e+02, threshold=5.025e+02, percent-clipped=4.0 2022-12-08 14:48:17,491 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=138742.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:48:21,893 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=138747.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:48:31,386 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=138758.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:49:14,391 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.6912, 2.5805, 3.1490, 2.1536, 2.0492, 2.6647, 1.5053, 2.7707], device='cuda:3'), covar=tensor([0.0935, 0.1074, 0.0728, 0.1962, 0.2224, 0.1026, 0.3465, 0.1072], device='cuda:3'), in_proj_covar=tensor([0.0088, 0.0104, 0.0097, 0.0101, 0.0116, 0.0092, 0.0118, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 14:49:23,532 INFO [train.py:873] (3/4) Epoch 19, batch 2700, loss[loss=0.1003, simple_loss=0.1423, pruned_loss=0.02916, over 14212.00 frames. ], tot_loss[loss=0.1034, simple_loss=0.1402, pruned_loss=0.03327, over 2016291.37 frames. ], batch size: 46, lr: 4.14e-03, grad_scale: 2.0 2022-12-08 14:49:41,057 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2022-12-08 14:49:43,049 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8013, 1.9740, 2.7566, 2.2250, 2.7049, 2.6790, 2.5207, 2.3949], device='cuda:3'), covar=tensor([0.0888, 0.3160, 0.0952, 0.1713, 0.0704, 0.1186, 0.1156, 0.1605], device='cuda:3'), in_proj_covar=tensor([0.0353, 0.0311, 0.0390, 0.0299, 0.0364, 0.0323, 0.0362, 0.0298], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 14:49:43,636 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.077e+02 2.157e+02 2.518e+02 3.000e+02 5.718e+02, threshold=5.036e+02, percent-clipped=2.0 2022-12-08 14:50:44,346 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=138909.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:50:51,208 INFO [train.py:873] (3/4) Epoch 19, batch 2800, loss[loss=0.1082, simple_loss=0.1464, pruned_loss=0.03503, over 14350.00 frames. ], tot_loss[loss=0.1041, simple_loss=0.1405, pruned_loss=0.03382, over 1987900.63 frames. ], batch size: 55, lr: 4.14e-03, grad_scale: 4.0 2022-12-08 14:50:59,263 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=138926.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:51:10,962 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.214e+02 2.185e+02 2.764e+02 3.570e+02 5.389e+02, threshold=5.528e+02, percent-clipped=1.0 2022-12-08 14:51:21,913 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.75 vs. limit=5.0 2022-12-08 14:51:37,426 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=138970.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 14:51:40,722 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=138974.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:51:44,283 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=138978.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:52:12,963 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.32 vs. limit=5.0 2022-12-08 14:52:15,100 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4189, 1.6827, 2.6029, 2.0928, 2.4789, 1.6983, 2.1921, 2.4719], device='cuda:3'), covar=tensor([0.2031, 0.3774, 0.0853, 0.3022, 0.1678, 0.2705, 0.1097, 0.1032], device='cuda:3'), in_proj_covar=tensor([0.0252, 0.0196, 0.0217, 0.0266, 0.0239, 0.0200, 0.0200, 0.0219], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0005], device='cuda:3') 2022-12-08 14:52:18,362 INFO [train.py:873] (3/4) Epoch 19, batch 2900, loss[loss=0.0928, simple_loss=0.1369, pruned_loss=0.02435, over 11281.00 frames. ], tot_loss[loss=0.1032, simple_loss=0.1401, pruned_loss=0.03315, over 2013875.81 frames. ], batch size: 14, lr: 4.14e-03, grad_scale: 4.0 2022-12-08 14:52:25,005 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2022-12-08 14:52:26,199 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=139026.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:52:35,629 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139037.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:52:38,099 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.182e+02 1.948e+02 2.422e+02 2.806e+02 5.439e+02, threshold=4.845e+02, percent-clipped=0.0 2022-12-08 14:52:39,981 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139042.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:52:49,567 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139053.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:53:24,803 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.78 vs. limit=2.0 2022-12-08 14:53:45,726 INFO [train.py:873] (3/4) Epoch 19, batch 3000, loss[loss=0.08975, simple_loss=0.1265, pruned_loss=0.02649, over 14128.00 frames. ], tot_loss[loss=0.1021, simple_loss=0.1394, pruned_loss=0.03243, over 2021986.87 frames. ], batch size: 19, lr: 4.14e-03, grad_scale: 4.0 2022-12-08 14:53:45,727 INFO [train.py:896] (3/4) Computing validation loss 2022-12-08 14:53:52,036 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2092, 2.2863, 4.7332, 4.5516, 4.2400, 4.7841, 4.4352, 4.8455], device='cuda:3'), covar=tensor([0.1453, 0.1287, 0.0095, 0.0146, 0.0212, 0.0102, 0.0124, 0.0079], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0156, 0.0131, 0.0169, 0.0148, 0.0143, 0.0126, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 14:53:54,218 INFO [train.py:905] (3/4) Epoch 19, validation: loss=0.142, simple_loss=0.1782, pruned_loss=0.05288, over 857387.00 frames. 2022-12-08 14:53:54,219 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-08 14:54:08,207 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.67 vs. limit=5.0 2022-12-08 14:54:14,226 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.279e+02 2.026e+02 2.406e+02 3.345e+02 1.533e+03, threshold=4.812e+02, percent-clipped=5.0 2022-12-08 14:55:22,156 INFO [train.py:873] (3/4) Epoch 19, batch 3100, loss[loss=0.124, simple_loss=0.1522, pruned_loss=0.04791, over 8650.00 frames. ], tot_loss[loss=0.1037, simple_loss=0.1405, pruned_loss=0.03345, over 2017694.06 frames. ], batch size: 100, lr: 4.14e-03, grad_scale: 4.0 2022-12-08 14:55:40,047 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139238.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:55:41,531 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.046e+02 2.036e+02 2.564e+02 2.982e+02 6.918e+02, threshold=5.128e+02, percent-clipped=2.0 2022-12-08 14:55:54,996 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5440, 4.4356, 4.1603, 4.5313, 4.1456, 3.8276, 4.5776, 4.3545], device='cuda:3'), covar=tensor([0.0592, 0.0795, 0.0848, 0.0606, 0.0761, 0.0705, 0.0599, 0.0789], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0149, 0.0150, 0.0166, 0.0153, 0.0126, 0.0174, 0.0153], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 14:56:01,994 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3141, 2.1432, 3.2000, 2.4991, 3.2180, 3.0722, 2.9907, 2.6488], device='cuda:3'), covar=tensor([0.0908, 0.3105, 0.0974, 0.1875, 0.0846, 0.1219, 0.1278, 0.1810], device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0307, 0.0387, 0.0295, 0.0361, 0.0320, 0.0359, 0.0295], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 14:56:03,657 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139265.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 14:56:06,456 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 2022-12-08 14:56:12,160 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8363, 3.6063, 3.2560, 2.6116, 3.2946, 3.5004, 3.9302, 3.2095], device='cuda:3'), covar=tensor([0.0482, 0.0824, 0.0783, 0.1092, 0.0652, 0.0715, 0.0514, 0.0870], device='cuda:3'), in_proj_covar=tensor([0.0154, 0.0168, 0.0140, 0.0125, 0.0144, 0.0155, 0.0138, 0.0142], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0007, 0.0006, 0.0006], device='cuda:3') 2022-12-08 14:56:13,038 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139276.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:56:24,412 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139289.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:56:33,166 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139299.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:56:48,455 INFO [train.py:873] (3/4) Epoch 19, batch 3200, loss[loss=0.1044, simple_loss=0.1274, pruned_loss=0.04069, over 5020.00 frames. ], tot_loss[loss=0.1042, simple_loss=0.1408, pruned_loss=0.03382, over 1984673.30 frames. ], batch size: 100, lr: 4.14e-03, grad_scale: 8.0 2022-12-08 14:56:57,548 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4856, 2.7940, 2.7220, 2.8475, 2.3050, 2.8196, 2.7029, 1.6438], device='cuda:3'), covar=tensor([0.1011, 0.0742, 0.0714, 0.0519, 0.0954, 0.0633, 0.0937, 0.1860], device='cuda:3'), in_proj_covar=tensor([0.0137, 0.0092, 0.0072, 0.0077, 0.0101, 0.0093, 0.0103, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0006, 0.0006, 0.0007, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 14:57:04,133 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.60 vs. limit=5.0 2022-12-08 14:57:06,635 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139337.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:57:06,697 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139337.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 14:57:08,977 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.774e+01 1.930e+02 2.439e+02 3.004e+02 8.294e+02, threshold=4.879e+02, percent-clipped=2.0 2022-12-08 14:57:10,927 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139342.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:57:17,635 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139350.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:57:20,021 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139353.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:57:20,112 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1236, 2.0931, 3.1738, 3.3432, 3.2472, 2.1503, 3.2186, 2.2835], device='cuda:3'), covar=tensor([0.0565, 0.1425, 0.1074, 0.0598, 0.0593, 0.2074, 0.0482, 0.1330], device='cuda:3'), in_proj_covar=tensor([0.0295, 0.0263, 0.0378, 0.0335, 0.0275, 0.0309, 0.0314, 0.0280], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 14:57:41,228 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8984, 1.5806, 1.9560, 1.6561, 1.9805, 1.7942, 1.5894, 1.8361], device='cuda:3'), covar=tensor([0.0651, 0.1434, 0.0506, 0.0447, 0.0525, 0.0944, 0.0321, 0.0389], device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0308, 0.0387, 0.0295, 0.0362, 0.0320, 0.0359, 0.0295], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 14:57:47,844 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=139385.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:57:52,101 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=139390.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:58:01,925 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=139401.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 14:58:16,251 INFO [train.py:873] (3/4) Epoch 19, batch 3300, loss[loss=0.1284, simple_loss=0.1553, pruned_loss=0.05078, over 8560.00 frames. ], tot_loss[loss=0.1039, simple_loss=0.1406, pruned_loss=0.03361, over 2031044.56 frames. ], batch size: 100, lr: 4.14e-03, grad_scale: 8.0 2022-12-08 14:58:35,036 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.212e+02 2.023e+02 2.414e+02 3.086e+02 5.603e+02, threshold=4.828e+02, percent-clipped=3.0 2022-12-08 14:58:44,369 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4869, 1.7274, 2.5952, 2.1377, 2.5540, 1.7152, 2.1972, 2.5457], device='cuda:3'), covar=tensor([0.1835, 0.3945, 0.0979, 0.2729, 0.1289, 0.2704, 0.1031, 0.1015], device='cuda:3'), in_proj_covar=tensor([0.0250, 0.0195, 0.0218, 0.0266, 0.0239, 0.0200, 0.0198, 0.0219], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0005], device='cuda:3') 2022-12-08 14:59:41,141 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8535, 1.9796, 4.5690, 2.5032, 4.4516, 4.9756, 4.5741, 5.3014], device='cuda:3'), covar=tensor([0.0184, 0.2799, 0.0359, 0.1852, 0.0284, 0.0328, 0.0283, 0.0149], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0154, 0.0160, 0.0168, 0.0168, 0.0178, 0.0133, 0.0152], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 14:59:41,884 INFO [train.py:873] (3/4) Epoch 19, batch 3400, loss[loss=0.1003, simple_loss=0.1288, pruned_loss=0.03588, over 14576.00 frames. ], tot_loss[loss=0.1034, simple_loss=0.1403, pruned_loss=0.03327, over 1978746.26 frames. ], batch size: 21, lr: 4.13e-03, grad_scale: 8.0 2022-12-08 15:00:02,110 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.225e+02 2.047e+02 2.601e+02 3.313e+02 6.688e+02, threshold=5.202e+02, percent-clipped=6.0 2022-12-08 15:00:24,226 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139565.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 15:00:29,836 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139571.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:00:39,085 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139582.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:00:49,366 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139594.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:01:06,699 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=139613.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:01:10,080 INFO [train.py:873] (3/4) Epoch 19, batch 3500, loss[loss=0.09346, simple_loss=0.1215, pruned_loss=0.03274, over 4979.00 frames. ], tot_loss[loss=0.1028, simple_loss=0.1401, pruned_loss=0.03276, over 1987136.92 frames. ], batch size: 100, lr: 4.13e-03, grad_scale: 8.0 2022-12-08 15:01:23,106 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139632.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 15:01:23,185 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139632.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:01:30,130 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.236e+02 2.003e+02 2.440e+02 2.964e+02 5.520e+02, threshold=4.880e+02, percent-clipped=2.0 2022-12-08 15:01:33,247 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139643.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:01:34,835 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139645.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:01:51,519 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1689, 1.4849, 3.2885, 1.6377, 3.0375, 3.2899, 2.4009, 3.5010], device='cuda:3'), covar=tensor([0.0273, 0.3006, 0.0383, 0.2262, 0.1287, 0.0445, 0.0945, 0.0235], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0155, 0.0161, 0.0169, 0.0169, 0.0179, 0.0133, 0.0153], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 15:02:01,854 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7000, 2.4043, 2.5631, 1.7415, 2.2432, 2.4879, 2.7152, 2.2899], device='cuda:3'), covar=tensor([0.0879, 0.0772, 0.0923, 0.1322, 0.1249, 0.0866, 0.0711, 0.1282], device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0169, 0.0141, 0.0126, 0.0145, 0.0156, 0.0140, 0.0143], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0007, 0.0006, 0.0006], device='cuda:3') 2022-12-08 15:02:37,620 INFO [train.py:873] (3/4) Epoch 19, batch 3600, loss[loss=0.084, simple_loss=0.1243, pruned_loss=0.02186, over 14265.00 frames. ], tot_loss[loss=0.1031, simple_loss=0.1402, pruned_loss=0.033, over 2005506.39 frames. ], batch size: 35, lr: 4.13e-03, grad_scale: 8.0 2022-12-08 15:02:57,975 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.754e+01 2.074e+02 2.504e+02 3.273e+02 8.629e+02, threshold=5.008e+02, percent-clipped=8.0 2022-12-08 15:03:22,616 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.48 vs. limit=5.0 2022-12-08 15:03:24,007 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139769.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:03:41,947 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139789.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:03:55,621 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139804.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:03:57,300 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139806.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:04:07,191 INFO [train.py:873] (3/4) Epoch 19, batch 3700, loss[loss=0.1117, simple_loss=0.1206, pruned_loss=0.05141, over 2614.00 frames. ], tot_loss[loss=0.1039, simple_loss=0.1406, pruned_loss=0.03363, over 1983535.99 frames. ], batch size: 100, lr: 4.13e-03, grad_scale: 8.0 2022-12-08 15:04:18,384 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139830.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:04:26,981 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.254e+02 1.959e+02 2.384e+02 2.847e+02 4.025e+02, threshold=4.768e+02, percent-clipped=0.0 2022-12-08 15:04:36,794 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139850.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:04:36,813 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139850.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:04:49,245 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139865.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:04:49,423 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.09 vs. limit=2.0 2022-12-08 15:04:51,058 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139867.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:05:14,695 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139894.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:05:20,752 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6984, 2.0763, 3.7333, 2.7500, 3.6453, 2.0017, 2.9614, 3.6056], device='cuda:3'), covar=tensor([0.0686, 0.3698, 0.0563, 0.4322, 0.0774, 0.3071, 0.1278, 0.0556], device='cuda:3'), in_proj_covar=tensor([0.0251, 0.0197, 0.0220, 0.0268, 0.0239, 0.0202, 0.0199, 0.0221], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0005], device='cuda:3') 2022-12-08 15:05:29,267 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139911.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:05:34,122 INFO [train.py:873] (3/4) Epoch 19, batch 3800, loss[loss=0.09198, simple_loss=0.1359, pruned_loss=0.02402, over 13893.00 frames. ], tot_loss[loss=0.1044, simple_loss=0.1408, pruned_loss=0.03399, over 1952173.72 frames. ], batch size: 20, lr: 4.13e-03, grad_scale: 8.0 2022-12-08 15:05:43,817 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139927.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:05:48,200 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139932.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 15:05:53,158 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139938.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:05:54,736 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.144e+02 2.015e+02 2.567e+02 3.063e+02 4.403e+02, threshold=5.135e+02, percent-clipped=0.0 2022-12-08 15:05:56,573 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=139942.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:05:59,177 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139945.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:06:29,966 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=139980.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:06:41,160 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=139993.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:07:04,931 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.64 vs. limit=2.0 2022-12-08 15:07:06,101 INFO [train.py:873] (3/4) Epoch 19, batch 3900, loss[loss=0.0877, simple_loss=0.1266, pruned_loss=0.02438, over 14060.00 frames. ], tot_loss[loss=0.1031, simple_loss=0.1399, pruned_loss=0.03319, over 1987656.20 frames. ], batch size: 19, lr: 4.13e-03, grad_scale: 8.0 2022-12-08 15:07:25,637 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.340e+02 2.144e+02 2.574e+02 3.333e+02 6.487e+02, threshold=5.147e+02, percent-clipped=6.0 2022-12-08 15:07:55,054 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1629, 2.0285, 1.8437, 1.9495, 2.0842, 2.1313, 2.1119, 2.0862], device='cuda:3'), covar=tensor([0.1174, 0.0902, 0.2533, 0.2511, 0.1336, 0.1254, 0.1419, 0.1127], device='cuda:3'), in_proj_covar=tensor([0.0396, 0.0274, 0.0448, 0.0566, 0.0353, 0.0458, 0.0389, 0.0398], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 15:08:11,609 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.6734, 4.5582, 4.3017, 4.6941, 4.3040, 4.0002, 4.7586, 4.4900], device='cuda:3'), covar=tensor([0.0621, 0.0770, 0.0821, 0.0506, 0.0698, 0.0639, 0.0590, 0.0774], device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0148, 0.0151, 0.0166, 0.0153, 0.0125, 0.0173, 0.0153], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 15:08:21,820 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140103.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:08:27,944 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140110.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:08:33,683 INFO [train.py:873] (3/4) Epoch 19, batch 4000, loss[loss=0.1641, simple_loss=0.1506, pruned_loss=0.08882, over 1237.00 frames. ], tot_loss[loss=0.1033, simple_loss=0.1399, pruned_loss=0.03338, over 1954279.19 frames. ], batch size: 100, lr: 4.13e-03, grad_scale: 8.0 2022-12-08 15:08:41,253 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140125.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:08:54,551 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.68 vs. limit=2.0 2022-12-08 15:08:54,677 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.294e+02 2.130e+02 2.579e+02 3.097e+02 6.234e+02, threshold=5.158e+02, percent-clipped=1.0 2022-12-08 15:08:59,205 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140145.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:09:05,117 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8281, 1.5371, 2.0342, 1.6458, 1.9723, 1.4512, 1.6575, 1.9289], device='cuda:3'), covar=tensor([0.3454, 0.2645, 0.0809, 0.1919, 0.1594, 0.1614, 0.1444, 0.1022], device='cuda:3'), in_proj_covar=tensor([0.0252, 0.0199, 0.0219, 0.0267, 0.0241, 0.0202, 0.0200, 0.0220], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0005], device='cuda:3') 2022-12-08 15:09:12,090 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140160.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:09:14,057 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140162.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:09:15,788 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140164.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:09:22,499 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140171.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:09:44,485 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3782, 2.2236, 2.5943, 1.6206, 1.7941, 2.4374, 1.4269, 2.3510], device='cuda:3'), covar=tensor([0.0839, 0.1218, 0.0785, 0.2158, 0.2204, 0.0793, 0.2756, 0.0848], device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0103, 0.0096, 0.0100, 0.0116, 0.0092, 0.0117, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 15:09:52,864 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140206.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:10:02,907 INFO [train.py:873] (3/4) Epoch 19, batch 4100, loss[loss=0.1396, simple_loss=0.1423, pruned_loss=0.06844, over 1246.00 frames. ], tot_loss[loss=0.1045, simple_loss=0.1407, pruned_loss=0.03418, over 1938269.79 frames. ], batch size: 100, lr: 4.12e-03, grad_scale: 8.0 2022-12-08 15:10:11,531 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140227.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:10:21,020 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140238.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:10:22,567 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.264e+02 2.026e+02 2.683e+02 3.306e+02 6.239e+02, threshold=5.366e+02, percent-clipped=2.0 2022-12-08 15:10:31,619 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9472, 3.7470, 3.5084, 3.6978, 3.8927, 3.9224, 3.9562, 3.9626], device='cuda:3'), covar=tensor([0.0930, 0.0618, 0.2077, 0.2488, 0.0708, 0.0899, 0.0935, 0.0829], device='cuda:3'), in_proj_covar=tensor([0.0398, 0.0276, 0.0451, 0.0569, 0.0356, 0.0463, 0.0392, 0.0401], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 15:10:39,358 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140259.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:10:52,816 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140275.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:11:02,836 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140286.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:11:15,988 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3911, 3.8076, 2.8386, 4.6270, 4.1721, 4.3758, 3.6516, 3.1108], device='cuda:3'), covar=tensor([0.0590, 0.1113, 0.3374, 0.0452, 0.0985, 0.1103, 0.1213, 0.2628], device='cuda:3'), in_proj_covar=tensor([0.0282, 0.0290, 0.0260, 0.0292, 0.0323, 0.0303, 0.0257, 0.0243], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 15:11:27,117 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.6104, 4.4773, 4.2431, 4.6068, 4.2364, 3.9978, 4.6667, 4.4319], device='cuda:3'), covar=tensor([0.0561, 0.0806, 0.0788, 0.0537, 0.0742, 0.0545, 0.0539, 0.0745], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0148, 0.0150, 0.0165, 0.0152, 0.0126, 0.0171, 0.0152], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 15:11:29,479 INFO [train.py:873] (3/4) Epoch 19, batch 4200, loss[loss=0.1158, simple_loss=0.1413, pruned_loss=0.0452, over 3891.00 frames. ], tot_loss[loss=0.1035, simple_loss=0.1403, pruned_loss=0.0333, over 1981071.38 frames. ], batch size: 100, lr: 4.12e-03, grad_scale: 8.0 2022-12-08 15:11:30,436 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1478, 1.9521, 2.2638, 2.3661, 2.0614, 1.9181, 2.3254, 2.0572], device='cuda:3'), covar=tensor([0.0376, 0.0820, 0.0400, 0.0424, 0.0686, 0.1313, 0.0459, 0.0512], device='cuda:3'), in_proj_covar=tensor([0.0294, 0.0262, 0.0378, 0.0333, 0.0273, 0.0310, 0.0314, 0.0278], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 15:11:31,993 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140320.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:11:49,279 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.317e+02 2.052e+02 2.429e+02 3.115e+02 5.934e+02, threshold=4.859e+02, percent-clipped=1.0 2022-12-08 15:12:04,453 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4574, 4.1769, 3.8835, 4.1170, 4.2567, 4.3749, 4.4341, 4.4175], device='cuda:3'), covar=tensor([0.0729, 0.0481, 0.1924, 0.2220, 0.0709, 0.0726, 0.0794, 0.0704], device='cuda:3'), in_proj_covar=tensor([0.0399, 0.0276, 0.0452, 0.0572, 0.0359, 0.0464, 0.0392, 0.0402], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 15:12:04,500 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9701, 1.9882, 4.2211, 3.9101, 3.9396, 4.3387, 3.7696, 4.3035], device='cuda:3'), covar=tensor([0.1619, 0.1440, 0.0127, 0.0226, 0.0250, 0.0137, 0.0216, 0.0118], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0158, 0.0132, 0.0170, 0.0149, 0.0143, 0.0127, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 15:12:29,811 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140387.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:12:56,649 INFO [train.py:873] (3/4) Epoch 19, batch 4300, loss[loss=0.12, simple_loss=0.1251, pruned_loss=0.05741, over 2696.00 frames. ], tot_loss[loss=0.1044, simple_loss=0.1409, pruned_loss=0.03394, over 1963686.15 frames. ], batch size: 100, lr: 4.12e-03, grad_scale: 8.0 2022-12-08 15:13:03,478 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140425.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:13:16,351 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.481e+01 2.038e+02 2.405e+02 2.912e+02 6.757e+02, threshold=4.809e+02, percent-clipped=4.0 2022-12-08 15:13:17,434 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8278, 3.0298, 3.0118, 3.1315, 2.3460, 3.1662, 2.9919, 1.8623], device='cuda:3'), covar=tensor([0.0905, 0.0856, 0.0701, 0.0452, 0.0895, 0.0421, 0.0700, 0.1524], device='cuda:3'), in_proj_covar=tensor([0.0137, 0.0092, 0.0072, 0.0078, 0.0102, 0.0094, 0.0103, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0006, 0.0006, 0.0007, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 15:13:21,059 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140445.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:13:23,802 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140448.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:13:33,475 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140459.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:13:34,338 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140460.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:13:36,106 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140462.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:13:39,464 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140466.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:13:42,592 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.71 vs. limit=5.0 2022-12-08 15:13:45,319 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140473.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:13:53,145 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8286, 1.7601, 1.7887, 1.9831, 1.8332, 1.2939, 1.5435, 1.7618], device='cuda:3'), covar=tensor([0.0700, 0.0715, 0.0584, 0.0521, 0.0772, 0.0823, 0.0967, 0.0654], device='cuda:3'), in_proj_covar=tensor([0.0038, 0.0036, 0.0041, 0.0035, 0.0037, 0.0051, 0.0038, 0.0041], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-08 15:13:55,005 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.6370, 5.1723, 5.0572, 5.6028, 5.1797, 4.7349, 5.5287, 4.5887], device='cuda:3'), covar=tensor([0.0296, 0.0821, 0.0385, 0.0360, 0.0693, 0.0340, 0.0494, 0.0499], device='cuda:3'), in_proj_covar=tensor([0.0181, 0.0277, 0.0205, 0.0202, 0.0188, 0.0162, 0.0293, 0.0172], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 15:13:56,784 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0178, 2.9155, 2.1980, 3.0362, 2.8764, 2.8990, 2.6124, 2.3106], device='cuda:3'), covar=tensor([0.0792, 0.1144, 0.2730, 0.0898, 0.0994, 0.1089, 0.1325, 0.2467], device='cuda:3'), in_proj_covar=tensor([0.0281, 0.0288, 0.0258, 0.0291, 0.0322, 0.0303, 0.0255, 0.0242], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 15:13:59,664 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140489.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:14:03,140 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140493.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:14:14,443 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140506.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:14:16,051 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140508.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:14:17,019 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140509.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:14:17,797 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140510.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:14:23,506 INFO [train.py:873] (3/4) Epoch 19, batch 4400, loss[loss=0.1032, simple_loss=0.1467, pruned_loss=0.02983, over 14213.00 frames. ], tot_loss[loss=0.1038, simple_loss=0.1408, pruned_loss=0.03342, over 2005346.73 frames. ], batch size: 32, lr: 4.12e-03, grad_scale: 8.0 2022-12-08 15:14:43,959 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.073e+02 2.032e+02 2.425e+02 2.812e+02 5.381e+02, threshold=4.850e+02, percent-clipped=2.0 2022-12-08 15:14:44,248 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3363, 2.5996, 4.3704, 3.0275, 4.1852, 4.1244, 3.9100, 3.3727], device='cuda:3'), covar=tensor([0.0701, 0.3540, 0.0649, 0.1830, 0.0703, 0.0918, 0.1546, 0.2621], device='cuda:3'), in_proj_covar=tensor([0.0345, 0.0305, 0.0382, 0.0294, 0.0359, 0.0318, 0.0358, 0.0293], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 15:14:45,775 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9264, 3.0603, 3.1332, 2.9901, 3.0694, 2.8617, 1.5383, 2.8527], device='cuda:3'), covar=tensor([0.0515, 0.0423, 0.0385, 0.0412, 0.0335, 0.0799, 0.2803, 0.0373], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0176, 0.0147, 0.0150, 0.0209, 0.0143, 0.0157, 0.0197], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 15:14:52,752 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140550.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:14:56,037 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140554.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:15:05,365 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.8449, 5.7766, 5.3607, 5.8990, 5.3907, 5.3485, 6.0070, 5.5699], device='cuda:3'), covar=tensor([0.0604, 0.0609, 0.0715, 0.0419, 0.0728, 0.0376, 0.0526, 0.0696], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0149, 0.0151, 0.0167, 0.0153, 0.0127, 0.0174, 0.0153], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 15:15:09,918 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140570.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:15:50,192 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140615.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:15:51,860 INFO [train.py:873] (3/4) Epoch 19, batch 4500, loss[loss=0.07939, simple_loss=0.1213, pruned_loss=0.01873, over 14392.00 frames. ], tot_loss[loss=0.1027, simple_loss=0.1401, pruned_loss=0.03268, over 1998497.07 frames. ], batch size: 18, lr: 4.12e-03, grad_scale: 4.0 2022-12-08 15:16:12,049 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.060e+02 2.100e+02 2.600e+02 3.124e+02 5.170e+02, threshold=5.201e+02, percent-clipped=3.0 2022-12-08 15:16:24,882 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.5599, 1.7361, 1.9459, 1.9294, 1.7681, 1.8515, 1.5768, 1.3973], device='cuda:3'), covar=tensor([0.0893, 0.1150, 0.0417, 0.0692, 0.1287, 0.1058, 0.1700, 0.1805], device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0092, 0.0072, 0.0077, 0.0102, 0.0093, 0.0103, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0006, 0.0006, 0.0007, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 15:17:17,738 INFO [train.py:873] (3/4) Epoch 19, batch 4600, loss[loss=0.1078, simple_loss=0.141, pruned_loss=0.0373, over 14179.00 frames. ], tot_loss[loss=0.1032, simple_loss=0.1405, pruned_loss=0.03291, over 1998717.44 frames. ], batch size: 99, lr: 4.12e-03, grad_scale: 4.0 2022-12-08 15:17:20,086 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.38 vs. limit=5.0 2022-12-08 15:17:23,443 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140723.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:17:39,454 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.258e+02 2.073e+02 2.575e+02 3.535e+02 1.946e+03, threshold=5.149e+02, percent-clipped=6.0 2022-12-08 15:17:41,343 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140743.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:17:55,003 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140759.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:18:00,906 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140766.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:18:17,077 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140784.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:18:36,906 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140807.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:18:38,954 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140809.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:18:43,025 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140814.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:18:45,759 INFO [train.py:873] (3/4) Epoch 19, batch 4700, loss[loss=0.09743, simple_loss=0.139, pruned_loss=0.02792, over 10382.00 frames. ], tot_loss[loss=0.1035, simple_loss=0.1404, pruned_loss=0.03332, over 1971056.03 frames. ], batch size: 100, lr: 4.11e-03, grad_scale: 4.0 2022-12-08 15:19:06,570 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.370e+02 2.144e+02 2.562e+02 3.057e+02 6.222e+02, threshold=5.124e+02, percent-clipped=4.0 2022-12-08 15:19:10,107 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140845.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:19:28,316 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140865.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:19:32,458 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140870.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:19:39,090 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9888, 3.5508, 2.7279, 4.1523, 3.9465, 3.9480, 3.4623, 2.8781], device='cuda:3'), covar=tensor([0.0695, 0.1143, 0.3000, 0.0578, 0.0872, 0.1242, 0.1151, 0.2561], device='cuda:3'), in_proj_covar=tensor([0.0281, 0.0289, 0.0259, 0.0292, 0.0323, 0.0302, 0.0256, 0.0242], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 15:19:45,316 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5717, 2.2783, 2.9522, 1.9825, 1.9753, 2.6990, 1.5519, 2.6649], device='cuda:3'), covar=tensor([0.0921, 0.1509, 0.0844, 0.1826, 0.2058, 0.0928, 0.2971, 0.0973], device='cuda:3'), in_proj_covar=tensor([0.0088, 0.0104, 0.0097, 0.0101, 0.0115, 0.0092, 0.0117, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 15:20:07,466 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8242, 3.6135, 3.3660, 3.5552, 3.7340, 3.7745, 3.8046, 3.8165], device='cuda:3'), covar=tensor([0.0879, 0.0559, 0.2082, 0.2184, 0.0727, 0.0880, 0.0867, 0.0765], device='cuda:3'), in_proj_covar=tensor([0.0396, 0.0276, 0.0450, 0.0567, 0.0357, 0.0459, 0.0393, 0.0402], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 15:20:11,728 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140915.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:20:13,318 INFO [train.py:873] (3/4) Epoch 19, batch 4800, loss[loss=0.1183, simple_loss=0.1305, pruned_loss=0.05308, over 3926.00 frames. ], tot_loss[loss=0.1026, simple_loss=0.14, pruned_loss=0.03267, over 2028493.15 frames. ], batch size: 100, lr: 4.11e-03, grad_scale: 8.0 2022-12-08 15:20:13,436 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140917.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:20:34,742 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.343e+02 2.115e+02 2.464e+02 2.978e+02 7.209e+02, threshold=4.928e+02, percent-clipped=2.0 2022-12-08 15:20:38,740 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.95 vs. limit=5.0 2022-12-08 15:20:39,976 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140947.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:20:53,888 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140963.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:21:03,016 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2022-12-08 15:21:07,419 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140978.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:21:24,449 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9341, 1.5450, 3.8078, 1.9888, 3.8814, 3.9899, 2.9715, 4.3331], device='cuda:3'), covar=tensor([0.0241, 0.3072, 0.0474, 0.2069, 0.0492, 0.0472, 0.0765, 0.0181], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0157, 0.0163, 0.0170, 0.0169, 0.0181, 0.0135, 0.0155], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 15:21:34,143 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141008.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:21:41,897 INFO [train.py:873] (3/4) Epoch 19, batch 4900, loss[loss=0.08251, simple_loss=0.126, pruned_loss=0.0195, over 14057.00 frames. ], tot_loss[loss=0.1018, simple_loss=0.1392, pruned_loss=0.03222, over 1996567.33 frames. ], batch size: 22, lr: 4.11e-03, grad_scale: 8.0 2022-12-08 15:22:02,360 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.082e+02 2.070e+02 2.764e+02 3.531e+02 1.295e+03, threshold=5.527e+02, percent-clipped=7.0 2022-12-08 15:22:04,138 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141043.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:22:10,050 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2022-12-08 15:22:11,984 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.8673, 5.7260, 5.2851, 5.8584, 5.4821, 5.2754, 5.9729, 5.6665], device='cuda:3'), covar=tensor([0.0565, 0.0548, 0.0818, 0.0505, 0.0641, 0.0422, 0.0408, 0.0609], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0147, 0.0149, 0.0164, 0.0151, 0.0126, 0.0171, 0.0151], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 15:22:14,335 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.62 vs. limit=2.0 2022-12-08 15:22:34,878 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141079.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:22:45,513 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=141091.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:23:08,043 INFO [train.py:873] (3/4) Epoch 19, batch 5000, loss[loss=0.1146, simple_loss=0.1398, pruned_loss=0.04463, over 4971.00 frames. ], tot_loss[loss=0.1023, simple_loss=0.1396, pruned_loss=0.03256, over 2009094.56 frames. ], batch size: 100, lr: 4.11e-03, grad_scale: 8.0 2022-12-08 15:23:17,682 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7626, 2.2447, 3.7103, 3.8427, 3.6458, 2.2716, 3.7729, 2.8106], device='cuda:3'), covar=tensor([0.0542, 0.1431, 0.0975, 0.0522, 0.0632, 0.2098, 0.0559, 0.1168], device='cuda:3'), in_proj_covar=tensor([0.0294, 0.0262, 0.0377, 0.0333, 0.0273, 0.0308, 0.0313, 0.0278], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 15:23:28,878 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.297e+02 2.058e+02 2.477e+02 3.160e+02 7.208e+02, threshold=4.953e+02, percent-clipped=1.0 2022-12-08 15:23:30,762 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141143.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:23:32,277 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141145.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:23:40,853 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8922, 2.4674, 3.7331, 2.8715, 3.7693, 3.6025, 3.4406, 3.1868], device='cuda:3'), covar=tensor([0.1112, 0.3133, 0.1101, 0.1796, 0.0805, 0.1169, 0.1531, 0.1578], device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0310, 0.0388, 0.0300, 0.0363, 0.0324, 0.0364, 0.0298], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 15:23:49,064 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141165.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:23:49,128 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141165.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:24:13,407 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=141193.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:24:22,795 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141204.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:24:30,573 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=141213.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:24:34,009 INFO [train.py:873] (3/4) Epoch 19, batch 5100, loss[loss=0.1207, simple_loss=0.1569, pruned_loss=0.04223, over 14482.00 frames. ], tot_loss[loss=0.1034, simple_loss=0.14, pruned_loss=0.03338, over 1996185.12 frames. ], batch size: 51, lr: 4.11e-03, grad_scale: 8.0 2022-12-08 15:24:54,061 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.191e+02 2.078e+02 2.511e+02 3.062e+02 5.602e+02, threshold=5.021e+02, percent-clipped=1.0 2022-12-08 15:25:14,881 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.58 vs. limit=5.0 2022-12-08 15:25:21,708 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141273.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:25:34,499 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.59 vs. limit=2.0 2022-12-08 15:25:46,038 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.71 vs. limit=5.0 2022-12-08 15:25:47,975 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141303.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:25:59,794 INFO [train.py:873] (3/4) Epoch 19, batch 5200, loss[loss=0.08551, simple_loss=0.1307, pruned_loss=0.02016, over 14411.00 frames. ], tot_loss[loss=0.1037, simple_loss=0.1403, pruned_loss=0.03356, over 2023332.63 frames. ], batch size: 53, lr: 4.11e-03, grad_scale: 8.0 2022-12-08 15:26:10,587 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141329.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:26:16,307 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.58 vs. limit=2.0 2022-12-08 15:26:16,309 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.78 vs. limit=2.0 2022-12-08 15:26:20,994 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.019e+02 2.014e+02 2.549e+02 3.340e+02 7.347e+02, threshold=5.099e+02, percent-clipped=2.0 2022-12-08 15:26:54,457 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.6811, 4.3801, 4.2064, 4.6968, 4.3676, 4.1486, 4.7097, 4.0083], device='cuda:3'), covar=tensor([0.0407, 0.0953, 0.0452, 0.0438, 0.0817, 0.0779, 0.0560, 0.0492], device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0279, 0.0204, 0.0202, 0.0187, 0.0162, 0.0293, 0.0172], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 15:26:54,524 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141379.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:27:04,060 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141390.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:27:08,384 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8292, 1.4030, 1.8418, 1.3794, 1.5827, 1.9249, 1.6914, 1.6797], device='cuda:3'), covar=tensor([0.1068, 0.0898, 0.0908, 0.1314, 0.1514, 0.1090, 0.0911, 0.1634], device='cuda:3'), in_proj_covar=tensor([0.0157, 0.0171, 0.0142, 0.0127, 0.0146, 0.0158, 0.0141, 0.0145], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0007, 0.0006, 0.0006], device='cuda:3') 2022-12-08 15:27:19,039 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3251, 2.1624, 4.3196, 3.1094, 4.0979, 2.0878, 3.2872, 4.1601], device='cuda:3'), covar=tensor([0.0675, 0.3855, 0.0529, 0.5289, 0.0914, 0.3344, 0.1356, 0.0606], device='cuda:3'), in_proj_covar=tensor([0.0252, 0.0197, 0.0219, 0.0267, 0.0242, 0.0200, 0.0199, 0.0220], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0005], device='cuda:3') 2022-12-08 15:27:28,121 INFO [train.py:873] (3/4) Epoch 19, batch 5300, loss[loss=0.08416, simple_loss=0.1258, pruned_loss=0.02126, over 13605.00 frames. ], tot_loss[loss=0.1035, simple_loss=0.1403, pruned_loss=0.03335, over 2015108.56 frames. ], batch size: 17, lr: 4.11e-03, grad_scale: 8.0 2022-12-08 15:27:28,316 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141417.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 15:27:36,863 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=141427.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:27:48,828 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.160e+02 2.222e+02 2.556e+02 2.994e+02 5.858e+02, threshold=5.112e+02, percent-clipped=4.0 2022-12-08 15:27:49,066 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141441.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:27:50,738 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7753, 1.6973, 1.7096, 1.9244, 1.8392, 1.2485, 1.5286, 1.5973], device='cuda:3'), covar=tensor([0.0733, 0.0657, 0.0697, 0.0700, 0.0421, 0.0801, 0.0731, 0.0893], device='cuda:3'), in_proj_covar=tensor([0.0039, 0.0036, 0.0042, 0.0035, 0.0037, 0.0051, 0.0039, 0.0041], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-08 15:28:10,346 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141465.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:28:21,433 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141478.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 15:28:39,750 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141499.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:28:42,510 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141502.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:28:51,518 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=141513.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:28:54,902 INFO [train.py:873] (3/4) Epoch 19, batch 5400, loss[loss=0.14, simple_loss=0.1372, pruned_loss=0.07144, over 1231.00 frames. ], tot_loss[loss=0.1035, simple_loss=0.1403, pruned_loss=0.03333, over 1969375.06 frames. ], batch size: 100, lr: 4.10e-03, grad_scale: 8.0 2022-12-08 15:29:02,829 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8099, 3.6568, 3.3318, 2.8235, 3.3346, 3.5929, 3.8982, 3.2660], device='cuda:3'), covar=tensor([0.0551, 0.0957, 0.0798, 0.1031, 0.0744, 0.0582, 0.0772, 0.0829], device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0170, 0.0141, 0.0126, 0.0145, 0.0156, 0.0140, 0.0143], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0007, 0.0006, 0.0006], device='cuda:3') 2022-12-08 15:29:16,144 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.046e+02 2.017e+02 2.549e+02 3.465e+02 7.895e+02, threshold=5.098e+02, percent-clipped=4.0 2022-12-08 15:29:20,116 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8545, 2.3832, 3.7014, 2.8647, 3.7606, 3.5742, 3.5130, 3.0623], device='cuda:3'), covar=tensor([0.1098, 0.3136, 0.1000, 0.1803, 0.0686, 0.1061, 0.1274, 0.1797], device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0308, 0.0387, 0.0297, 0.0361, 0.0323, 0.0359, 0.0296], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 15:29:43,846 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141573.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:30:10,178 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141603.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:30:22,126 INFO [train.py:873] (3/4) Epoch 19, batch 5500, loss[loss=0.1186, simple_loss=0.1218, pruned_loss=0.05775, over 2568.00 frames. ], tot_loss[loss=0.1023, simple_loss=0.1394, pruned_loss=0.0326, over 1968173.31 frames. ], batch size: 100, lr: 4.10e-03, grad_scale: 8.0 2022-12-08 15:30:24,017 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8312, 0.8323, 0.7025, 0.8622, 0.8240, 0.3720, 0.7960, 0.8213], device='cuda:3'), covar=tensor([0.0409, 0.0524, 0.0500, 0.0400, 0.0376, 0.0316, 0.0832, 0.0737], device='cuda:3'), in_proj_covar=tensor([0.0039, 0.0036, 0.0042, 0.0035, 0.0037, 0.0051, 0.0039, 0.0041], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-08 15:30:25,887 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=141621.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:30:43,271 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.230e+02 2.074e+02 2.608e+02 3.444e+02 6.357e+02, threshold=5.216e+02, percent-clipped=6.0 2022-12-08 15:30:52,130 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=141651.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:30:53,214 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7626, 1.8692, 1.9289, 1.7650, 1.7872, 1.7949, 1.4679, 1.3275], device='cuda:3'), covar=tensor([0.0223, 0.0286, 0.0194, 0.0274, 0.0244, 0.0280, 0.0278, 0.0429], device='cuda:3'), in_proj_covar=tensor([0.0024, 0.0024, 0.0022, 0.0023, 0.0023, 0.0035, 0.0029, 0.0034], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:3') 2022-12-08 15:31:21,360 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141685.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:31:28,434 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.59 vs. limit=5.0 2022-12-08 15:31:48,910 INFO [train.py:873] (3/4) Epoch 19, batch 5600, loss[loss=0.1741, simple_loss=0.1619, pruned_loss=0.09316, over 1215.00 frames. ], tot_loss[loss=0.103, simple_loss=0.14, pruned_loss=0.03295, over 1970028.59 frames. ], batch size: 100, lr: 4.10e-03, grad_scale: 8.0 2022-12-08 15:32:09,418 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.697e+01 2.227e+02 2.853e+02 3.557e+02 6.540e+02, threshold=5.705e+02, percent-clipped=5.0 2022-12-08 15:32:25,806 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2022-12-08 15:32:27,942 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7380, 1.3638, 1.7029, 1.2473, 1.5067, 1.8134, 1.5620, 1.5611], device='cuda:3'), covar=tensor([0.0828, 0.0801, 0.0723, 0.0869, 0.1508, 0.0952, 0.0938, 0.1840], device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0170, 0.0141, 0.0126, 0.0146, 0.0156, 0.0140, 0.0144], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0007, 0.0006, 0.0006], device='cuda:3') 2022-12-08 15:32:37,279 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141773.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 15:32:37,362 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2570, 1.7147, 2.4518, 2.0084, 2.3151, 1.6515, 2.0449, 2.3380], device='cuda:3'), covar=tensor([0.2451, 0.3075, 0.0788, 0.1899, 0.1558, 0.2014, 0.1082, 0.1008], device='cuda:3'), in_proj_covar=tensor([0.0255, 0.0198, 0.0220, 0.0268, 0.0241, 0.0201, 0.0201, 0.0223], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0005], device='cuda:3') 2022-12-08 15:32:58,671 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141797.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:33:00,336 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141799.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:33:16,297 INFO [train.py:873] (3/4) Epoch 19, batch 5700, loss[loss=0.09722, simple_loss=0.1334, pruned_loss=0.0305, over 10360.00 frames. ], tot_loss[loss=0.1025, simple_loss=0.1397, pruned_loss=0.0326, over 1991170.31 frames. ], batch size: 100, lr: 4.10e-03, grad_scale: 8.0 2022-12-08 15:33:36,775 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.084e+02 2.066e+02 2.688e+02 3.167e+02 5.846e+02, threshold=5.376e+02, percent-clipped=2.0 2022-12-08 15:33:41,756 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=141847.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:33:54,898 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7571, 2.6824, 2.0476, 2.8541, 2.6272, 2.6794, 2.4035, 2.2017], device='cuda:3'), covar=tensor([0.1021, 0.1287, 0.3111, 0.0883, 0.1251, 0.1155, 0.1645, 0.2422], device='cuda:3'), in_proj_covar=tensor([0.0282, 0.0289, 0.0258, 0.0292, 0.0322, 0.0303, 0.0256, 0.0243], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 15:34:17,091 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 2022-12-08 15:34:31,120 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.71 vs. limit=5.0 2022-12-08 15:34:41,420 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141916.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:34:42,168 INFO [train.py:873] (3/4) Epoch 19, batch 5800, loss[loss=0.1184, simple_loss=0.1224, pruned_loss=0.05718, over 1164.00 frames. ], tot_loss[loss=0.1039, simple_loss=0.1403, pruned_loss=0.03377, over 1922760.61 frames. ], batch size: 100, lr: 4.10e-03, grad_scale: 8.0 2022-12-08 15:34:47,971 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.67 vs. limit=5.0 2022-12-08 15:34:51,854 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141928.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:35:03,273 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.310e+01 2.008e+02 2.595e+02 3.223e+02 5.261e+02, threshold=5.190e+02, percent-clipped=0.0 2022-12-08 15:35:35,489 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141977.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:35:43,150 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141985.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:35:46,598 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141989.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:36:11,184 INFO [train.py:873] (3/4) Epoch 19, batch 5900, loss[loss=0.1109, simple_loss=0.1446, pruned_loss=0.03858, over 9494.00 frames. ], tot_loss[loss=0.1025, simple_loss=0.1397, pruned_loss=0.0327, over 1988735.43 frames. ], batch size: 100, lr: 4.10e-03, grad_scale: 8.0 2022-12-08 15:36:11,854 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2022-12-08 15:36:24,798 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=142033.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:36:31,706 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.115e+02 2.065e+02 2.424e+02 3.179e+02 4.807e+02, threshold=4.849e+02, percent-clipped=0.0 2022-12-08 15:36:50,714 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8593, 1.5253, 3.8537, 1.8980, 3.8281, 4.0943, 3.0572, 4.2492], device='cuda:3'), covar=tensor([0.0252, 0.3321, 0.0493, 0.2162, 0.0498, 0.0407, 0.0778, 0.0229], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0156, 0.0161, 0.0168, 0.0167, 0.0179, 0.0133, 0.0153], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 15:37:00,279 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=142073.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 15:37:21,240 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=142097.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:37:38,674 INFO [train.py:873] (3/4) Epoch 19, batch 6000, loss[loss=0.1231, simple_loss=0.1594, pruned_loss=0.0434, over 14298.00 frames. ], tot_loss[loss=0.1023, simple_loss=0.1395, pruned_loss=0.03254, over 1969850.90 frames. ], batch size: 39, lr: 4.10e-03, grad_scale: 8.0 2022-12-08 15:37:38,674 INFO [train.py:896] (3/4) Computing validation loss 2022-12-08 15:37:47,303 INFO [train.py:905] (3/4) Epoch 19, validation: loss=0.1418, simple_loss=0.1782, pruned_loss=0.05266, over 857387.00 frames. 2022-12-08 15:37:47,304 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-08 15:37:50,876 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=142121.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 15:38:08,226 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.241e+02 1.928e+02 2.466e+02 3.087e+02 6.019e+02, threshold=4.931e+02, percent-clipped=3.0 2022-12-08 15:38:11,917 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=142145.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:38:46,197 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.12 vs. limit=2.0 2022-12-08 15:39:15,265 INFO [train.py:873] (3/4) Epoch 19, batch 6100, loss[loss=0.1289, simple_loss=0.1313, pruned_loss=0.06324, over 1217.00 frames. ], tot_loss[loss=0.1022, simple_loss=0.1396, pruned_loss=0.03241, over 1988840.15 frames. ], batch size: 100, lr: 4.09e-03, grad_scale: 8.0 2022-12-08 15:39:36,596 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.188e+02 2.000e+02 2.533e+02 3.049e+02 1.114e+03, threshold=5.065e+02, percent-clipped=3.0 2022-12-08 15:39:50,024 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.10 vs. limit=2.0 2022-12-08 15:39:57,617 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5508, 4.0429, 3.1319, 4.9477, 4.3910, 4.6444, 4.1972, 3.3293], device='cuda:3'), covar=tensor([0.0654, 0.1077, 0.3156, 0.0423, 0.0782, 0.1192, 0.0975, 0.2714], device='cuda:3'), in_proj_covar=tensor([0.0281, 0.0286, 0.0256, 0.0291, 0.0320, 0.0301, 0.0254, 0.0242], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 15:40:03,723 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=142272.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:40:14,703 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=142284.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:40:42,867 INFO [train.py:873] (3/4) Epoch 19, batch 6200, loss[loss=0.1058, simple_loss=0.1227, pruned_loss=0.04443, over 2592.00 frames. ], tot_loss[loss=0.1024, simple_loss=0.1392, pruned_loss=0.03282, over 1928833.24 frames. ], batch size: 100, lr: 4.09e-03, grad_scale: 8.0 2022-12-08 15:41:04,427 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.271e+02 2.036e+02 2.651e+02 3.192e+02 8.704e+02, threshold=5.303e+02, percent-clipped=4.0 2022-12-08 15:42:06,726 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8099, 1.5645, 1.8483, 1.9806, 1.5407, 1.7403, 1.7133, 1.8076], device='cuda:3'), covar=tensor([0.0284, 0.0416, 0.0286, 0.0241, 0.0550, 0.0614, 0.0362, 0.0283], device='cuda:3'), in_proj_covar=tensor([0.0293, 0.0260, 0.0375, 0.0329, 0.0271, 0.0307, 0.0312, 0.0275], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 15:42:11,863 INFO [train.py:873] (3/4) Epoch 19, batch 6300, loss[loss=0.1077, simple_loss=0.1554, pruned_loss=0.02997, over 14184.00 frames. ], tot_loss[loss=0.1027, simple_loss=0.1399, pruned_loss=0.03274, over 1973069.87 frames. ], batch size: 25, lr: 4.09e-03, grad_scale: 8.0 2022-12-08 15:42:27,745 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.67 vs. limit=2.0 2022-12-08 15:42:32,398 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.018e+02 2.111e+02 2.568e+02 3.235e+02 6.197e+02, threshold=5.136e+02, percent-clipped=2.0 2022-12-08 15:42:43,683 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.9314, 2.7194, 2.4711, 2.6738, 2.8311, 2.8623, 2.8614, 2.8688], device='cuda:3'), covar=tensor([0.0944, 0.0753, 0.2464, 0.2263, 0.1008, 0.1050, 0.1434, 0.0921], device='cuda:3'), in_proj_covar=tensor([0.0395, 0.0277, 0.0452, 0.0568, 0.0357, 0.0462, 0.0394, 0.0400], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 15:43:38,809 INFO [train.py:873] (3/4) Epoch 19, batch 6400, loss[loss=0.1095, simple_loss=0.1455, pruned_loss=0.03681, over 14178.00 frames. ], tot_loss[loss=0.1031, simple_loss=0.1403, pruned_loss=0.03294, over 2001784.37 frames. ], batch size: 84, lr: 4.09e-03, grad_scale: 8.0 2022-12-08 15:44:00,397 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.252e+02 1.982e+02 2.595e+02 3.310e+02 8.116e+02, threshold=5.191e+02, percent-clipped=3.0 2022-12-08 15:44:26,974 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4396, 2.2271, 3.3016, 2.4994, 3.2313, 3.2071, 3.0519, 2.7493], device='cuda:3'), covar=tensor([0.0894, 0.3102, 0.1017, 0.1890, 0.0842, 0.1115, 0.1139, 0.1748], device='cuda:3'), in_proj_covar=tensor([0.0347, 0.0307, 0.0384, 0.0296, 0.0362, 0.0321, 0.0358, 0.0295], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 15:44:27,794 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=142572.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:44:28,640 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=142573.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:44:38,129 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=142584.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:44:42,402 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8044, 1.8438, 4.7241, 2.4166, 4.3730, 4.8176, 4.4982, 5.2071], device='cuda:3'), covar=tensor([0.0206, 0.3092, 0.0379, 0.1980, 0.0328, 0.0415, 0.0321, 0.0158], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0157, 0.0163, 0.0170, 0.0168, 0.0181, 0.0135, 0.0155], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 15:44:56,975 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2022-12-08 15:45:03,969 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2991, 2.2624, 1.9103, 2.4238, 2.2461, 2.3002, 2.1460, 1.9327], device='cuda:3'), covar=tensor([0.1062, 0.0985, 0.1876, 0.0909, 0.1320, 0.0746, 0.1370, 0.1622], device='cuda:3'), in_proj_covar=tensor([0.0283, 0.0290, 0.0260, 0.0293, 0.0323, 0.0304, 0.0257, 0.0244], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 15:45:07,275 INFO [train.py:873] (3/4) Epoch 19, batch 6500, loss[loss=0.1096, simple_loss=0.1213, pruned_loss=0.04893, over 2670.00 frames. ], tot_loss[loss=0.1038, simple_loss=0.1405, pruned_loss=0.03359, over 1907039.33 frames. ], batch size: 100, lr: 4.09e-03, grad_scale: 16.0 2022-12-08 15:45:09,997 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=142620.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:45:19,994 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=142632.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:45:21,752 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=142634.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:45:27,970 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.114e+02 2.216e+02 2.673e+02 3.413e+02 8.329e+02, threshold=5.346e+02, percent-clipped=5.0 2022-12-08 15:45:38,443 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.6292, 2.3669, 2.8714, 1.7873, 1.9448, 2.5120, 1.5001, 2.5090], device='cuda:3'), covar=tensor([0.0947, 0.1733, 0.0639, 0.2394, 0.2216, 0.1051, 0.3169, 0.1044], device='cuda:3'), in_proj_covar=tensor([0.0088, 0.0104, 0.0098, 0.0102, 0.0116, 0.0093, 0.0117, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 15:46:34,425 INFO [train.py:873] (3/4) Epoch 19, batch 6600, loss[loss=0.1378, simple_loss=0.164, pruned_loss=0.05577, over 9495.00 frames. ], tot_loss[loss=0.1034, simple_loss=0.1399, pruned_loss=0.03349, over 1913460.06 frames. ], batch size: 100, lr: 4.09e-03, grad_scale: 8.0 2022-12-08 15:46:40,965 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=142724.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:46:49,632 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2022-12-08 15:46:56,546 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.271e+02 2.141e+02 2.630e+02 3.221e+02 9.486e+02, threshold=5.260e+02, percent-clipped=4.0 2022-12-08 15:47:29,051 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4083, 1.4686, 3.5316, 1.6540, 3.2826, 3.5381, 2.4752, 3.7592], device='cuda:3'), covar=tensor([0.0252, 0.3016, 0.0347, 0.2154, 0.0799, 0.0372, 0.0936, 0.0182], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0156, 0.0161, 0.0169, 0.0167, 0.0179, 0.0134, 0.0154], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 15:47:34,325 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=142785.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:47:54,158 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.5334, 1.5572, 1.5426, 1.6573, 1.7002, 1.0894, 1.4376, 1.4271], device='cuda:3'), covar=tensor([0.0638, 0.0781, 0.0641, 0.0546, 0.0517, 0.0940, 0.0776, 0.0657], device='cuda:3'), in_proj_covar=tensor([0.0039, 0.0037, 0.0042, 0.0035, 0.0037, 0.0051, 0.0039, 0.0041], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-08 15:48:02,763 INFO [train.py:873] (3/4) Epoch 19, batch 6700, loss[loss=0.09791, simple_loss=0.1294, pruned_loss=0.03322, over 6050.00 frames. ], tot_loss[loss=0.1039, simple_loss=0.1405, pruned_loss=0.03365, over 1947431.13 frames. ], batch size: 100, lr: 4.09e-03, grad_scale: 8.0 2022-12-08 15:48:20,882 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7787, 1.7017, 1.7430, 1.9025, 1.8177, 1.2362, 1.6389, 1.6896], device='cuda:3'), covar=tensor([0.0776, 0.0610, 0.0765, 0.0813, 0.0530, 0.0822, 0.0758, 0.0637], device='cuda:3'), in_proj_covar=tensor([0.0039, 0.0037, 0.0042, 0.0035, 0.0037, 0.0052, 0.0039, 0.0041], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-08 15:48:23,750 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.068e+02 1.995e+02 2.478e+02 2.934e+02 5.263e+02, threshold=4.957e+02, percent-clipped=1.0 2022-12-08 15:49:28,807 INFO [train.py:873] (3/4) Epoch 19, batch 6800, loss[loss=0.09116, simple_loss=0.1305, pruned_loss=0.02589, over 14384.00 frames. ], tot_loss[loss=0.1039, simple_loss=0.1405, pruned_loss=0.03366, over 1934615.25 frames. ], batch size: 55, lr: 4.08e-03, grad_scale: 8.0 2022-12-08 15:49:39,800 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=142929.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:49:50,756 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.071e+02 1.868e+02 2.461e+02 3.045e+02 8.260e+02, threshold=4.922e+02, percent-clipped=6.0 2022-12-08 15:50:02,713 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7279, 3.8600, 4.0722, 3.6007, 3.9156, 3.8772, 1.6631, 3.6933], device='cuda:3'), covar=tensor([0.0370, 0.0375, 0.0303, 0.0497, 0.0296, 0.0381, 0.3009, 0.0298], device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0177, 0.0148, 0.0150, 0.0209, 0.0144, 0.0158, 0.0198], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 15:50:44,028 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.8746, 5.6964, 5.2231, 5.8073, 5.4721, 5.2777, 5.9438, 5.5646], device='cuda:3'), covar=tensor([0.0528, 0.0728, 0.0725, 0.0550, 0.0605, 0.0419, 0.0537, 0.0724], device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0148, 0.0150, 0.0165, 0.0152, 0.0127, 0.0171, 0.0152], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 15:50:56,728 INFO [train.py:873] (3/4) Epoch 19, batch 6900, loss[loss=0.1302, simple_loss=0.1251, pruned_loss=0.06766, over 1255.00 frames. ], tot_loss[loss=0.1031, simple_loss=0.14, pruned_loss=0.03306, over 1969993.91 frames. ], batch size: 100, lr: 4.08e-03, grad_scale: 4.0 2022-12-08 15:51:11,694 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.55 vs. limit=5.0 2022-12-08 15:51:18,510 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.205e+02 2.146e+02 2.489e+02 3.023e+02 7.661e+02, threshold=4.978e+02, percent-clipped=3.0 2022-12-08 15:51:24,596 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1387, 1.5282, 1.6951, 1.6937, 1.5966, 1.6575, 1.3981, 1.2644], device='cuda:3'), covar=tensor([0.1304, 0.1182, 0.0410, 0.0575, 0.1284, 0.1034, 0.1359, 0.1770], device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0093, 0.0072, 0.0078, 0.0102, 0.0092, 0.0103, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0006, 0.0006, 0.0007, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 15:51:51,242 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=143080.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:52:23,547 INFO [train.py:873] (3/4) Epoch 19, batch 7000, loss[loss=0.1207, simple_loss=0.1439, pruned_loss=0.04873, over 7798.00 frames. ], tot_loss[loss=0.1031, simple_loss=0.14, pruned_loss=0.03312, over 1951506.35 frames. ], batch size: 100, lr: 4.08e-03, grad_scale: 4.0 2022-12-08 15:52:46,400 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.312e+02 1.976e+02 2.529e+02 3.068e+02 1.002e+03, threshold=5.057e+02, percent-clipped=4.0 2022-12-08 15:53:30,311 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8410, 1.9312, 2.0420, 1.4772, 1.5251, 1.8679, 1.2381, 1.8932], device='cuda:3'), covar=tensor([0.1392, 0.1680, 0.1099, 0.2693, 0.3041, 0.1209, 0.3566, 0.1401], device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0103, 0.0096, 0.0101, 0.0115, 0.0092, 0.0116, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 15:53:32,650 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7970, 2.8532, 3.0169, 2.9333, 2.9841, 2.6861, 1.5570, 2.7562], device='cuda:3'), covar=tensor([0.0497, 0.0458, 0.0351, 0.0349, 0.0328, 0.1009, 0.2640, 0.0370], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0175, 0.0147, 0.0149, 0.0208, 0.0143, 0.0157, 0.0197], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 15:53:49,876 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9392, 4.5571, 4.4766, 4.9338, 4.6059, 4.3953, 4.9546, 4.2362], device='cuda:3'), covar=tensor([0.0375, 0.0996, 0.0441, 0.0430, 0.0863, 0.0539, 0.0485, 0.0515], device='cuda:3'), in_proj_covar=tensor([0.0185, 0.0282, 0.0208, 0.0205, 0.0190, 0.0164, 0.0297, 0.0174], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 15:53:50,608 INFO [train.py:873] (3/4) Epoch 19, batch 7100, loss[loss=0.1127, simple_loss=0.1539, pruned_loss=0.03576, over 14234.00 frames. ], tot_loss[loss=0.1022, simple_loss=0.1395, pruned_loss=0.03245, over 1974830.24 frames. ], batch size: 60, lr: 4.08e-03, grad_scale: 4.0 2022-12-08 15:54:00,930 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=143229.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:54:12,484 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.837e+01 2.042e+02 2.415e+02 3.007e+02 1.079e+03, threshold=4.829e+02, percent-clipped=5.0 2022-12-08 15:54:14,686 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3408, 1.3687, 3.3859, 1.5822, 3.2326, 3.4274, 2.4622, 3.6499], device='cuda:3'), covar=tensor([0.0276, 0.3189, 0.0428, 0.2328, 0.0847, 0.0473, 0.0926, 0.0229], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0155, 0.0160, 0.0167, 0.0165, 0.0177, 0.0133, 0.0152], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 15:54:27,676 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.0625, 2.6532, 4.9575, 3.4842, 4.8189, 2.3277, 3.6863, 4.8163], device='cuda:3'), covar=tensor([0.0483, 0.3207, 0.0381, 0.5303, 0.0424, 0.3060, 0.1315, 0.0480], device='cuda:3'), in_proj_covar=tensor([0.0252, 0.0198, 0.0219, 0.0265, 0.0237, 0.0200, 0.0200, 0.0219], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0005], device='cuda:3') 2022-12-08 15:54:42,675 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=143277.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:54:59,340 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3884, 3.9873, 2.8979, 4.6698, 4.2403, 4.4734, 4.0594, 3.4367], device='cuda:3'), covar=tensor([0.0671, 0.1086, 0.3498, 0.0721, 0.0851, 0.1949, 0.1129, 0.2566], device='cuda:3'), in_proj_covar=tensor([0.0280, 0.0284, 0.0256, 0.0289, 0.0317, 0.0299, 0.0255, 0.0240], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 15:55:17,810 INFO [train.py:873] (3/4) Epoch 19, batch 7200, loss[loss=0.09569, simple_loss=0.1348, pruned_loss=0.02829, over 14163.00 frames. ], tot_loss[loss=0.1016, simple_loss=0.139, pruned_loss=0.03209, over 1961678.72 frames. ], batch size: 99, lr: 4.08e-03, grad_scale: 8.0 2022-12-08 15:55:40,829 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.306e+02 2.274e+02 2.664e+02 3.314e+02 6.390e+02, threshold=5.328e+02, percent-clipped=7.0 2022-12-08 15:56:13,726 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=143380.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:56:30,791 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=143400.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 15:56:46,182 INFO [train.py:873] (3/4) Epoch 19, batch 7300, loss[loss=0.1005, simple_loss=0.1408, pruned_loss=0.0301, over 14296.00 frames. ], tot_loss[loss=0.1012, simple_loss=0.1381, pruned_loss=0.03209, over 1967857.00 frames. ], batch size: 69, lr: 4.08e-03, grad_scale: 8.0 2022-12-08 15:56:55,463 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=143428.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 15:57:08,204 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.099e+02 2.101e+02 2.553e+02 3.059e+02 5.821e+02, threshold=5.106e+02, percent-clipped=2.0 2022-12-08 15:57:25,016 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=143461.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 15:57:35,019 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=143473.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 15:58:13,339 INFO [train.py:873] (3/4) Epoch 19, batch 7400, loss[loss=0.08529, simple_loss=0.1276, pruned_loss=0.02147, over 14320.00 frames. ], tot_loss[loss=0.1021, simple_loss=0.139, pruned_loss=0.03261, over 2013862.35 frames. ], batch size: 44, lr: 4.08e-03, grad_scale: 8.0 2022-12-08 15:58:29,063 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=143534.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 15:58:36,573 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.395e+02 2.037e+02 2.578e+02 3.136e+02 8.501e+02, threshold=5.155e+02, percent-clipped=4.0 2022-12-08 15:58:37,441 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5421, 3.6268, 3.8427, 3.4435, 3.7314, 3.5883, 1.5646, 3.5226], device='cuda:3'), covar=tensor([0.0378, 0.0339, 0.0303, 0.0521, 0.0285, 0.0484, 0.2967, 0.0297], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0176, 0.0147, 0.0150, 0.0209, 0.0143, 0.0157, 0.0197], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 15:58:45,079 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5105, 4.3994, 4.0314, 4.1967, 4.4306, 4.4807, 4.5758, 4.5382], device='cuda:3'), covar=tensor([0.0989, 0.0434, 0.2188, 0.2586, 0.0765, 0.0821, 0.0752, 0.0797], device='cuda:3'), in_proj_covar=tensor([0.0402, 0.0282, 0.0464, 0.0584, 0.0365, 0.0469, 0.0401, 0.0407], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004], device='cuda:3') 2022-12-08 15:59:41,171 INFO [train.py:873] (3/4) Epoch 19, batch 7500, loss[loss=0.1252, simple_loss=0.1292, pruned_loss=0.06057, over 1224.00 frames. ], tot_loss[loss=0.1022, simple_loss=0.139, pruned_loss=0.03274, over 1956830.92 frames. ], batch size: 100, lr: 4.07e-03, grad_scale: 8.0 2022-12-08 16:00:03,425 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.287e+02 2.124e+02 2.640e+02 3.469e+02 7.311e+02, threshold=5.280e+02, percent-clipped=6.0 2022-12-08 16:00:07,252 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2022-12-08 16:00:12,603 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8464, 4.5698, 4.4680, 4.7937, 4.4830, 4.1729, 4.9183, 4.6596], device='cuda:3'), covar=tensor([0.0492, 0.0861, 0.0757, 0.0501, 0.0740, 0.0591, 0.0512, 0.0661], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0145, 0.0148, 0.0162, 0.0150, 0.0126, 0.0169, 0.0149], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 16:00:22,722 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 2022-12-08 16:00:24,200 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=6.44 vs. limit=5.0 2022-12-08 16:01:11,361 INFO [train.py:873] (3/4) Epoch 20, batch 0, loss[loss=0.1489, simple_loss=0.1708, pruned_loss=0.06347, over 8618.00 frames. ], tot_loss[loss=0.1489, simple_loss=0.1708, pruned_loss=0.06347, over 8618.00 frames. ], batch size: 100, lr: 3.97e-03, grad_scale: 8.0 2022-12-08 16:01:11,362 INFO [train.py:896] (3/4) Computing validation loss 2022-12-08 16:01:18,887 INFO [train.py:905] (3/4) Epoch 20, validation: loss=0.1452, simple_loss=0.1824, pruned_loss=0.05396, over 857387.00 frames. 2022-12-08 16:01:18,887 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-08 16:01:43,607 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.8706, 5.2553, 5.2640, 5.7992, 5.4003, 4.9952, 5.7343, 4.7265], device='cuda:3'), covar=tensor([0.0222, 0.0932, 0.0309, 0.0325, 0.0683, 0.0327, 0.0473, 0.0498], device='cuda:3'), in_proj_covar=tensor([0.0182, 0.0278, 0.0204, 0.0201, 0.0187, 0.0161, 0.0293, 0.0171], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 16:02:06,356 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4638, 3.2457, 3.8830, 2.8644, 2.4945, 3.3720, 1.9276, 3.3135], device='cuda:3'), covar=tensor([0.0722, 0.0953, 0.0453, 0.1486, 0.1803, 0.0662, 0.2821, 0.1083], device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0104, 0.0097, 0.0102, 0.0116, 0.0092, 0.0116, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 16:02:15,749 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.236e+01 1.946e+02 2.466e+02 3.510e+02 7.205e+02, threshold=4.933e+02, percent-clipped=3.0 2022-12-08 16:02:21,741 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9300, 1.6530, 3.4108, 3.0858, 3.2363, 3.4207, 2.7450, 3.4150], device='cuda:3'), covar=tensor([0.1657, 0.1787, 0.0180, 0.0360, 0.0343, 0.0184, 0.0378, 0.0192], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0157, 0.0132, 0.0170, 0.0149, 0.0143, 0.0126, 0.0125], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 16:02:27,763 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=143756.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 16:02:47,467 INFO [train.py:873] (3/4) Epoch 20, batch 100, loss[loss=0.11, simple_loss=0.1439, pruned_loss=0.03806, over 14189.00 frames. ], tot_loss[loss=0.1016, simple_loss=0.1399, pruned_loss=0.03168, over 880633.87 frames. ], batch size: 84, lr: 3.97e-03, grad_scale: 8.0 2022-12-08 16:02:49,387 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.76 vs. limit=2.0 2022-12-08 16:02:51,723 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.1711, 5.0145, 4.7267, 5.2235, 4.8427, 4.7753, 5.3545, 4.9717], device='cuda:3'), covar=tensor([0.0586, 0.0829, 0.0861, 0.0459, 0.0802, 0.0540, 0.0412, 0.0721], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0146, 0.0149, 0.0162, 0.0150, 0.0126, 0.0169, 0.0149], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 16:03:03,869 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.6853, 2.9580, 2.8320, 2.9796, 2.3090, 3.0763, 2.8912, 1.6887], device='cuda:3'), covar=tensor([0.1165, 0.0967, 0.1009, 0.0500, 0.1014, 0.0435, 0.0872, 0.1801], device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0092, 0.0071, 0.0077, 0.0102, 0.0092, 0.0103, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0006, 0.0006, 0.0007, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 16:03:27,182 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5099, 4.2829, 3.9222, 4.1946, 4.3223, 4.4470, 4.5095, 4.5017], device='cuda:3'), covar=tensor([0.0802, 0.0623, 0.2236, 0.2533, 0.0729, 0.0878, 0.0866, 0.0780], device='cuda:3'), in_proj_covar=tensor([0.0401, 0.0282, 0.0460, 0.0581, 0.0363, 0.0467, 0.0402, 0.0404], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004], device='cuda:3') 2022-12-08 16:03:31,994 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=143829.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 16:03:43,924 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.291e+02 2.111e+02 2.727e+02 3.354e+02 8.934e+02, threshold=5.454e+02, percent-clipped=6.0 2022-12-08 16:04:15,260 INFO [train.py:873] (3/4) Epoch 20, batch 200, loss[loss=0.09041, simple_loss=0.1339, pruned_loss=0.02347, over 14096.00 frames. ], tot_loss[loss=0.1039, simple_loss=0.1408, pruned_loss=0.03348, over 1309243.92 frames. ], batch size: 22, lr: 3.97e-03, grad_scale: 8.0 2022-12-08 16:04:17,291 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3081, 2.7057, 4.2612, 4.3863, 4.0992, 2.6248, 4.3918, 3.3724], device='cuda:3'), covar=tensor([0.0461, 0.1264, 0.0949, 0.0428, 0.0600, 0.1921, 0.0424, 0.0968], device='cuda:3'), in_proj_covar=tensor([0.0297, 0.0264, 0.0379, 0.0335, 0.0274, 0.0310, 0.0315, 0.0279], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 16:04:25,874 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8727, 3.7276, 3.5791, 3.9601, 3.7337, 3.6648, 4.0101, 3.2857], device='cuda:3'), covar=tensor([0.0567, 0.0915, 0.0501, 0.0458, 0.0754, 0.1009, 0.0526, 0.0609], device='cuda:3'), in_proj_covar=tensor([0.0183, 0.0280, 0.0205, 0.0202, 0.0187, 0.0162, 0.0294, 0.0172], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 16:05:02,107 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2022-12-08 16:05:11,776 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.198e+02 1.866e+02 2.261e+02 3.101e+02 5.918e+02, threshold=4.522e+02, percent-clipped=1.0 2022-12-08 16:05:32,098 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3408, 2.6897, 2.5828, 2.6017, 2.2286, 2.6851, 2.5554, 1.5917], device='cuda:3'), covar=tensor([0.1039, 0.0762, 0.0782, 0.0740, 0.0900, 0.0646, 0.0864, 0.1648], device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0093, 0.0072, 0.0078, 0.0101, 0.0092, 0.0103, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0006, 0.0006, 0.0007, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 16:05:43,234 INFO [train.py:873] (3/4) Epoch 20, batch 300, loss[loss=0.1102, simple_loss=0.1507, pruned_loss=0.03483, over 14230.00 frames. ], tot_loss[loss=0.1029, simple_loss=0.1399, pruned_loss=0.03294, over 1578578.74 frames. ], batch size: 76, lr: 3.96e-03, grad_scale: 8.0 2022-12-08 16:06:39,227 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.977e+01 2.046e+02 2.429e+02 3.312e+02 9.305e+02, threshold=4.857e+02, percent-clipped=8.0 2022-12-08 16:06:50,851 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=144056.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 16:06:58,011 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8096, 1.5370, 3.7107, 1.8232, 3.7779, 3.9590, 2.6964, 4.2074], device='cuda:3'), covar=tensor([0.0252, 0.3150, 0.0496, 0.2160, 0.0501, 0.0400, 0.0867, 0.0185], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0156, 0.0162, 0.0169, 0.0166, 0.0179, 0.0133, 0.0153], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 16:07:02,973 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.59 vs. limit=2.0 2022-12-08 16:07:10,843 INFO [train.py:873] (3/4) Epoch 20, batch 400, loss[loss=0.08614, simple_loss=0.1307, pruned_loss=0.02078, over 14230.00 frames. ], tot_loss[loss=0.102, simple_loss=0.1394, pruned_loss=0.03232, over 1768000.66 frames. ], batch size: 69, lr: 3.96e-03, grad_scale: 8.0 2022-12-08 16:07:14,298 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0921, 1.6888, 3.9793, 2.1346, 3.9272, 4.1789, 3.4544, 4.4908], device='cuda:3'), covar=tensor([0.0269, 0.3116, 0.0523, 0.2159, 0.0467, 0.0518, 0.0607, 0.0225], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0156, 0.0162, 0.0169, 0.0167, 0.0179, 0.0133, 0.0153], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 16:07:32,232 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=144104.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 16:07:54,121 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=144129.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 16:07:57,775 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.4818, 5.3579, 4.9804, 5.5373, 5.1111, 5.0146, 5.5959, 5.3236], device='cuda:3'), covar=tensor([0.0559, 0.0757, 0.0806, 0.0459, 0.0630, 0.0335, 0.0442, 0.0658], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0146, 0.0149, 0.0163, 0.0150, 0.0126, 0.0169, 0.0150], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 16:08:07,211 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.157e+02 2.224e+02 2.595e+02 3.172e+02 7.454e+02, threshold=5.190e+02, percent-clipped=6.0 2022-12-08 16:08:35,920 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=144177.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 16:08:37,521 INFO [train.py:873] (3/4) Epoch 20, batch 500, loss[loss=0.1252, simple_loss=0.1281, pruned_loss=0.06111, over 1233.00 frames. ], tot_loss[loss=0.1016, simple_loss=0.1392, pruned_loss=0.032, over 1856480.06 frames. ], batch size: 100, lr: 3.96e-03, grad_scale: 4.0 2022-12-08 16:09:34,775 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.234e+02 2.083e+02 2.620e+02 3.196e+02 5.834e+02, threshold=5.241e+02, percent-clipped=3.0 2022-12-08 16:09:56,831 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5595, 2.5499, 2.0008, 2.6613, 2.5328, 2.5328, 2.3094, 2.1386], device='cuda:3'), covar=tensor([0.1157, 0.1182, 0.2805, 0.0937, 0.1231, 0.1001, 0.1677, 0.2018], device='cuda:3'), in_proj_covar=tensor([0.0281, 0.0286, 0.0257, 0.0290, 0.0321, 0.0303, 0.0256, 0.0240], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 16:10:05,214 INFO [train.py:873] (3/4) Epoch 20, batch 600, loss[loss=0.09219, simple_loss=0.1343, pruned_loss=0.02505, over 14332.00 frames. ], tot_loss[loss=0.1016, simple_loss=0.1391, pruned_loss=0.03202, over 1894159.03 frames. ], batch size: 73, lr: 3.96e-03, grad_scale: 4.0 2022-12-08 16:10:34,321 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.12 vs. limit=2.0 2022-12-08 16:11:02,325 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.523e+02 2.036e+02 2.548e+02 2.967e+02 5.900e+02, threshold=5.097e+02, percent-clipped=2.0 2022-12-08 16:11:33,549 INFO [train.py:873] (3/4) Epoch 20, batch 700, loss[loss=0.1444, simple_loss=0.147, pruned_loss=0.07091, over 1251.00 frames. ], tot_loss[loss=0.1012, simple_loss=0.1389, pruned_loss=0.03179, over 1972037.70 frames. ], batch size: 100, lr: 3.96e-03, grad_scale: 4.0 2022-12-08 16:12:02,246 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=144411.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:12:20,073 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0133, 2.1060, 1.9278, 2.1361, 1.8109, 2.0437, 2.0746, 2.0483], device='cuda:3'), covar=tensor([0.1044, 0.1312, 0.1261, 0.0998, 0.1529, 0.0822, 0.1155, 0.1027], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0149, 0.0151, 0.0166, 0.0153, 0.0129, 0.0172, 0.0153], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 16:12:31,474 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.258e+02 1.983e+02 2.498e+02 3.051e+02 6.384e+02, threshold=4.995e+02, percent-clipped=1.0 2022-12-08 16:12:35,208 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3647, 2.9921, 3.8212, 2.7251, 2.3892, 3.4011, 1.8259, 3.3481], device='cuda:3'), covar=tensor([0.0802, 0.1032, 0.0493, 0.1707, 0.1882, 0.0764, 0.2691, 0.0937], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0105, 0.0098, 0.0102, 0.0116, 0.0093, 0.0118, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 16:12:56,260 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=144472.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:13:02,250 INFO [train.py:873] (3/4) Epoch 20, batch 800, loss[loss=0.09177, simple_loss=0.1286, pruned_loss=0.02747, over 13478.00 frames. ], tot_loss[loss=0.1015, simple_loss=0.1387, pruned_loss=0.03213, over 1991384.50 frames. ], batch size: 100, lr: 3.96e-03, grad_scale: 8.0 2022-12-08 16:13:49,901 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1581, 2.0659, 1.8224, 1.9505, 2.0864, 2.1187, 2.0891, 2.1025], device='cuda:3'), covar=tensor([0.1076, 0.0895, 0.2677, 0.2359, 0.1221, 0.1193, 0.1592, 0.1015], device='cuda:3'), in_proj_covar=tensor([0.0394, 0.0280, 0.0453, 0.0572, 0.0356, 0.0462, 0.0393, 0.0396], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 16:13:58,406 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.122e+02 2.092e+02 2.630e+02 3.168e+02 5.138e+02, threshold=5.259e+02, percent-clipped=1.0 2022-12-08 16:13:59,025 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.67 vs. limit=2.0 2022-12-08 16:14:12,220 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8322, 0.8710, 0.7116, 0.8331, 0.8611, 0.3835, 0.8162, 0.8145], device='cuda:3'), covar=tensor([0.0466, 0.0560, 0.0549, 0.0513, 0.0392, 0.0368, 0.1014, 0.0855], device='cuda:3'), in_proj_covar=tensor([0.0039, 0.0037, 0.0042, 0.0035, 0.0037, 0.0051, 0.0039, 0.0041], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-08 16:14:29,536 INFO [train.py:873] (3/4) Epoch 20, batch 900, loss[loss=0.1143, simple_loss=0.126, pruned_loss=0.05131, over 1310.00 frames. ], tot_loss[loss=0.1014, simple_loss=0.1389, pruned_loss=0.03194, over 1996572.00 frames. ], batch size: 100, lr: 3.96e-03, grad_scale: 8.0 2022-12-08 16:14:41,791 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8668, 3.0166, 4.6727, 3.5318, 4.6749, 4.5392, 4.5783, 3.9796], device='cuda:3'), covar=tensor([0.0661, 0.2813, 0.0884, 0.1559, 0.0633, 0.0866, 0.1021, 0.1713], device='cuda:3'), in_proj_covar=tensor([0.0347, 0.0305, 0.0384, 0.0295, 0.0362, 0.0319, 0.0357, 0.0294], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 16:15:27,790 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.200e+02 2.044e+02 2.378e+02 2.904e+02 6.661e+02, threshold=4.757e+02, percent-clipped=2.0 2022-12-08 16:15:47,020 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4146, 2.3856, 4.3724, 3.0024, 4.2445, 2.1337, 3.2726, 4.2429], device='cuda:3'), covar=tensor([0.0610, 0.3430, 0.0451, 0.5204, 0.0604, 0.3072, 0.1321, 0.0427], device='cuda:3'), in_proj_covar=tensor([0.0253, 0.0198, 0.0221, 0.0265, 0.0239, 0.0201, 0.0202, 0.0219], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0005], device='cuda:3') 2022-12-08 16:15:58,945 INFO [train.py:873] (3/4) Epoch 20, batch 1000, loss[loss=0.1014, simple_loss=0.1387, pruned_loss=0.03203, over 13554.00 frames. ], tot_loss[loss=0.1016, simple_loss=0.1393, pruned_loss=0.03198, over 2031149.33 frames. ], batch size: 100, lr: 3.96e-03, grad_scale: 8.0 2022-12-08 16:16:56,768 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.347e+02 1.902e+02 2.363e+02 2.821e+02 5.781e+02, threshold=4.726e+02, percent-clipped=3.0 2022-12-08 16:17:04,214 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.2395, 4.1457, 3.9538, 4.2450, 3.8767, 3.6920, 4.3183, 4.1122], device='cuda:3'), covar=tensor([0.0588, 0.0918, 0.0902, 0.0607, 0.0933, 0.0671, 0.0559, 0.0742], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0147, 0.0150, 0.0166, 0.0152, 0.0127, 0.0171, 0.0151], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 16:17:06,224 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8105, 3.4226, 2.7337, 4.0676, 3.8807, 3.9031, 3.4258, 2.8150], device='cuda:3'), covar=tensor([0.0848, 0.1384, 0.2819, 0.0565, 0.0837, 0.1184, 0.1210, 0.2559], device='cuda:3'), in_proj_covar=tensor([0.0282, 0.0286, 0.0257, 0.0290, 0.0320, 0.0303, 0.0256, 0.0240], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 16:17:17,213 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=144767.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:17:27,641 INFO [train.py:873] (3/4) Epoch 20, batch 1100, loss[loss=0.09839, simple_loss=0.1368, pruned_loss=0.03001, over 6945.00 frames. ], tot_loss[loss=0.1017, simple_loss=0.1392, pruned_loss=0.03209, over 1980762.35 frames. ], batch size: 100, lr: 3.95e-03, grad_scale: 8.0 2022-12-08 16:18:25,977 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.093e+02 2.171e+02 2.584e+02 3.315e+02 6.802e+02, threshold=5.169e+02, percent-clipped=3.0 2022-12-08 16:18:56,029 INFO [train.py:873] (3/4) Epoch 20, batch 1200, loss[loss=0.09622, simple_loss=0.1378, pruned_loss=0.02734, over 14468.00 frames. ], tot_loss[loss=0.1015, simple_loss=0.1393, pruned_loss=0.03188, over 1992556.20 frames. ], batch size: 51, lr: 3.95e-03, grad_scale: 8.0 2022-12-08 16:19:15,524 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 2022-12-08 16:19:45,061 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.43 vs. limit=2.0 2022-12-08 16:19:54,171 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.060e+02 1.946e+02 2.467e+02 2.965e+02 6.907e+02, threshold=4.934e+02, percent-clipped=1.0 2022-12-08 16:20:23,994 INFO [train.py:873] (3/4) Epoch 20, batch 1300, loss[loss=0.09697, simple_loss=0.1372, pruned_loss=0.02836, over 14176.00 frames. ], tot_loss[loss=0.1013, simple_loss=0.1389, pruned_loss=0.03183, over 1995176.94 frames. ], batch size: 35, lr: 3.95e-03, grad_scale: 8.0 2022-12-08 16:20:32,976 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.69 vs. limit=5.0 2022-12-08 16:20:39,320 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=144996.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:20:58,127 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0239, 2.9142, 2.1943, 3.0658, 2.9138, 2.9583, 2.6496, 2.3253], device='cuda:3'), covar=tensor([0.0890, 0.1279, 0.3136, 0.0929, 0.1037, 0.1046, 0.1561, 0.2706], device='cuda:3'), in_proj_covar=tensor([0.0283, 0.0288, 0.0259, 0.0292, 0.0322, 0.0304, 0.0257, 0.0242], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 16:20:58,840 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.2480, 2.3827, 4.0286, 3.0801, 4.0237, 3.8370, 3.8159, 3.3471], device='cuda:3'), covar=tensor([0.0754, 0.3290, 0.1093, 0.1589, 0.0804, 0.1027, 0.1543, 0.1707], device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0308, 0.0387, 0.0297, 0.0364, 0.0320, 0.0359, 0.0295], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 16:21:10,366 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145027.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:21:26,336 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.036e+02 2.136e+02 2.595e+02 3.096e+02 6.124e+02, threshold=5.190e+02, percent-clipped=3.0 2022-12-08 16:21:36,891 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145057.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 16:21:44,332 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145066.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:21:45,142 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=145067.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:21:55,280 INFO [train.py:873] (3/4) Epoch 20, batch 1400, loss[loss=0.1143, simple_loss=0.1163, pruned_loss=0.05614, over 1282.00 frames. ], tot_loss[loss=0.1005, simple_loss=0.1385, pruned_loss=0.03119, over 2035491.82 frames. ], batch size: 100, lr: 3.95e-03, grad_scale: 8.0 2022-12-08 16:22:03,125 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145088.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:22:15,484 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.66 vs. limit=2.0 2022-12-08 16:22:26,075 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=145115.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:22:27,104 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5268, 2.2878, 2.1565, 2.3159, 2.1645, 1.6217, 2.2277, 2.4198], device='cuda:3'), covar=tensor([0.0811, 0.0698, 0.0745, 0.1172, 0.0771, 0.0654, 0.0999, 0.0539], device='cuda:3'), in_proj_covar=tensor([0.0039, 0.0037, 0.0041, 0.0035, 0.0037, 0.0050, 0.0039, 0.0041], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-08 16:22:36,068 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145127.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:22:52,063 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.811e+01 2.056e+02 2.726e+02 3.195e+02 8.712e+02, threshold=5.453e+02, percent-clipped=3.0 2022-12-08 16:23:08,405 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.65 vs. limit=2.0 2022-12-08 16:23:20,737 INFO [train.py:873] (3/4) Epoch 20, batch 1500, loss[loss=0.09934, simple_loss=0.1393, pruned_loss=0.02969, over 14511.00 frames. ], tot_loss[loss=0.1009, simple_loss=0.1384, pruned_loss=0.03166, over 1994846.56 frames. ], batch size: 49, lr: 3.95e-03, grad_scale: 4.0 2022-12-08 16:24:19,275 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.332e+01 2.024e+02 2.567e+02 3.376e+02 1.316e+03, threshold=5.133e+02, percent-clipped=5.0 2022-12-08 16:24:48,112 INFO [train.py:873] (3/4) Epoch 20, batch 1600, loss[loss=0.08695, simple_loss=0.1329, pruned_loss=0.02051, over 14284.00 frames. ], tot_loss[loss=0.1015, simple_loss=0.1387, pruned_loss=0.03213, over 1999895.08 frames. ], batch size: 39, lr: 3.95e-03, grad_scale: 8.0 2022-12-08 16:24:53,074 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.43 vs. limit=5.0 2022-12-08 16:25:46,272 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.072e+02 1.921e+02 2.439e+02 2.878e+02 7.278e+02, threshold=4.878e+02, percent-clipped=1.0 2022-12-08 16:25:52,010 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145352.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 16:26:08,302 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145371.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:26:15,271 INFO [train.py:873] (3/4) Epoch 20, batch 1700, loss[loss=0.1193, simple_loss=0.1457, pruned_loss=0.04643, over 8648.00 frames. ], tot_loss[loss=0.1015, simple_loss=0.139, pruned_loss=0.03196, over 2013337.02 frames. ], batch size: 100, lr: 3.95e-03, grad_scale: 8.0 2022-12-08 16:26:18,717 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145383.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:26:39,167 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1351, 1.3005, 1.3072, 1.0488, 0.8345, 1.1030, 0.9292, 1.2318], device='cuda:3'), covar=tensor([0.1803, 0.2330, 0.1196, 0.2346, 0.3040, 0.1387, 0.2073, 0.1321], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0104, 0.0097, 0.0102, 0.0115, 0.0093, 0.0118, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 16:26:52,721 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145422.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:27:01,934 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145432.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:27:05,278 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7163, 3.8827, 4.0183, 3.6867, 3.9057, 3.9246, 1.6823, 3.7386], device='cuda:3'), covar=tensor([0.0379, 0.0337, 0.0288, 0.0460, 0.0303, 0.0354, 0.2926, 0.0281], device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0176, 0.0148, 0.0150, 0.0209, 0.0143, 0.0158, 0.0197], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 16:27:13,922 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.159e+01 2.091e+02 2.554e+02 3.165e+02 6.798e+02, threshold=5.109e+02, percent-clipped=6.0 2022-12-08 16:27:26,412 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1095, 1.1618, 0.9951, 1.0716, 1.1639, 0.7878, 1.0563, 1.0336], device='cuda:3'), covar=tensor([0.0669, 0.0579, 0.0592, 0.0502, 0.0424, 0.0409, 0.0836, 0.0873], device='cuda:3'), in_proj_covar=tensor([0.0039, 0.0037, 0.0042, 0.0035, 0.0037, 0.0051, 0.0039, 0.0041], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-08 16:27:42,938 INFO [train.py:873] (3/4) Epoch 20, batch 1800, loss[loss=0.1376, simple_loss=0.1359, pruned_loss=0.06961, over 1332.00 frames. ], tot_loss[loss=0.1013, simple_loss=0.139, pruned_loss=0.03179, over 2029151.78 frames. ], batch size: 100, lr: 3.94e-03, grad_scale: 8.0 2022-12-08 16:28:12,935 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.9057, 4.7458, 4.4909, 4.8837, 4.5666, 4.3481, 4.9655, 4.6458], device='cuda:3'), covar=tensor([0.0527, 0.0742, 0.0788, 0.0530, 0.0762, 0.0580, 0.0505, 0.0709], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0148, 0.0151, 0.0167, 0.0152, 0.0128, 0.0173, 0.0152], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 16:28:41,420 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.279e+02 2.270e+02 2.808e+02 3.331e+02 1.195e+03, threshold=5.616e+02, percent-clipped=5.0 2022-12-08 16:29:10,285 INFO [train.py:873] (3/4) Epoch 20, batch 1900, loss[loss=0.1044, simple_loss=0.1345, pruned_loss=0.03708, over 5960.00 frames. ], tot_loss[loss=0.102, simple_loss=0.1393, pruned_loss=0.03229, over 2050929.70 frames. ], batch size: 100, lr: 3.94e-03, grad_scale: 8.0 2022-12-08 16:29:23,492 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145594.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:29:31,001 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.67 vs. limit=2.0 2022-12-08 16:29:33,762 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2022-12-08 16:30:07,013 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145643.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:30:09,685 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.288e+02 2.010e+02 2.453e+02 2.850e+02 4.730e+02, threshold=4.905e+02, percent-clipped=0.0 2022-12-08 16:30:14,984 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=145652.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 16:30:17,522 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145655.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 16:30:38,453 INFO [train.py:873] (3/4) Epoch 20, batch 2000, loss[loss=0.1106, simple_loss=0.1457, pruned_loss=0.03773, over 10318.00 frames. ], tot_loss[loss=0.102, simple_loss=0.1394, pruned_loss=0.03226, over 2011540.13 frames. ], batch size: 100, lr: 3.94e-03, grad_scale: 8.0 2022-12-08 16:30:42,038 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=145683.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:30:52,615 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7189, 3.4225, 3.2742, 2.4292, 3.2615, 3.4173, 3.7947, 2.9160], device='cuda:3'), covar=tensor([0.0618, 0.0991, 0.0862, 0.1321, 0.0799, 0.0728, 0.0721, 0.1225], device='cuda:3'), in_proj_covar=tensor([0.0154, 0.0168, 0.0141, 0.0125, 0.0145, 0.0158, 0.0140, 0.0143], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0007, 0.0006, 0.0006], device='cuda:3') 2022-12-08 16:30:56,733 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=145700.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:31:00,306 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145704.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:31:08,607 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.7571, 2.3645, 4.7327, 3.2552, 4.5461, 2.2896, 3.5182, 4.6385], device='cuda:3'), covar=tensor([0.0423, 0.3392, 0.0340, 0.5124, 0.0466, 0.2967, 0.1232, 0.0255], device='cuda:3'), in_proj_covar=tensor([0.0254, 0.0199, 0.0221, 0.0266, 0.0241, 0.0202, 0.0202, 0.0220], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0005], device='cuda:3') 2022-12-08 16:31:16,206 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=145722.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:31:20,718 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145727.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:31:23,991 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=145731.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:31:37,254 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.155e+02 1.878e+02 2.490e+02 3.077e+02 5.719e+02, threshold=4.981e+02, percent-clipped=2.0 2022-12-08 16:31:37,531 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3372, 2.2221, 3.2147, 2.4611, 3.1599, 3.0372, 2.9575, 2.6570], device='cuda:3'), covar=tensor([0.0829, 0.2973, 0.1008, 0.1815, 0.0810, 0.1276, 0.1180, 0.1568], device='cuda:3'), in_proj_covar=tensor([0.0354, 0.0310, 0.0392, 0.0301, 0.0368, 0.0323, 0.0361, 0.0298], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 16:31:58,229 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=145770.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:32:01,909 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9412, 1.8907, 2.2797, 1.8063, 1.8911, 1.7257, 1.7853, 1.2670], device='cuda:3'), covar=tensor([0.0249, 0.0498, 0.0178, 0.0382, 0.0249, 0.0324, 0.0329, 0.0564], device='cuda:3'), in_proj_covar=tensor([0.0025, 0.0024, 0.0022, 0.0023, 0.0023, 0.0036, 0.0030, 0.0034], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:3') 2022-12-08 16:32:05,946 INFO [train.py:873] (3/4) Epoch 20, batch 2100, loss[loss=0.09051, simple_loss=0.1288, pruned_loss=0.02611, over 13891.00 frames. ], tot_loss[loss=0.1016, simple_loss=0.139, pruned_loss=0.03203, over 2105575.92 frames. ], batch size: 23, lr: 3.94e-03, grad_scale: 8.0 2022-12-08 16:32:43,299 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145821.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 16:33:04,938 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.307e+02 2.012e+02 2.500e+02 3.223e+02 4.888e+02, threshold=5.000e+02, percent-clipped=0.0 2022-12-08 16:33:33,925 INFO [train.py:873] (3/4) Epoch 20, batch 2200, loss[loss=0.1053, simple_loss=0.1348, pruned_loss=0.03789, over 6937.00 frames. ], tot_loss[loss=0.1021, simple_loss=0.1395, pruned_loss=0.03239, over 2125141.61 frames. ], batch size: 100, lr: 3.94e-03, grad_scale: 8.0 2022-12-08 16:33:36,646 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145882.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 16:33:39,389 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9426, 1.7182, 3.7258, 3.4396, 3.5320, 3.7823, 3.0634, 3.7340], device='cuda:3'), covar=tensor([0.1639, 0.1598, 0.0141, 0.0285, 0.0299, 0.0160, 0.0361, 0.0152], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0157, 0.0132, 0.0171, 0.0150, 0.0144, 0.0126, 0.0126], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 16:34:09,954 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1434, 3.6231, 3.5302, 3.4429, 2.5386, 3.6561, 3.3789, 1.7941], device='cuda:3'), covar=tensor([0.1253, 0.0707, 0.0700, 0.0845, 0.0928, 0.0415, 0.0839, 0.1752], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0093, 0.0072, 0.0077, 0.0102, 0.0092, 0.0104, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0006, 0.0006, 0.0007, 0.0006, 0.0007, 0.0006], device='cuda:3') 2022-12-08 16:34:32,821 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.066e+02 2.107e+02 2.702e+02 3.321e+02 7.589e+02, threshold=5.404e+02, percent-clipped=7.0 2022-12-08 16:34:36,613 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145950.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 16:34:52,486 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.2986, 2.5533, 4.0663, 2.9627, 4.0876, 3.8904, 3.8275, 3.4159], device='cuda:3'), covar=tensor([0.0651, 0.3207, 0.0946, 0.1764, 0.0760, 0.1102, 0.1442, 0.1706], device='cuda:3'), in_proj_covar=tensor([0.0348, 0.0305, 0.0387, 0.0297, 0.0362, 0.0317, 0.0355, 0.0293], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 16:35:01,455 INFO [train.py:873] (3/4) Epoch 20, batch 2300, loss[loss=0.1369, simple_loss=0.1406, pruned_loss=0.06661, over 1247.00 frames. ], tot_loss[loss=0.102, simple_loss=0.1392, pruned_loss=0.03239, over 2025581.90 frames. ], batch size: 100, lr: 3.94e-03, grad_scale: 8.0 2022-12-08 16:35:19,658 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145999.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:35:43,563 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2264, 2.2091, 3.2076, 3.3452, 3.1602, 2.2039, 3.2358, 2.4892], device='cuda:3'), covar=tensor([0.0535, 0.1473, 0.0863, 0.0580, 0.0685, 0.1978, 0.0568, 0.1213], device='cuda:3'), in_proj_covar=tensor([0.0296, 0.0263, 0.0378, 0.0333, 0.0273, 0.0309, 0.0315, 0.0280], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 16:35:44,243 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146027.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:36:00,973 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.336e+02 2.169e+02 2.442e+02 2.958e+02 6.970e+02, threshold=4.884e+02, percent-clipped=2.0 2022-12-08 16:36:26,855 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=146075.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:36:30,341 INFO [train.py:873] (3/4) Epoch 20, batch 2400, loss[loss=0.09161, simple_loss=0.1248, pruned_loss=0.02923, over 5960.00 frames. ], tot_loss[loss=0.1027, simple_loss=0.1396, pruned_loss=0.03286, over 1952602.58 frames. ], batch size: 100, lr: 3.94e-03, grad_scale: 8.0 2022-12-08 16:37:11,170 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.8318, 2.3642, 3.6523, 2.8196, 3.6548, 3.5064, 3.4822, 2.9919], device='cuda:3'), covar=tensor([0.0899, 0.3024, 0.0979, 0.1801, 0.0917, 0.1186, 0.1247, 0.1820], device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0307, 0.0387, 0.0298, 0.0363, 0.0319, 0.0357, 0.0294], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 16:37:29,040 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.227e+02 1.997e+02 2.339e+02 2.913e+02 8.016e+02, threshold=4.679e+02, percent-clipped=1.0 2022-12-08 16:37:32,263 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=146149.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:37:56,523 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=146177.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 16:37:58,097 INFO [train.py:873] (3/4) Epoch 20, batch 2500, loss[loss=0.1468, simple_loss=0.14, pruned_loss=0.07681, over 1239.00 frames. ], tot_loss[loss=0.1006, simple_loss=0.1385, pruned_loss=0.03134, over 2001383.27 frames. ], batch size: 100, lr: 3.93e-03, grad_scale: 8.0 2022-12-08 16:38:26,587 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=146210.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:38:58,168 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.933e+01 2.087e+02 2.444e+02 3.038e+02 5.934e+02, threshold=4.888e+02, percent-clipped=3.0 2022-12-08 16:39:01,876 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146250.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:39:07,149 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1308, 2.7250, 2.8414, 2.0379, 2.6099, 2.8085, 3.1241, 2.5927], device='cuda:3'), covar=tensor([0.0708, 0.0813, 0.0913, 0.1270, 0.0948, 0.0758, 0.0606, 0.1121], device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0169, 0.0142, 0.0126, 0.0145, 0.0157, 0.0139, 0.0143], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0007, 0.0006, 0.0006], device='cuda:3') 2022-12-08 16:39:22,242 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8899, 1.4674, 3.3289, 3.0152, 3.1619, 3.3241, 2.6582, 3.3215], device='cuda:3'), covar=tensor([0.1644, 0.1745, 0.0168, 0.0374, 0.0330, 0.0192, 0.0400, 0.0185], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0156, 0.0132, 0.0170, 0.0149, 0.0143, 0.0126, 0.0125], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 16:39:27,232 INFO [train.py:873] (3/4) Epoch 20, batch 2600, loss[loss=0.1131, simple_loss=0.1468, pruned_loss=0.03969, over 14315.00 frames. ], tot_loss[loss=0.1007, simple_loss=0.1384, pruned_loss=0.03146, over 2007737.54 frames. ], batch size: 46, lr: 3.93e-03, grad_scale: 8.0 2022-12-08 16:39:32,652 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3848, 2.4311, 2.5820, 2.4789, 2.5107, 2.1380, 1.5137, 2.2887], device='cuda:3'), covar=tensor([0.0698, 0.0586, 0.0429, 0.0485, 0.0461, 0.1498, 0.2418, 0.0449], device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0177, 0.0149, 0.0151, 0.0209, 0.0144, 0.0158, 0.0197], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 16:39:43,724 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=146298.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:39:44,695 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146299.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:40:25,675 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.546e+01 2.007e+02 2.593e+02 2.996e+02 8.778e+02, threshold=5.186e+02, percent-clipped=5.0 2022-12-08 16:40:26,682 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=146347.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:40:54,845 INFO [train.py:873] (3/4) Epoch 20, batch 2700, loss[loss=0.09768, simple_loss=0.1385, pruned_loss=0.02841, over 14295.00 frames. ], tot_loss[loss=0.1018, simple_loss=0.1388, pruned_loss=0.03241, over 1985924.19 frames. ], batch size: 25, lr: 3.93e-03, grad_scale: 8.0 2022-12-08 16:41:01,557 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 2022-12-08 16:41:34,929 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4320, 1.5532, 3.4550, 1.7003, 3.2564, 3.5253, 2.4825, 3.7345], device='cuda:3'), covar=tensor([0.0239, 0.3160, 0.0433, 0.2217, 0.0961, 0.0419, 0.0913, 0.0211], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0158, 0.0163, 0.0170, 0.0168, 0.0181, 0.0134, 0.0154], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 16:41:53,288 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.210e+02 2.024e+02 2.634e+02 3.181e+02 1.190e+03, threshold=5.268e+02, percent-clipped=3.0 2022-12-08 16:41:59,814 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.0066, 1.5169, 3.8684, 1.8751, 3.9897, 4.0574, 3.0864, 4.3710], device='cuda:3'), covar=tensor([0.0233, 0.3242, 0.0409, 0.2129, 0.0394, 0.0431, 0.0681, 0.0160], device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0159, 0.0164, 0.0171, 0.0169, 0.0182, 0.0135, 0.0154], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 16:42:02,852 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.2624, 4.0499, 3.9327, 4.3304, 4.0874, 3.7841, 4.3293, 3.6811], device='cuda:3'), covar=tensor([0.0450, 0.0992, 0.0477, 0.0427, 0.0805, 0.1181, 0.0558, 0.0557], device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0275, 0.0204, 0.0201, 0.0186, 0.0162, 0.0289, 0.0171], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 16:42:03,941 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2022-12-08 16:42:20,396 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146477.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 16:42:21,927 INFO [train.py:873] (3/4) Epoch 20, batch 2800, loss[loss=0.1452, simple_loss=0.1457, pruned_loss=0.07242, over 1168.00 frames. ], tot_loss[loss=0.1021, simple_loss=0.1391, pruned_loss=0.03259, over 1922601.59 frames. ], batch size: 100, lr: 3.93e-03, grad_scale: 8.0 2022-12-08 16:42:25,739 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.3908, 3.2381, 3.9253, 2.8246, 2.4772, 3.3611, 1.9309, 3.4829], device='cuda:3'), covar=tensor([0.0806, 0.0702, 0.0585, 0.2458, 0.1653, 0.0707, 0.2473, 0.0656], device='cuda:3'), in_proj_covar=tensor([0.0088, 0.0102, 0.0096, 0.0101, 0.0114, 0.0092, 0.0116, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 16:42:44,677 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=146505.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:43:02,233 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=146525.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 16:43:20,289 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.384e+02 1.986e+02 2.533e+02 3.212e+02 8.104e+02, threshold=5.065e+02, percent-clipped=2.0 2022-12-08 16:43:26,313 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.34 vs. limit=5.0 2022-12-08 16:43:48,734 INFO [train.py:873] (3/4) Epoch 20, batch 2900, loss[loss=0.09345, simple_loss=0.1105, pruned_loss=0.03817, over 2593.00 frames. ], tot_loss[loss=0.1025, simple_loss=0.1393, pruned_loss=0.03285, over 1954530.02 frames. ], batch size: 100, lr: 3.93e-03, grad_scale: 8.0 2022-12-08 16:43:58,466 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([0.8583, 0.8529, 0.7476, 0.8453, 0.8567, 0.4931, 0.7565, 0.8326], device='cuda:3'), covar=tensor([0.0497, 0.0565, 0.0751, 0.0527, 0.0348, 0.0371, 0.0980, 0.0718], device='cuda:3'), in_proj_covar=tensor([0.0039, 0.0037, 0.0042, 0.0034, 0.0037, 0.0051, 0.0039, 0.0041], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-08 16:43:59,402 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=146591.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:44:30,227 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=146626.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:44:47,463 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.360e+02 2.123e+02 2.667e+02 3.296e+02 6.992e+02, threshold=5.334e+02, percent-clipped=2.0 2022-12-08 16:44:52,790 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=146652.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:45:16,679 INFO [train.py:873] (3/4) Epoch 20, batch 3000, loss[loss=0.08485, simple_loss=0.1228, pruned_loss=0.02347, over 5944.00 frames. ], tot_loss[loss=0.1025, simple_loss=0.1397, pruned_loss=0.03264, over 2050564.48 frames. ], batch size: 100, lr: 3.93e-03, grad_scale: 8.0 2022-12-08 16:45:16,679 INFO [train.py:896] (3/4) Computing validation loss 2022-12-08 16:45:25,137 INFO [train.py:905] (3/4) Epoch 20, validation: loss=0.1444, simple_loss=0.1794, pruned_loss=0.05469, over 857387.00 frames. 2022-12-08 16:45:25,137 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-08 16:45:31,156 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7077, 3.5447, 3.2390, 3.4150, 3.6203, 3.6473, 3.6633, 3.6802], device='cuda:3'), covar=tensor([0.0765, 0.0594, 0.2213, 0.2304, 0.0789, 0.0982, 0.0990, 0.0791], device='cuda:3'), in_proj_covar=tensor([0.0397, 0.0282, 0.0453, 0.0573, 0.0358, 0.0464, 0.0396, 0.0398], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 16:45:32,051 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=146687.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 16:46:23,235 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.146e+02 2.030e+02 2.519e+02 3.160e+02 5.604e+02, threshold=5.039e+02, percent-clipped=1.0 2022-12-08 16:46:52,205 INFO [train.py:873] (3/4) Epoch 20, batch 3100, loss[loss=0.1025, simple_loss=0.1471, pruned_loss=0.02898, over 14661.00 frames. ], tot_loss[loss=0.1009, simple_loss=0.1385, pruned_loss=0.03166, over 2074837.88 frames. ], batch size: 33, lr: 3.93e-03, grad_scale: 8.0 2022-12-08 16:47:14,834 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146805.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:47:38,112 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.31 vs. limit=5.0 2022-12-08 16:47:49,268 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=146844.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:47:50,641 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.355e+02 2.017e+02 2.599e+02 3.064e+02 9.305e+02, threshold=5.198e+02, percent-clipped=5.0 2022-12-08 16:47:56,839 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=146853.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:48:19,589 INFO [train.py:873] (3/4) Epoch 20, batch 3200, loss[loss=0.1022, simple_loss=0.1349, pruned_loss=0.0348, over 6969.00 frames. ], tot_loss[loss=0.1011, simple_loss=0.1384, pruned_loss=0.03189, over 1988296.02 frames. ], batch size: 100, lr: 3.93e-03, grad_scale: 8.0 2022-12-08 16:48:43,347 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=146905.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:49:16,515 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.2504, 2.1833, 2.2131, 2.2163, 2.2313, 1.5203, 1.9269, 2.1960], device='cuda:3'), covar=tensor([0.0582, 0.0777, 0.0491, 0.1140, 0.0545, 0.0781, 0.1018, 0.1199], device='cuda:3'), in_proj_covar=tensor([0.0039, 0.0037, 0.0042, 0.0034, 0.0037, 0.0051, 0.0039, 0.0041], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-08 16:49:16,818 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2022-12-08 16:49:18,825 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.275e+02 2.042e+02 2.442e+02 3.021e+02 7.374e+02, threshold=4.884e+02, percent-clipped=4.0 2022-12-08 16:49:19,701 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=146947.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:49:40,670 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.47 vs. limit=5.0 2022-12-08 16:49:47,631 INFO [train.py:873] (3/4) Epoch 20, batch 3300, loss[loss=0.1044, simple_loss=0.1411, pruned_loss=0.03387, over 14214.00 frames. ], tot_loss[loss=0.1006, simple_loss=0.1377, pruned_loss=0.03176, over 1910373.17 frames. ], batch size: 94, lr: 3.92e-03, grad_scale: 8.0 2022-12-08 16:49:50,254 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=146982.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 16:50:02,396 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2022-12-08 16:50:46,822 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5243, 1.6652, 2.6637, 2.2080, 2.5320, 1.6860, 2.2397, 2.5296], device='cuda:3'), covar=tensor([0.1640, 0.3518, 0.0813, 0.2465, 0.1260, 0.2952, 0.0931, 0.1141], device='cuda:3'), in_proj_covar=tensor([0.0256, 0.0199, 0.0224, 0.0267, 0.0242, 0.0204, 0.0203, 0.0223], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0005], device='cuda:3') 2022-12-08 16:50:47,402 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.340e+02 2.122e+02 2.526e+02 3.271e+02 7.839e+02, threshold=5.052e+02, percent-clipped=6.0 2022-12-08 16:51:14,964 INFO [train.py:873] (3/4) Epoch 20, batch 3400, loss[loss=0.1339, simple_loss=0.1309, pruned_loss=0.06845, over 1244.00 frames. ], tot_loss[loss=0.1013, simple_loss=0.1381, pruned_loss=0.03224, over 1942860.77 frames. ], batch size: 100, lr: 3.92e-03, grad_scale: 4.0 2022-12-08 16:51:44,509 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2022-12-08 16:52:16,822 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.131e+02 1.939e+02 2.278e+02 2.850e+02 6.486e+02, threshold=4.556e+02, percent-clipped=1.0 2022-12-08 16:52:19,567 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.2682, 1.3561, 1.2366, 1.3481, 1.4580, 0.9018, 1.1069, 1.1692], device='cuda:3'), covar=tensor([0.1030, 0.0782, 0.0787, 0.0623, 0.0509, 0.0928, 0.1100, 0.0960], device='cuda:3'), in_proj_covar=tensor([0.0039, 0.0037, 0.0042, 0.0035, 0.0037, 0.0051, 0.0039, 0.0041], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-08 16:52:45,346 INFO [train.py:873] (3/4) Epoch 20, batch 3500, loss[loss=0.1001, simple_loss=0.1389, pruned_loss=0.03061, over 14276.00 frames. ], tot_loss[loss=0.101, simple_loss=0.1378, pruned_loss=0.03209, over 1861964.38 frames. ], batch size: 80, lr: 3.92e-03, grad_scale: 4.0 2022-12-08 16:53:03,627 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147200.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:53:33,180 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2022-12-08 16:53:39,338 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147241.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:53:44,666 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.154e+02 2.013e+02 2.574e+02 3.119e+02 7.865e+02, threshold=5.147e+02, percent-clipped=11.0 2022-12-08 16:53:44,871 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147247.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:54:12,622 INFO [train.py:873] (3/4) Epoch 20, batch 3600, loss[loss=0.09015, simple_loss=0.1384, pruned_loss=0.02093, over 14409.00 frames. ], tot_loss[loss=0.1008, simple_loss=0.138, pruned_loss=0.03183, over 1886286.34 frames. ], batch size: 41, lr: 3.92e-03, grad_scale: 8.0 2022-12-08 16:54:12,725 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4754, 4.1910, 4.0949, 4.5248, 4.2032, 3.9977, 4.5436, 3.8175], device='cuda:3'), covar=tensor([0.0434, 0.0901, 0.0429, 0.0400, 0.0787, 0.0900, 0.0440, 0.0518], device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0276, 0.0204, 0.0201, 0.0187, 0.0161, 0.0289, 0.0171], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 16:54:15,377 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147282.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:54:26,918 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=147295.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:54:34,071 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147302.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:54:40,978 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147310.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:54:58,636 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=147330.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:55:04,993 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.5681, 3.6871, 3.8083, 3.4278, 3.7246, 3.7189, 1.6784, 3.5427], device='cuda:3'), covar=tensor([0.0385, 0.0369, 0.0360, 0.0528, 0.0339, 0.0461, 0.3059, 0.0311], device='cuda:3'), in_proj_covar=tensor([0.0179, 0.0179, 0.0150, 0.0153, 0.0213, 0.0145, 0.0160, 0.0199], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 16:55:14,167 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.052e+02 2.112e+02 2.582e+02 3.106e+02 5.993e+02, threshold=5.164e+02, percent-clipped=4.0 2022-12-08 16:55:34,702 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147371.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:55:41,612 INFO [train.py:873] (3/4) Epoch 20, batch 3700, loss[loss=0.0984, simple_loss=0.1374, pruned_loss=0.02967, over 14271.00 frames. ], tot_loss[loss=0.1018, simple_loss=0.1389, pruned_loss=0.03233, over 1959829.20 frames. ], batch size: 76, lr: 3.92e-03, grad_scale: 4.0 2022-12-08 16:56:06,101 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147407.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:56:19,255 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.7333, 2.7837, 2.6329, 2.8718, 2.4660, 2.6157, 2.8164, 2.7051], device='cuda:3'), covar=tensor([0.0846, 0.1248, 0.1036, 0.0883, 0.1270, 0.0834, 0.0922, 0.1131], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0150, 0.0151, 0.0166, 0.0152, 0.0128, 0.0174, 0.0153], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 16:56:26,782 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6037, 3.3562, 4.1489, 3.0742, 2.4538, 3.5453, 2.1271, 3.6413], device='cuda:3'), covar=tensor([0.0937, 0.1164, 0.0420, 0.1821, 0.2022, 0.0939, 0.2629, 0.0875], device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0102, 0.0095, 0.0101, 0.0115, 0.0093, 0.0116, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 16:56:27,576 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6167, 3.4249, 3.1587, 3.3100, 3.5286, 3.5459, 3.5340, 3.5801], device='cuda:3'), covar=tensor([0.0831, 0.0606, 0.1928, 0.2269, 0.0827, 0.0918, 0.1088, 0.0848], device='cuda:3'), in_proj_covar=tensor([0.0397, 0.0286, 0.0456, 0.0575, 0.0362, 0.0467, 0.0404, 0.0402], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004], device='cuda:3') 2022-12-08 16:56:29,310 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147434.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:56:41,247 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.389e+02 2.266e+02 2.765e+02 3.309e+02 4.888e+02, threshold=5.530e+02, percent-clipped=0.0 2022-12-08 16:56:50,130 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.6151, 2.7083, 2.8144, 2.7351, 2.7687, 2.4898, 1.5729, 2.5107], device='cuda:3'), covar=tensor([0.0602, 0.0491, 0.0416, 0.0477, 0.0399, 0.1104, 0.2631, 0.0434], device='cuda:3'), in_proj_covar=tensor([0.0179, 0.0179, 0.0151, 0.0153, 0.0212, 0.0145, 0.0160, 0.0199], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 16:56:58,736 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147468.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:57:07,330 INFO [train.py:873] (3/4) Epoch 20, batch 3800, loss[loss=0.1252, simple_loss=0.1508, pruned_loss=0.04976, over 12744.00 frames. ], tot_loss[loss=0.1027, simple_loss=0.1394, pruned_loss=0.03298, over 1951407.99 frames. ], batch size: 100, lr: 3.92e-03, grad_scale: 4.0 2022-12-08 16:57:21,582 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147495.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:57:25,973 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147500.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:58:07,481 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.824e+01 2.099e+02 2.660e+02 3.307e+02 5.236e+02, threshold=5.319e+02, percent-clipped=0.0 2022-12-08 16:58:07,559 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=147548.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:58:19,739 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2022-12-08 16:58:34,640 INFO [train.py:873] (3/4) Epoch 20, batch 3900, loss[loss=0.1101, simple_loss=0.1174, pruned_loss=0.0514, over 1275.00 frames. ], tot_loss[loss=0.1023, simple_loss=0.139, pruned_loss=0.03275, over 1970910.35 frames. ], batch size: 100, lr: 3.92e-03, grad_scale: 4.0 2022-12-08 16:58:50,282 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147597.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:59:35,033 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.207e+02 1.905e+02 2.298e+02 3.086e+02 1.551e+03, threshold=4.597e+02, percent-clipped=3.0 2022-12-08 16:59:45,226 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147659.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 16:59:51,042 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147666.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:00:01,898 INFO [train.py:873] (3/4) Epoch 20, batch 4000, loss[loss=0.093, simple_loss=0.1345, pruned_loss=0.02575, over 14140.00 frames. ], tot_loss[loss=0.1019, simple_loss=0.1389, pruned_loss=0.03244, over 1986676.40 frames. ], batch size: 99, lr: 3.91e-03, grad_scale: 8.0 2022-12-08 17:00:18,383 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0502, 1.8944, 2.1607, 2.1907, 1.8892, 1.8384, 1.6932, 1.8293], device='cuda:3'), covar=tensor([0.0370, 0.0781, 0.0373, 0.0245, 0.0447, 0.0662, 0.0555, 0.0535], device='cuda:3'), in_proj_covar=tensor([0.0025, 0.0024, 0.0022, 0.0024, 0.0023, 0.0036, 0.0030, 0.0034], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:3') 2022-12-08 17:00:37,470 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147720.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:00:39,610 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9501, 1.9651, 4.2345, 3.8782, 3.9145, 4.2716, 3.6753, 4.2560], device='cuda:3'), covar=tensor([0.1607, 0.1437, 0.0115, 0.0243, 0.0233, 0.0138, 0.0214, 0.0114], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0157, 0.0131, 0.0170, 0.0147, 0.0142, 0.0124, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 17:00:58,450 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.09 vs. limit=2.0 2022-12-08 17:01:01,455 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.046e+02 2.183e+02 2.642e+02 3.333e+02 1.015e+03, threshold=5.284e+02, percent-clipped=7.0 2022-12-08 17:01:14,058 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147763.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:01:27,732 INFO [train.py:873] (3/4) Epoch 20, batch 4100, loss[loss=0.1348, simple_loss=0.1415, pruned_loss=0.06401, over 2611.00 frames. ], tot_loss[loss=0.1016, simple_loss=0.1389, pruned_loss=0.03214, over 2012417.90 frames. ], batch size: 100, lr: 3.91e-03, grad_scale: 8.0 2022-12-08 17:01:37,673 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147790.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:01:59,421 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147815.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:02:09,312 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0432, 1.9379, 4.4936, 4.1793, 4.0342, 4.5755, 4.0956, 4.5485], device='cuda:3'), covar=tensor([0.1556, 0.1560, 0.0120, 0.0243, 0.0275, 0.0125, 0.0215, 0.0120], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0157, 0.0132, 0.0171, 0.0148, 0.0142, 0.0125, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 17:02:20,320 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1461, 1.3157, 1.2573, 1.0697, 0.8542, 1.1003, 0.9060, 1.1958], device='cuda:3'), covar=tensor([0.1714, 0.2663, 0.1373, 0.2388, 0.3225, 0.1365, 0.1861, 0.1589], device='cuda:3'), in_proj_covar=tensor([0.0088, 0.0103, 0.0097, 0.0102, 0.0116, 0.0094, 0.0116, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2022-12-08 17:02:27,883 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.192e+02 2.007e+02 2.574e+02 3.284e+02 8.087e+02, threshold=5.147e+02, percent-clipped=4.0 2022-12-08 17:02:36,121 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147857.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:02:52,257 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147876.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:02:54,613 INFO [train.py:873] (3/4) Epoch 20, batch 4200, loss[loss=0.09525, simple_loss=0.1374, pruned_loss=0.02656, over 13836.00 frames. ], tot_loss[loss=0.1007, simple_loss=0.1383, pruned_loss=0.03155, over 1969136.43 frames. ], batch size: 20, lr: 3.91e-03, grad_scale: 8.0 2022-12-08 17:02:57,191 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3885, 2.1874, 2.4080, 1.6805, 1.9952, 2.4471, 2.3642, 2.0799], device='cuda:3'), covar=tensor([0.0767, 0.0583, 0.0860, 0.1181, 0.1251, 0.0867, 0.0759, 0.1225], device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0169, 0.0142, 0.0126, 0.0147, 0.0157, 0.0141, 0.0144], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0007, 0.0006, 0.0006], device='cuda:3') 2022-12-08 17:03:10,375 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147897.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:03:22,203 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147911.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:03:28,194 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147918.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:03:43,988 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8320, 1.7691, 1.8817, 1.9729, 1.9657, 1.2266, 1.4018, 1.8008], device='cuda:3'), covar=tensor([0.0647, 0.0687, 0.0512, 0.0526, 0.0512, 0.0830, 0.1067, 0.0576], device='cuda:3'), in_proj_covar=tensor([0.0039, 0.0038, 0.0043, 0.0035, 0.0038, 0.0052, 0.0040, 0.0042], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-08 17:03:44,784 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8962, 2.3467, 2.6475, 1.6911, 2.3789, 2.6363, 2.8597, 2.2801], device='cuda:3'), covar=tensor([0.0717, 0.0822, 0.1023, 0.1520, 0.1096, 0.0782, 0.0723, 0.1377], device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0169, 0.0142, 0.0126, 0.0147, 0.0158, 0.0141, 0.0144], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0007, 0.0006, 0.0006], device='cuda:3') 2022-12-08 17:03:52,192 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=147945.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:03:55,541 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.345e+02 1.943e+02 2.282e+02 3.039e+02 5.429e+02, threshold=4.564e+02, percent-clipped=1.0 2022-12-08 17:04:10,355 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147966.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:04:15,456 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147972.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:04:21,716 INFO [train.py:873] (3/4) Epoch 20, batch 4300, loss[loss=0.1116, simple_loss=0.12, pruned_loss=0.05162, over 1281.00 frames. ], tot_loss[loss=0.1023, simple_loss=0.1388, pruned_loss=0.03288, over 1868685.14 frames. ], batch size: 100, lr: 3.91e-03, grad_scale: 4.0 2022-12-08 17:04:52,067 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=148014.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:04:52,941 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148015.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:04:58,980 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 2022-12-08 17:05:00,796 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 2022-12-08 17:05:22,415 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.032e+02 2.071e+02 2.502e+02 3.240e+02 4.957e+02, threshold=5.004e+02, percent-clipped=3.0 2022-12-08 17:05:34,995 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148063.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:05:40,981 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2022-12-08 17:05:48,147 INFO [train.py:873] (3/4) Epoch 20, batch 4400, loss[loss=0.1064, simple_loss=0.1462, pruned_loss=0.03329, over 14255.00 frames. ], tot_loss[loss=0.103, simple_loss=0.1395, pruned_loss=0.03329, over 1880343.51 frames. ], batch size: 25, lr: 3.91e-03, grad_scale: 8.0 2022-12-08 17:05:58,199 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148090.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:06:16,514 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=148111.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:06:39,830 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=148138.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:06:49,733 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.267e+02 2.050e+02 2.451e+02 3.050e+02 8.671e+02, threshold=4.901e+02, percent-clipped=4.0 2022-12-08 17:07:08,006 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148171.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:07:15,617 INFO [train.py:873] (3/4) Epoch 20, batch 4500, loss[loss=0.09274, simple_loss=0.1289, pruned_loss=0.02826, over 5970.00 frames. ], tot_loss[loss=0.1017, simple_loss=0.1387, pruned_loss=0.03234, over 1918661.61 frames. ], batch size: 100, lr: 3.91e-03, grad_scale: 4.0 2022-12-08 17:07:29,085 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.1935, 1.1612, 1.2141, 1.0820, 1.0487, 0.7824, 0.9288, 0.9792], device='cuda:3'), covar=tensor([0.0215, 0.0215, 0.0189, 0.0222, 0.0229, 0.0371, 0.0266, 0.0356], device='cuda:3'), in_proj_covar=tensor([0.0025, 0.0024, 0.0022, 0.0024, 0.0023, 0.0036, 0.0030, 0.0034], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:3') 2022-12-08 17:07:45,070 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148213.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:08:00,261 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.8759, 4.7079, 4.3031, 4.5342, 4.6240, 4.8528, 4.8670, 4.8748], device='cuda:3'), covar=tensor([0.0721, 0.0399, 0.1959, 0.2520, 0.0734, 0.0769, 0.0678, 0.0693], device='cuda:3'), in_proj_covar=tensor([0.0401, 0.0285, 0.0457, 0.0578, 0.0365, 0.0470, 0.0401, 0.0403], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0004], device='cuda:3') 2022-12-08 17:08:17,146 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.374e+02 2.029e+02 2.580e+02 3.112e+02 6.429e+02, threshold=5.160e+02, percent-clipped=2.0 2022-12-08 17:08:32,472 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148267.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:08:42,532 INFO [train.py:873] (3/4) Epoch 20, batch 4600, loss[loss=0.09504, simple_loss=0.1261, pruned_loss=0.03198, over 4986.00 frames. ], tot_loss[loss=0.1019, simple_loss=0.1391, pruned_loss=0.03237, over 1960943.93 frames. ], batch size: 100, lr: 3.91e-03, grad_scale: 4.0 2022-12-08 17:08:49,600 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148287.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:09:13,751 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148315.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:09:16,349 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1985, 2.4368, 2.4916, 2.5137, 2.1094, 2.5164, 2.3529, 1.5810], device='cuda:3'), covar=tensor([0.0918, 0.0841, 0.0732, 0.0443, 0.0885, 0.0699, 0.1154, 0.1655], device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0094, 0.0073, 0.0078, 0.0101, 0.0092, 0.0104, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0006, 0.0006, 0.0007, 0.0006, 0.0007, 0.0006], device='cuda:3') 2022-12-08 17:09:24,112 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4412, 2.3916, 3.3958, 3.5685, 3.3693, 2.3405, 3.4228, 2.6624], device='cuda:3'), covar=tensor([0.0587, 0.1333, 0.1105, 0.0595, 0.0697, 0.1966, 0.0541, 0.1225], device='cuda:3'), in_proj_covar=tensor([0.0296, 0.0263, 0.0379, 0.0335, 0.0273, 0.0312, 0.0317, 0.0282], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 17:09:35,122 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.3505, 5.0197, 4.8900, 5.3517, 4.9180, 4.5726, 5.3383, 4.5007], device='cuda:3'), covar=tensor([0.0357, 0.0891, 0.0434, 0.0422, 0.0820, 0.0471, 0.0469, 0.0502], device='cuda:3'), in_proj_covar=tensor([0.0182, 0.0276, 0.0205, 0.0203, 0.0187, 0.0164, 0.0293, 0.0172], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0004, 0.0002], device='cuda:3') 2022-12-08 17:09:42,764 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148348.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:09:44,260 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.292e+02 2.179e+02 2.688e+02 3.360e+02 6.935e+02, threshold=5.375e+02, percent-clipped=4.0 2022-12-08 17:09:55,131 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=148363.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:10:08,872 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1543, 1.8611, 3.2054, 2.4768, 3.1462, 1.9084, 2.5536, 3.1395], device='cuda:3'), covar=tensor([0.0877, 0.3800, 0.0833, 0.4036, 0.0967, 0.2898, 0.1403, 0.0688], device='cuda:3'), in_proj_covar=tensor([0.0250, 0.0197, 0.0221, 0.0263, 0.0239, 0.0199, 0.0200, 0.0220], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0005], device='cuda:3') 2022-12-08 17:10:09,364 INFO [train.py:873] (3/4) Epoch 20, batch 4700, loss[loss=0.134, simple_loss=0.1605, pruned_loss=0.05378, over 8599.00 frames. ], tot_loss[loss=0.1017, simple_loss=0.1393, pruned_loss=0.03207, over 2054981.69 frames. ], batch size: 100, lr: 3.91e-03, grad_scale: 4.0 2022-12-08 17:10:52,694 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148429.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:11:10,544 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.275e+02 2.097e+02 2.538e+02 3.093e+02 5.864e+02, threshold=5.077e+02, percent-clipped=2.0 2022-12-08 17:11:28,968 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148471.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:11:35,538 INFO [train.py:873] (3/4) Epoch 20, batch 4800, loss[loss=0.114, simple_loss=0.1436, pruned_loss=0.04217, over 5998.00 frames. ], tot_loss[loss=0.1017, simple_loss=0.1391, pruned_loss=0.03219, over 1937752.57 frames. ], batch size: 100, lr: 3.90e-03, grad_scale: 8.0 2022-12-08 17:11:45,398 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148490.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:11:52,327 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148498.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:11:56,324 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4482, 2.6620, 4.4668, 4.5630, 4.2483, 2.5897, 4.5537, 3.4124], device='cuda:3'), covar=tensor([0.0415, 0.1227, 0.0758, 0.0401, 0.0558, 0.2017, 0.0431, 0.0979], device='cuda:3'), in_proj_covar=tensor([0.0293, 0.0261, 0.0375, 0.0333, 0.0271, 0.0309, 0.0315, 0.0280], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 17:12:05,531 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148513.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:12:09,265 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2022-12-08 17:12:10,302 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=148519.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:12:37,249 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.358e+02 2.096e+02 2.575e+02 3.014e+02 5.003e+02, threshold=5.149e+02, percent-clipped=0.0 2022-12-08 17:12:45,249 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148559.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:12:46,812 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=148561.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:12:46,884 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8186, 1.2807, 1.7142, 1.2869, 1.5034, 1.8488, 1.5608, 1.5415], device='cuda:3'), covar=tensor([0.0670, 0.1015, 0.0768, 0.0780, 0.1607, 0.1071, 0.0883, 0.1516], device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0168, 0.0140, 0.0125, 0.0146, 0.0157, 0.0140, 0.0142], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0007, 0.0006, 0.0006], device='cuda:3') 2022-12-08 17:12:52,301 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148567.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:13:02,973 INFO [train.py:873] (3/4) Epoch 20, batch 4900, loss[loss=0.08108, simple_loss=0.1287, pruned_loss=0.01675, over 14355.00 frames. ], tot_loss[loss=0.1012, simple_loss=0.1387, pruned_loss=0.03188, over 1930559.02 frames. ], batch size: 28, lr: 3.90e-03, grad_scale: 4.0 2022-12-08 17:13:33,593 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=148615.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:13:57,698 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148643.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:14:04,643 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.369e+02 1.967e+02 2.434e+02 3.108e+02 5.007e+02, threshold=4.868e+02, percent-clipped=0.0 2022-12-08 17:14:04,885 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148651.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:14:29,319 INFO [train.py:873] (3/4) Epoch 20, batch 5000, loss[loss=0.08923, simple_loss=0.1326, pruned_loss=0.02294, over 14271.00 frames. ], tot_loss[loss=0.1017, simple_loss=0.139, pruned_loss=0.03216, over 1912886.48 frames. ], batch size: 31, lr: 3.90e-03, grad_scale: 4.0 2022-12-08 17:14:58,945 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148712.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:15:02,458 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148716.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:15:13,074 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1863, 1.9996, 2.1528, 2.3318, 1.9481, 2.0323, 2.1017, 2.1370], device='cuda:3'), covar=tensor([0.0431, 0.0721, 0.0427, 0.0431, 0.0676, 0.0969, 0.0536, 0.0479], device='cuda:3'), in_proj_covar=tensor([0.0295, 0.0263, 0.0378, 0.0334, 0.0274, 0.0311, 0.0317, 0.0281], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 17:15:33,632 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.005e+02 2.005e+02 2.534e+02 3.099e+02 5.530e+02, threshold=5.067e+02, percent-clipped=2.0 2022-12-08 17:15:41,494 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1201, 2.3477, 2.3683, 2.4433, 2.0476, 2.4138, 2.2941, 1.5127], device='cuda:3'), covar=tensor([0.0634, 0.0875, 0.0698, 0.0544, 0.1013, 0.0652, 0.0977, 0.1579], device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0094, 0.0073, 0.0078, 0.0102, 0.0093, 0.0104, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0006, 0.0006, 0.0007, 0.0006, 0.0007, 0.0006], device='cuda:3') 2022-12-08 17:15:50,360 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1593, 1.9360, 2.1700, 2.3427, 1.9281, 1.9867, 2.1210, 2.0692], device='cuda:3'), covar=tensor([0.0427, 0.0922, 0.0408, 0.0409, 0.0755, 0.1127, 0.0591, 0.0553], device='cuda:3'), in_proj_covar=tensor([0.0295, 0.0263, 0.0379, 0.0335, 0.0274, 0.0311, 0.0318, 0.0282], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 17:15:56,403 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148777.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:15:58,226 INFO [train.py:873] (3/4) Epoch 20, batch 5100, loss[loss=0.1399, simple_loss=0.1364, pruned_loss=0.07168, over 1327.00 frames. ], tot_loss[loss=0.09981, simple_loss=0.1377, pruned_loss=0.03095, over 1938138.45 frames. ], batch size: 100, lr: 3.90e-03, grad_scale: 4.0 2022-12-08 17:16:03,516 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148785.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:16:05,388 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.1078, 3.9107, 3.5476, 2.8701, 3.5256, 3.7614, 4.3549, 3.4784], device='cuda:3'), covar=tensor([0.0499, 0.1156, 0.0867, 0.1161, 0.0825, 0.0744, 0.0483, 0.0865], device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0168, 0.0141, 0.0126, 0.0146, 0.0158, 0.0140, 0.0142], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0007, 0.0006, 0.0006], device='cuda:3') 2022-12-08 17:17:01,004 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.416e+02 2.285e+02 2.852e+02 3.620e+02 6.279e+02, threshold=5.703e+02, percent-clipped=4.0 2022-12-08 17:17:01,174 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2384, 3.3643, 3.4979, 3.3804, 3.4348, 3.1083, 1.5739, 3.2560], device='cuda:3'), covar=tensor([0.0444, 0.0355, 0.0353, 0.0376, 0.0336, 0.0701, 0.2934, 0.0311], device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0178, 0.0149, 0.0152, 0.0210, 0.0145, 0.0158, 0.0198], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 17:17:03,865 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148854.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:17:25,638 INFO [train.py:873] (3/4) Epoch 20, batch 5200, loss[loss=0.1014, simple_loss=0.1354, pruned_loss=0.03366, over 4979.00 frames. ], tot_loss[loss=0.1007, simple_loss=0.1382, pruned_loss=0.03159, over 1918263.90 frames. ], batch size: 100, lr: 3.90e-03, grad_scale: 8.0 2022-12-08 17:17:32,901 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7668, 2.2960, 3.5876, 2.6868, 3.5920, 3.4300, 3.3968, 2.9201], device='cuda:3'), covar=tensor([0.0885, 0.3197, 0.1066, 0.1969, 0.0887, 0.1251, 0.1420, 0.1923], device='cuda:3'), in_proj_covar=tensor([0.0344, 0.0301, 0.0384, 0.0293, 0.0360, 0.0316, 0.0352, 0.0290], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 17:17:49,790 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.7830, 2.8572, 4.6327, 3.6912, 4.6444, 4.3684, 4.2918, 4.0471], device='cuda:3'), covar=tensor([0.0573, 0.2981, 0.0711, 0.1500, 0.0690, 0.0948, 0.1408, 0.1359], device='cuda:3'), in_proj_covar=tensor([0.0344, 0.0301, 0.0383, 0.0292, 0.0359, 0.0315, 0.0352, 0.0289], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 17:18:20,689 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.69 vs. limit=2.0 2022-12-08 17:18:21,910 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148943.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:18:29,832 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.154e+02 1.963e+02 2.482e+02 3.012e+02 7.254e+02, threshold=4.965e+02, percent-clipped=1.0 2022-12-08 17:18:53,064 INFO [train.py:873] (3/4) Epoch 20, batch 5300, loss[loss=0.07776, simple_loss=0.117, pruned_loss=0.01925, over 13912.00 frames. ], tot_loss[loss=0.1011, simple_loss=0.1385, pruned_loss=0.03182, over 1895732.43 frames. ], batch size: 20, lr: 3.90e-03, grad_scale: 4.0 2022-12-08 17:19:03,323 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=148991.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:19:13,677 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.6409, 3.0975, 2.9644, 3.0661, 2.2945, 3.0402, 2.9367, 1.6641], device='cuda:3'), covar=tensor([0.0837, 0.0620, 0.0676, 0.0601, 0.0914, 0.0554, 0.0693, 0.1695], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0093, 0.0072, 0.0077, 0.0101, 0.0092, 0.0103, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0006, 0.0006, 0.0007, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 17:19:17,281 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=149007.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:19:57,355 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.142e+02 2.154e+02 2.538e+02 2.985e+02 9.704e+02, threshold=5.077e+02, percent-clipped=4.0 2022-12-08 17:20:14,504 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=149072.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:20:20,461 INFO [train.py:873] (3/4) Epoch 20, batch 5400, loss[loss=0.0964, simple_loss=0.1393, pruned_loss=0.02674, over 14264.00 frames. ], tot_loss[loss=0.1006, simple_loss=0.1383, pruned_loss=0.03141, over 1924341.47 frames. ], batch size: 57, lr: 3.90e-03, grad_scale: 2.0 2022-12-08 17:20:25,683 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=149085.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:20:53,440 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2022-12-08 17:21:07,674 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=149133.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:21:24,433 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.5614, 1.5129, 2.7622, 1.5335, 2.7600, 2.7154, 2.0130, 2.8895], device='cuda:3'), covar=tensor([0.0300, 0.2775, 0.0404, 0.1942, 0.0399, 0.0547, 0.1123, 0.0275], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0157, 0.0162, 0.0170, 0.0167, 0.0181, 0.0134, 0.0154], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 17:21:25,134 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.118e+02 1.972e+02 2.518e+02 2.987e+02 6.531e+02, threshold=5.036e+02, percent-clipped=3.0 2022-12-08 17:21:26,189 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=149154.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:21:31,259 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.1417, 3.5346, 3.1152, 3.3535, 2.5684, 3.4052, 3.2476, 1.9741], device='cuda:3'), covar=tensor([0.1384, 0.0579, 0.1357, 0.0564, 0.0921, 0.0531, 0.1007, 0.1566], device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0094, 0.0073, 0.0078, 0.0102, 0.0093, 0.0104, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0006, 0.0006, 0.0007, 0.0006, 0.0007, 0.0006], device='cuda:3') 2022-12-08 17:21:48,145 INFO [train.py:873] (3/4) Epoch 20, batch 5500, loss[loss=0.1007, simple_loss=0.1286, pruned_loss=0.03642, over 5973.00 frames. ], tot_loss[loss=0.1006, simple_loss=0.1382, pruned_loss=0.03146, over 1927758.21 frames. ], batch size: 100, lr: 3.90e-03, grad_scale: 2.0 2022-12-08 17:22:07,130 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.1392, 2.0069, 1.8373, 1.8809, 2.0601, 2.1088, 2.0781, 2.0879], device='cuda:3'), covar=tensor([0.1205, 0.1170, 0.2643, 0.2679, 0.1487, 0.1323, 0.1632, 0.1139], device='cuda:3'), in_proj_covar=tensor([0.0399, 0.0285, 0.0452, 0.0572, 0.0360, 0.0462, 0.0399, 0.0399], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 17:22:07,923 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=149202.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:22:52,116 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.123e+02 2.034e+02 2.619e+02 3.145e+02 6.128e+02, threshold=5.238e+02, percent-clipped=4.0 2022-12-08 17:23:14,999 INFO [train.py:873] (3/4) Epoch 20, batch 5600, loss[loss=0.08567, simple_loss=0.1325, pruned_loss=0.0194, over 14214.00 frames. ], tot_loss[loss=0.1016, simple_loss=0.1387, pruned_loss=0.03224, over 1944111.87 frames. ], batch size: 32, lr: 3.89e-03, grad_scale: 4.0 2022-12-08 17:23:39,926 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=149307.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:24:07,580 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3258, 4.1349, 3.7774, 3.9653, 4.2016, 4.2745, 4.3098, 4.3288], device='cuda:3'), covar=tensor([0.0974, 0.0545, 0.2099, 0.3157, 0.0805, 0.0878, 0.1073, 0.0830], device='cuda:3'), in_proj_covar=tensor([0.0398, 0.0284, 0.0449, 0.0572, 0.0360, 0.0460, 0.0397, 0.0398], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 17:24:19,225 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.71 vs. limit=2.0 2022-12-08 17:24:19,494 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.122e+02 1.989e+02 2.324e+02 2.779e+02 6.746e+02, threshold=4.648e+02, percent-clipped=1.0 2022-12-08 17:24:19,669 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9112, 1.7114, 3.4820, 3.1263, 3.3033, 3.5011, 2.8935, 3.4784], device='cuda:3'), covar=tensor([0.1676, 0.1676, 0.0168, 0.0365, 0.0345, 0.0197, 0.0336, 0.0179], device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0155, 0.0130, 0.0168, 0.0147, 0.0141, 0.0124, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 17:24:21,362 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=149355.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:24:36,242 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=149372.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:24:42,618 INFO [train.py:873] (3/4) Epoch 20, batch 5700, loss[loss=0.1208, simple_loss=0.1342, pruned_loss=0.05371, over 2617.00 frames. ], tot_loss[loss=0.1018, simple_loss=0.1387, pruned_loss=0.0324, over 1893186.55 frames. ], batch size: 100, lr: 3.89e-03, grad_scale: 4.0 2022-12-08 17:25:18,494 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=149420.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:25:47,186 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 7.719e+01 2.141e+02 2.467e+02 2.944e+02 6.062e+02, threshold=4.935e+02, percent-clipped=6.0 2022-12-08 17:25:47,779 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 2022-12-08 17:25:52,874 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.8002, 1.8054, 1.9161, 1.8287, 1.7859, 1.7831, 1.5078, 1.2225], device='cuda:3'), covar=tensor([0.0268, 0.0358, 0.0235, 0.0235, 0.0289, 0.0336, 0.0315, 0.0431], device='cuda:3'), in_proj_covar=tensor([0.0025, 0.0024, 0.0022, 0.0024, 0.0023, 0.0036, 0.0030, 0.0035], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:3') 2022-12-08 17:26:10,321 INFO [train.py:873] (3/4) Epoch 20, batch 5800, loss[loss=0.09998, simple_loss=0.1421, pruned_loss=0.02891, over 14243.00 frames. ], tot_loss[loss=0.1011, simple_loss=0.1386, pruned_loss=0.03184, over 1925694.53 frames. ], batch size: 46, lr: 3.89e-03, grad_scale: 4.0 2022-12-08 17:26:33,953 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7530, 1.3047, 2.5437, 2.2282, 2.3449, 2.5423, 1.7803, 2.5486], device='cuda:3'), covar=tensor([0.1147, 0.1560, 0.0253, 0.0613, 0.0659, 0.0311, 0.0750, 0.0284], device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0155, 0.0130, 0.0168, 0.0148, 0.0140, 0.0123, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2022-12-08 17:26:40,997 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=149514.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:26:45,521 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.80 vs. limit=5.0 2022-12-08 17:26:54,126 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.6272, 2.7393, 4.4755, 3.4344, 4.4626, 4.2743, 4.2543, 3.8161], device='cuda:3'), covar=tensor([0.0637, 0.3008, 0.0731, 0.1635, 0.0642, 0.0938, 0.1359, 0.1517], device='cuda:3'), in_proj_covar=tensor([0.0348, 0.0305, 0.0386, 0.0296, 0.0361, 0.0318, 0.0357, 0.0292], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 17:26:57,655 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.12 vs. limit=2.0 2022-12-08 17:27:06,481 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.4108, 2.4816, 2.4141, 2.2837, 2.1746, 2.0083, 1.8674, 2.0284], device='cuda:3'), covar=tensor([0.0338, 0.0509, 0.0416, 0.0324, 0.0372, 0.0489, 0.0411, 0.0449], device='cuda:3'), in_proj_covar=tensor([0.0025, 0.0024, 0.0022, 0.0024, 0.0023, 0.0036, 0.0030, 0.0034], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:3') 2022-12-08 17:27:08,933 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.2770, 3.4286, 3.5972, 3.3978, 3.5111, 3.2793, 1.4889, 3.2714], device='cuda:3'), covar=tensor([0.0442, 0.0336, 0.0326, 0.0414, 0.0323, 0.0595, 0.3012, 0.0301], device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0176, 0.0148, 0.0151, 0.0209, 0.0145, 0.0158, 0.0196], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 17:27:15,062 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.429e+02 2.003e+02 2.561e+02 2.963e+02 6.960e+02, threshold=5.123e+02, percent-clipped=3.0 2022-12-08 17:27:25,781 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=149565.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:27:34,448 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=149575.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:27:38,142 INFO [train.py:873] (3/4) Epoch 20, batch 5900, loss[loss=0.1002, simple_loss=0.1198, pruned_loss=0.04028, over 3892.00 frames. ], tot_loss[loss=0.1006, simple_loss=0.1378, pruned_loss=0.03168, over 1923070.64 frames. ], batch size: 100, lr: 3.89e-03, grad_scale: 4.0 2022-12-08 17:27:50,410 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.24 vs. limit=5.0 2022-12-08 17:28:18,217 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.55 vs. limit=2.0 2022-12-08 17:28:20,630 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=149626.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 17:28:24,792 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.3858, 2.4750, 2.5385, 2.5270, 2.5163, 2.1345, 1.5556, 2.2592], device='cuda:3'), covar=tensor([0.0646, 0.0510, 0.0471, 0.0391, 0.0381, 0.1352, 0.2427, 0.0452], device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0177, 0.0150, 0.0152, 0.0211, 0.0146, 0.0159, 0.0197], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 17:28:43,779 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.268e+02 2.112e+02 2.459e+02 3.260e+02 5.779e+02, threshold=4.919e+02, percent-clipped=5.0 2022-12-08 17:28:50,229 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2022-12-08 17:29:07,359 INFO [train.py:873] (3/4) Epoch 20, batch 6000, loss[loss=0.09072, simple_loss=0.1323, pruned_loss=0.02456, over 14274.00 frames. ], tot_loss[loss=0.09917, simple_loss=0.1368, pruned_loss=0.03077, over 1945841.30 frames. ], batch size: 76, lr: 3.89e-03, grad_scale: 8.0 2022-12-08 17:29:07,359 INFO [train.py:896] (3/4) Computing validation loss 2022-12-08 17:29:22,043 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8979, 1.8106, 3.0619, 2.0298, 3.0384, 2.9469, 1.8986, 3.1237], device='cuda:3'), covar=tensor([0.0184, 0.2444, 0.0310, 0.1380, 0.0316, 0.0474, 0.0772, 0.0316], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0157, 0.0161, 0.0169, 0.0167, 0.0180, 0.0133, 0.0154], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 17:29:22,627 INFO [train.py:905] (3/4) Epoch 20, validation: loss=0.1445, simple_loss=0.1806, pruned_loss=0.0542, over 857387.00 frames. 2022-12-08 17:29:22,627 INFO [train.py:906] (3/4) Maximum memory allocated so far is 17954MB 2022-12-08 17:30:22,021 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=149746.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 17:30:27,889 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.146e+02 2.065e+02 2.623e+02 3.098e+02 9.406e+02, threshold=5.245e+02, percent-clipped=5.0 2022-12-08 17:30:50,977 INFO [train.py:873] (3/4) Epoch 20, batch 6100, loss[loss=0.08651, simple_loss=0.1104, pruned_loss=0.03132, over 3816.00 frames. ], tot_loss[loss=0.1007, simple_loss=0.1379, pruned_loss=0.03177, over 1942208.58 frames. ], batch size: 100, lr: 3.89e-03, grad_scale: 8.0 2022-12-08 17:31:09,796 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9289, 1.3719, 2.0274, 1.3021, 1.9509, 2.0505, 1.7124, 2.1350], device='cuda:3'), covar=tensor([0.0382, 0.2277, 0.0587, 0.2101, 0.0714, 0.0706, 0.1298, 0.0475], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0157, 0.0162, 0.0169, 0.0167, 0.0181, 0.0134, 0.0154], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 17:31:16,110 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=149807.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 17:31:28,620 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.8518, 2.9835, 3.1328, 2.9895, 3.0981, 2.9151, 1.6082, 2.8721], device='cuda:3'), covar=tensor([0.0495, 0.0480, 0.0366, 0.0429, 0.0326, 0.0821, 0.2677, 0.0344], device='cuda:3'), in_proj_covar=tensor([0.0179, 0.0178, 0.0151, 0.0153, 0.0211, 0.0146, 0.0159, 0.0197], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 17:31:56,635 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.295e+02 2.144e+02 2.583e+02 3.219e+02 5.785e+02, threshold=5.165e+02, percent-clipped=1.0 2022-12-08 17:32:11,625 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=149870.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:32:11,655 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.5450, 4.6424, 5.0444, 4.2390, 4.8636, 5.1277, 2.0087, 4.4853], device='cuda:3'), covar=tensor([0.0347, 0.0326, 0.0356, 0.0412, 0.0273, 0.0148, 0.3145, 0.0289], device='cuda:3'), in_proj_covar=tensor([0.0179, 0.0178, 0.0151, 0.0153, 0.0211, 0.0147, 0.0159, 0.0198], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 17:32:19,291 INFO [train.py:873] (3/4) Epoch 20, batch 6200, loss[loss=0.1031, simple_loss=0.1424, pruned_loss=0.03189, over 13537.00 frames. ], tot_loss[loss=0.1004, simple_loss=0.1383, pruned_loss=0.03128, over 2036388.44 frames. ], batch size: 100, lr: 3.89e-03, grad_scale: 8.0 2022-12-08 17:32:57,040 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=149921.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 17:33:24,558 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.165e+02 2.069e+02 2.540e+02 3.004e+02 8.039e+02, threshold=5.080e+02, percent-clipped=2.0 2022-12-08 17:33:31,438 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.63 vs. limit=2.0 2022-12-08 17:33:47,752 INFO [train.py:873] (3/4) Epoch 20, batch 6300, loss[loss=0.1539, simple_loss=0.1423, pruned_loss=0.08278, over 1275.00 frames. ], tot_loss[loss=0.09955, simple_loss=0.1376, pruned_loss=0.03073, over 1993875.53 frames. ], batch size: 100, lr: 3.88e-03, grad_scale: 8.0 2022-12-08 17:33:51,357 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.4411, 4.0774, 3.5131, 3.6854, 2.6361, 3.9184, 3.6319, 2.0767], device='cuda:3'), covar=tensor([0.1425, 0.0491, 0.1428, 0.0580, 0.0916, 0.0410, 0.0887, 0.1743], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0092, 0.0071, 0.0077, 0.0099, 0.0091, 0.0101, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0008, 0.0006, 0.0005, 0.0006, 0.0006, 0.0006, 0.0006, 0.0006], device='cuda:3') 2022-12-08 17:33:53,417 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.57 vs. limit=5.0 2022-12-08 17:34:31,039 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.0605, 1.4179, 3.2584, 1.7130, 3.1005, 3.2323, 2.2855, 3.4161], device='cuda:3'), covar=tensor([0.0305, 0.3183, 0.0446, 0.2162, 0.0916, 0.0463, 0.1098, 0.0255], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0158, 0.0162, 0.0170, 0.0167, 0.0181, 0.0134, 0.0154], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 17:34:56,193 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.169e+02 2.076e+02 2.693e+02 3.406e+02 8.300e+02, threshold=5.385e+02, percent-clipped=3.0 2022-12-08 17:35:08,502 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2022-12-08 17:35:18,743 INFO [train.py:873] (3/4) Epoch 20, batch 6400, loss[loss=0.07628, simple_loss=0.1212, pruned_loss=0.01567, over 14324.00 frames. ], tot_loss[loss=0.0986, simple_loss=0.137, pruned_loss=0.03009, over 1998578.83 frames. ], batch size: 28, lr: 3.88e-03, grad_scale: 8.0 2022-12-08 17:35:18,914 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=150079.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:35:39,396 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=150102.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 17:35:55,429 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.3417, 4.0652, 3.7492, 4.0105, 4.1764, 4.2246, 4.3143, 4.2744], device='cuda:3'), covar=tensor([0.0739, 0.0534, 0.2212, 0.2326, 0.0803, 0.0862, 0.0858, 0.0832], device='cuda:3'), in_proj_covar=tensor([0.0399, 0.0285, 0.0453, 0.0576, 0.0363, 0.0465, 0.0399, 0.0403], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 17:36:05,980 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7483, 1.2497, 1.7211, 1.3008, 1.4983, 1.8110, 1.5626, 1.5520], device='cuda:3'), covar=tensor([0.0945, 0.1164, 0.0793, 0.0949, 0.1645, 0.0867, 0.0856, 0.1791], device='cuda:3'), in_proj_covar=tensor([0.0154, 0.0165, 0.0138, 0.0123, 0.0143, 0.0154, 0.0138, 0.0141], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0007, 0.0006, 0.0006], device='cuda:3') 2022-12-08 17:36:06,886 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.7323, 1.7259, 1.7467, 1.8227, 1.8036, 1.2993, 1.4730, 1.7792], device='cuda:3'), covar=tensor([0.0755, 0.0705, 0.0548, 0.0619, 0.0606, 0.0842, 0.0776, 0.0468], device='cuda:3'), in_proj_covar=tensor([0.0040, 0.0038, 0.0043, 0.0035, 0.0038, 0.0052, 0.0040, 0.0042], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:3') 2022-12-08 17:36:12,265 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=150140.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:36:23,032 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.274e+02 2.042e+02 2.543e+02 3.172e+02 7.027e+02, threshold=5.086e+02, percent-clipped=3.0 2022-12-08 17:36:38,293 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=150170.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:36:46,072 INFO [train.py:873] (3/4) Epoch 20, batch 6500, loss[loss=0.08377, simple_loss=0.1229, pruned_loss=0.0223, over 13997.00 frames. ], tot_loss[loss=0.1001, simple_loss=0.1381, pruned_loss=0.03103, over 1990807.41 frames. ], batch size: 19, lr: 3.88e-03, grad_scale: 8.0 2022-12-08 17:37:12,638 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2022-12-08 17:37:18,351 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.6769, 1.4122, 3.6657, 1.7613, 3.5441, 3.7874, 2.7677, 4.0407], device='cuda:3'), covar=tensor([0.0232, 0.3312, 0.0407, 0.2235, 0.0686, 0.0418, 0.0882, 0.0184], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0158, 0.0163, 0.0170, 0.0168, 0.0182, 0.0134, 0.0155], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2022-12-08 17:37:20,272 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=150218.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:37:23,018 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=150221.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:37:50,499 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.402e+02 2.155e+02 2.479e+02 3.022e+02 7.876e+02, threshold=4.957e+02, percent-clipped=4.0 2022-12-08 17:37:51,138 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.74 vs. limit=5.0 2022-12-08 17:38:04,672 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=150269.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:38:08,974 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=150274.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:38:10,637 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.7344, 5.5720, 5.1902, 5.6968, 5.2490, 5.1900, 5.8152, 5.5028], device='cuda:3'), covar=tensor([0.0497, 0.0699, 0.0653, 0.0459, 0.0774, 0.0328, 0.0430, 0.0548], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0147, 0.0149, 0.0166, 0.0150, 0.0127, 0.0172, 0.0152], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 17:38:13,123 INFO [train.py:873] (3/4) Epoch 20, batch 6600, loss[loss=0.1006, simple_loss=0.1406, pruned_loss=0.03031, over 14186.00 frames. ], tot_loss[loss=0.09934, simple_loss=0.1377, pruned_loss=0.03048, over 2043481.97 frames. ], batch size: 84, lr: 3.88e-03, grad_scale: 8.0 2022-12-08 17:38:34,101 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 2022-12-08 17:38:56,230 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2022-12-08 17:39:02,479 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=150335.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:39:18,406 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.127e+02 2.141e+02 2.452e+02 3.103e+02 5.286e+02, threshold=4.904e+02, percent-clipped=1.0 2022-12-08 17:39:22,182 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.59 vs. limit=2.0 2022-12-08 17:39:22,824 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([4.4451, 2.2403, 4.4328, 3.0334, 4.2148, 2.1477, 3.3094, 4.3073], device='cuda:3'), covar=tensor([0.0543, 0.3613, 0.0401, 0.4974, 0.0660, 0.3053, 0.1350, 0.0367], device='cuda:3'), in_proj_covar=tensor([0.0249, 0.0194, 0.0223, 0.0262, 0.0238, 0.0197, 0.0200, 0.0218], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0005], device='cuda:3') 2022-12-08 17:39:35,471 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.56 vs. limit=5.0 2022-12-08 17:39:41,070 INFO [train.py:873] (3/4) Epoch 20, batch 6700, loss[loss=0.08496, simple_loss=0.127, pruned_loss=0.02147, over 14501.00 frames. ], tot_loss[loss=0.09911, simple_loss=0.1373, pruned_loss=0.03047, over 1985480.21 frames. ], batch size: 21, lr: 3.88e-03, grad_scale: 8.0 2022-12-08 17:39:45,875 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 2022-12-08 17:40:01,130 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=150402.0, num_to_drop=1, layers_to_drop={0} 2022-12-08 17:40:29,792 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=150435.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:40:42,820 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=150450.0, num_to_drop=1, layers_to_drop={1} 2022-12-08 17:40:45,137 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.516e+02 2.053e+02 2.408e+02 2.928e+02 1.041e+03, threshold=4.817e+02, percent-clipped=2.0 2022-12-08 17:41:07,585 INFO [train.py:873] (3/4) Epoch 20, batch 6800, loss[loss=0.105, simple_loss=0.1415, pruned_loss=0.03423, over 14284.00 frames. ], tot_loss[loss=0.1005, simple_loss=0.1382, pruned_loss=0.03137, over 1996361.84 frames. ], batch size: 66, lr: 3.88e-03, grad_scale: 8.0 2022-12-08 17:41:57,986 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=150537.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:42:12,046 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.252e+02 2.195e+02 2.476e+02 3.043e+02 9.162e+02, threshold=4.952e+02, percent-clipped=3.0 2022-12-08 17:42:34,234 INFO [train.py:873] (3/4) Epoch 20, batch 6900, loss[loss=0.09423, simple_loss=0.1375, pruned_loss=0.02548, over 14281.00 frames. ], tot_loss[loss=0.1007, simple_loss=0.1382, pruned_loss=0.03158, over 1984177.47 frames. ], batch size: 60, lr: 3.88e-03, grad_scale: 8.0 2022-12-08 17:42:51,096 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=150598.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:43:19,320 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=150630.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:43:28,366 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0 2022-12-08 17:43:40,041 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.249e+02 1.852e+02 2.409e+02 3.250e+02 6.967e+02, threshold=4.817e+02, percent-clipped=2.0 2022-12-08 17:43:59,213 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2022-12-08 17:44:02,001 INFO [train.py:873] (3/4) Epoch 20, batch 7000, loss[loss=0.09625, simple_loss=0.1408, pruned_loss=0.02584, over 14207.00 frames. ], tot_loss[loss=0.1004, simple_loss=0.1381, pruned_loss=0.03137, over 2019700.55 frames. ], batch size: 25, lr: 3.88e-03, grad_scale: 4.0 2022-12-08 17:44:48,591 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=150732.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:44:50,887 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=150735.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:45:06,716 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.355e+02 2.298e+02 2.589e+02 3.066e+02 6.150e+02, threshold=5.179e+02, percent-clipped=4.0 2022-12-08 17:45:28,339 INFO [train.py:873] (3/4) Epoch 20, batch 7100, loss[loss=0.1313, simple_loss=0.1304, pruned_loss=0.06616, over 1276.00 frames. ], tot_loss[loss=0.1018, simple_loss=0.139, pruned_loss=0.0323, over 1967506.20 frames. ], batch size: 100, lr: 3.87e-03, grad_scale: 4.0 2022-12-08 17:45:31,473 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=150783.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:45:38,930 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.41 vs. limit=2.0 2022-12-08 17:45:40,433 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.2914, 1.2361, 1.3554, 1.2006, 1.1141, 1.0569, 1.1328, 1.0211], device='cuda:3'), covar=tensor([0.0253, 0.0282, 0.0207, 0.0280, 0.0269, 0.0472, 0.0309, 0.0421], device='cuda:3'), in_proj_covar=tensor([0.0025, 0.0025, 0.0022, 0.0024, 0.0023, 0.0037, 0.0030, 0.0035], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:3') 2022-12-08 17:45:40,438 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=150793.0, num_to_drop=1, layers_to_drop={3} 2022-12-08 17:46:14,858 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([1.9654, 2.0166, 2.0674, 2.0518, 2.0015, 1.6254, 1.3686, 1.8363], device='cuda:3'), covar=tensor([0.0849, 0.0648, 0.0541, 0.0467, 0.0564, 0.1658, 0.2312, 0.0568], device='cuda:3'), in_proj_covar=tensor([0.0179, 0.0178, 0.0150, 0.0153, 0.0212, 0.0146, 0.0159, 0.0198], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004], device='cuda:3') 2022-12-08 17:46:33,854 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.300e+02 2.271e+02 2.646e+02 3.296e+02 2.153e+03, threshold=5.293e+02, percent-clipped=6.0 2022-12-08 17:46:52,336 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([5.2100, 5.0183, 4.7427, 5.0895, 4.7876, 4.5793, 5.2479, 4.8686], device='cuda:3'), covar=tensor([0.0554, 0.0863, 0.0793, 0.0694, 0.0801, 0.0547, 0.0554, 0.0912], device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0147, 0.0148, 0.0165, 0.0149, 0.0126, 0.0171, 0.0152], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 17:46:55,670 INFO [train.py:873] (3/4) Epoch 20, batch 7200, loss[loss=0.1642, simple_loss=0.1538, pruned_loss=0.08735, over 1239.00 frames. ], tot_loss[loss=0.1013, simple_loss=0.1388, pruned_loss=0.03191, over 1975886.49 frames. ], batch size: 100, lr: 3.87e-03, grad_scale: 8.0 2022-12-08 17:47:07,850 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=150893.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:47:08,979 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.9472, 2.4042, 3.8588, 4.0441, 3.7201, 2.3485, 3.9288, 2.8899], device='cuda:3'), covar=tensor([0.0607, 0.1457, 0.1119, 0.0588, 0.0711, 0.2374, 0.0626, 0.1255], device='cuda:3'), in_proj_covar=tensor([0.0295, 0.0262, 0.0378, 0.0335, 0.0273, 0.0309, 0.0315, 0.0279], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:3') 2022-12-08 17:47:14,183 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=150900.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:47:40,832 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=150930.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:48:01,352 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.240e+02 1.977e+02 2.439e+02 3.091e+02 5.350e+02, threshold=4.878e+02, percent-clipped=1.0 2022-12-08 17:48:07,593 INFO [zipformer.py:626] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=150961.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:48:22,754 INFO [zipformer.py:626] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=150978.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:48:23,466 INFO [train.py:873] (3/4) Epoch 20, batch 7300, loss[loss=0.1085, simple_loss=0.126, pruned_loss=0.0455, over 2658.00 frames. ], tot_loss[loss=0.1014, simple_loss=0.1386, pruned_loss=0.03209, over 1946307.88 frames. ], batch size: 100, lr: 3.87e-03, grad_scale: 8.0 2022-12-08 17:49:29,351 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.047e+02 2.139e+02 2.523e+02 3.177e+02 9.954e+02, threshold=5.046e+02, percent-clipped=7.0 2022-12-08 17:49:31,552 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2022-12-08 17:49:51,032 INFO [train.py:873] (3/4) Epoch 20, batch 7400, loss[loss=0.1437, simple_loss=0.1375, pruned_loss=0.07491, over 1295.00 frames. ], tot_loss[loss=0.1015, simple_loss=0.1388, pruned_loss=0.03209, over 1966019.87 frames. ], batch size: 100, lr: 3.87e-03, grad_scale: 8.0 2022-12-08 17:49:55,616 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([3.7847, 3.7405, 3.5548, 3.8714, 3.4932, 3.3412, 3.8305, 3.6375], device='cuda:3'), covar=tensor([0.0703, 0.0969, 0.0889, 0.0631, 0.0986, 0.0721, 0.0653, 0.0906], device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0150, 0.0150, 0.0167, 0.0151, 0.0128, 0.0173, 0.0154], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:3') 2022-12-08 17:49:59,305 INFO [zipformer.py:626] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=151088.0, num_to_drop=1, layers_to_drop={2} 2022-12-08 17:50:55,891 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.036e+02 2.135e+02 2.491e+02 3.146e+02 1.107e+03, threshold=4.982e+02, percent-clipped=6.0 2022-12-08 17:50:57,044 INFO [zipformer.py:1414] (3/4) attn_weights_entropy = tensor([2.0661, 1.9503, 2.2887, 2.1209, 2.0368, 1.8337, 1.8881, 1.2778], device='cuda:3'), covar=tensor([0.0185, 0.0328, 0.0160, 0.0179, 0.0221, 0.0310, 0.0221, 0.0482], device='cuda:3'), in_proj_covar=tensor([0.0025, 0.0024, 0.0022, 0.0024, 0.0023, 0.0036, 0.0030, 0.0034], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:3') 2022-12-08 17:51:17,607 INFO [train.py:873] (3/4) Epoch 20, batch 7500, loss[loss=0.08406, simple_loss=0.1266, pruned_loss=0.02076, over 14286.00 frames. ], tot_loss[loss=0.1012, simple_loss=0.1384, pruned_loss=0.03196, over 1918833.55 frames. ], batch size: 31, lr: 3.87e-03, grad_scale: 8.0 2022-12-08 17:51:21,842 INFO [zipformer.py:626] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=151184.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:51:29,971 INFO [zipformer.py:626] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151193.0, num_to_drop=0, layers_to_drop=set() 2022-12-08 17:51:40,168 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2022-12-08 17:52:04,556 INFO [train.py:1091] (3/4) Done!