2022-11-21 08:54:32,239 INFO [decode.py:574] Decoding started
2022-11-21 08:54:32,240 INFO [decode.py:580] Device: cuda:0
2022-11-21 08:54:32,247 INFO [decode.py:590] {'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': '1.21', 'k2-build-type': 'Debug', 'k2-with-cuda': True, 'k2-git-sha1': 'f271e82ef30f75fecbae44b163e1244e53def116', 'k2-git-date': 'Fri Oct 28 05:02:16 2022', 'lhotse-version': '1.9.0.dev+git.97bf4b0.dirty', 'torch-version': '1.10.0+cu111', 'torch-cuda-available': True, 'torch-cuda-version': '11.1', 'python-version': '3.8', 'icefall-git-branch': 'ami_recipe', 'icefall-git-sha1': 'd1b5a16-dirty', 'icefall-git-date': 'Sun Nov 20 22:32:57 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': 'r2n06', 'IP address': '10.1.2.6'}, 'epoch': 14, 'iter': 0, 'avg': 8, 'use_averaged_model': True, 'exp_dir': PosixPath('pruned_transducer_stateless7/exp/v2'), 'lang_dir': PosixPath('data/lang_bpe_500'), 'decoding_method': 'greedy_search', 'beam_size': 4, 'beam': 4, 'ngram_lm_scale': 0.01, 'max_contexts': 4, 'max_states': 8, 'context_size': 2, 'max_sym_per_frame': 1, 'num_paths': 200, 'nbest_scale': 0.5, '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': 500, 'max_cuts': None, 'num_buckets': 50, 'on_the_fly_feats': False, 'shuffle': True, 'num_workers': 8, 'enable_spec_aug': True, 'spec_aug_time_warp_factor': 80, 'ihm_only': False, 'res_dir': PosixPath('pruned_transducer_stateless7/exp/v2/greedy_search'), 'suffix': 'epoch-14-avg-8-context-2-max-sym-per-frame-1', 'blank_id': 0, 'unk_id': 2, 'vocab_size': 500}
2022-11-21 08:54:32,247 INFO [decode.py:592] About to create model
2022-11-21 08:54:32,735 INFO [zipformer.py:179] 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.
2022-11-21 08:54:32,755 INFO [decode.py:659] Calculating the averaged model over epoch range from 6 (excluded) to 14
2022-11-21 08:54:46,206 INFO [decode.py:694] Number of model parameters: 70369391
2022-11-21 08:54:46,206 INFO [asr_datamodule.py:392] About to get AMI IHM dev cuts
2022-11-21 08:54:46,208 INFO [asr_datamodule.py:413] About to get AMI IHM test cuts
2022-11-21 08:54:46,209 INFO [asr_datamodule.py:398] About to get AMI SDM dev cuts
2022-11-21 08:54:46,210 INFO [asr_datamodule.py:419] About to get AMI SDM test cuts
2022-11-21 08:54:46,211 INFO [asr_datamodule.py:407] About to get AMI GSS-enhanced dev cuts
2022-11-21 08:54:46,212 INFO [asr_datamodule.py:428] About to get AMI GSS-enhanced test cuts
2022-11-21 08:54:48,328 INFO [decode.py:726] Decoding dev_ihm
2022-11-21 08:54:50,509 INFO [decode.py:469] batch 0/?, cuts processed until now is 72
2022-11-21 08:55:00,670 INFO [zipformer.py:1414] attn_weights_entropy = tensor([3.4141, 5.3941, 4.0147, 5.1413, 4.2180, 4.1089, 3.8119, 4.7968],
       device='cuda:0'), covar=tensor([0.0981, 0.0178, 0.0671, 0.0197, 0.0484, 0.0609, 0.1299, 0.0177],
       device='cuda:0'), in_proj_covar=tensor([0.0148, 0.0117, 0.0145, 0.0122, 0.0157, 0.0156, 0.0153, 0.0133],
       device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003],
       device='cuda:0')
2022-11-21 08:55:37,030 INFO [decode.py:485] The transcripts are stored in pruned_transducer_stateless7/exp/v2/greedy_search/recogs-dev_ihm-greedy_search-epoch-14-avg-8-context-2-max-sym-per-frame-1.txt
2022-11-21 08:55:37,218 INFO [utils.py:530] [dev_ihm-greedy_search] %WER 19.25% [18280 / 94940, 2799 ins, 3599 del, 11882 sub ]
2022-11-21 08:55:37,918 INFO [utils.py:530] [dev_ihm-greedy_search] %WER 12.01% [44413 / 369873, 10958 ins, 16172 del, 17283 sub ]
2022-11-21 08:55:38,949 INFO [decode.py:511] Wrote detailed error stats to pruned_transducer_stateless7/exp/v2/greedy_search/wers-dev_ihm-greedy_search-epoch-14-avg-8-context-2-max-sym-per-frame-1.txt
2022-11-21 08:55:38,950 INFO [decode.py:531] 
For dev_ihm, WER/CER of different settings are:
greedy_search	19.25	12.01	best for dev_ihm

2022-11-21 08:55:38,954 INFO [decode.py:726] Decoding test_ihm
2022-11-21 08:55:40,998 INFO [decode.py:469] batch 0/?, cuts processed until now is 69
2022-11-21 08:55:47,212 INFO [zipformer.py:1414] attn_weights_entropy = tensor([3.9011, 3.9179, 3.7190, 3.6107, 3.8823, 3.5968, 1.7153, 4.0888],
       device='cuda:0'), covar=tensor([0.0182, 0.0175, 0.0218, 0.0272, 0.0220, 0.0263, 0.2729, 0.0168],
       device='cuda:0'), in_proj_covar=tensor([0.0093, 0.0074, 0.0074, 0.0066, 0.0090, 0.0076, 0.0123, 0.0097],
       device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002],
       device='cuda:0')
2022-11-21 08:55:58,872 INFO [zipformer.py:1414] attn_weights_entropy = tensor([2.3999, 2.3089, 2.9419, 1.7908, 1.0976, 3.2808, 2.6477, 2.3227],
       device='cuda:0'), covar=tensor([0.0923, 0.1227, 0.0463, 0.3079, 0.4287, 0.1788, 0.1915, 0.1478],
       device='cuda:0'), in_proj_covar=tensor([0.0076, 0.0065, 0.0066, 0.0079, 0.0058, 0.0046, 0.0055, 0.0066],
       device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002],
       device='cuda:0')
2022-11-21 08:56:28,680 INFO [decode.py:485] The transcripts are stored in pruned_transducer_stateless7/exp/v2/greedy_search/recogs-test_ihm-greedy_search-epoch-14-avg-8-context-2-max-sym-per-frame-1.txt
2022-11-21 08:56:28,820 INFO [utils.py:530] [test_ihm-greedy_search] %WER 17.83% [15986 / 89659, 1991 ins, 3568 del, 10427 sub ]
2022-11-21 08:56:29,570 INFO [utils.py:530] [test_ihm-greedy_search] %WER 10.95% [38776 / 354205, 8770 ins, 15207 del, 14799 sub ]
2022-11-21 08:56:30,771 INFO [decode.py:511] Wrote detailed error stats to pruned_transducer_stateless7/exp/v2/greedy_search/wers-test_ihm-greedy_search-epoch-14-avg-8-context-2-max-sym-per-frame-1.txt
2022-11-21 08:56:30,773 INFO [decode.py:531] 
For test_ihm, WER/CER of different settings are:
greedy_search	17.83	10.95	best for test_ihm

2022-11-21 08:56:30,784 INFO [decode.py:726] Decoding dev_sdm
2022-11-21 08:56:32,521 INFO [decode.py:469] batch 0/?, cuts processed until now is 71
2022-11-21 08:56:39,302 INFO [zipformer.py:1414] attn_weights_entropy = tensor([2.2504, 1.6006, 1.7922, 1.4734, 1.7332, 2.0485, 1.6161, 1.4893],
       device='cuda:0'), covar=tensor([0.0036, 0.0058, 0.0063, 0.0051, 0.0107, 0.0068, 0.0035, 0.0053],
       device='cuda:0'), in_proj_covar=tensor([0.0018, 0.0018, 0.0018, 0.0025, 0.0021, 0.0019, 0.0024, 0.0024],
       device='cuda:0'), out_proj_covar=tensor([1.6600e-05, 1.6809e-05, 1.6082e-05, 2.4018e-05, 1.9485e-05, 1.8494e-05,
        2.3316e-05, 2.3270e-05], device='cuda:0')
2022-11-21 08:57:04,693 INFO [zipformer.py:1414] attn_weights_entropy = tensor([3.4126, 3.5695, 3.3951, 3.2795, 3.4601, 3.1680, 1.4567, 3.5835],
       device='cuda:0'), covar=tensor([0.0201, 0.0128, 0.0202, 0.0191, 0.0253, 0.0244, 0.2779, 0.0204],
       device='cuda:0'), in_proj_covar=tensor([0.0093, 0.0074, 0.0074, 0.0066, 0.0090, 0.0076, 0.0123, 0.0097],
       device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002],
       device='cuda:0')
2022-11-21 08:57:18,784 INFO [decode.py:485] The transcripts are stored in pruned_transducer_stateless7/exp/v2/greedy_search/recogs-dev_sdm-greedy_search-epoch-14-avg-8-context-2-max-sym-per-frame-1.txt
2022-11-21 08:57:18,939 INFO [utils.py:530] [dev_sdm-greedy_search] %WER 31.32% [29731 / 94940, 4512 ins, 7044 del, 18175 sub ]
2022-11-21 08:57:19,722 INFO [utils.py:530] [dev_sdm-greedy_search] %WER 22.44% [83014 / 369873, 19666 ins, 31039 del, 32309 sub ]
2022-11-21 08:57:20,845 INFO [decode.py:511] Wrote detailed error stats to pruned_transducer_stateless7/exp/v2/greedy_search/wers-dev_sdm-greedy_search-epoch-14-avg-8-context-2-max-sym-per-frame-1.txt
2022-11-21 08:57:20,866 INFO [decode.py:531] 
For dev_sdm, WER/CER of different settings are:
greedy_search	31.32	22.44	best for dev_sdm

2022-11-21 08:57:20,877 INFO [decode.py:726] Decoding test_sdm
2022-11-21 08:57:22,664 INFO [decode.py:469] batch 0/?, cuts processed until now is 69
2022-11-21 08:57:28,994 INFO [zipformer.py:1414] attn_weights_entropy = tensor([2.6137, 3.3224, 3.5908, 3.1441, 2.0380, 3.5516, 2.1972, 3.0422],
       device='cuda:0'), covar=tensor([0.0359, 0.0196, 0.0141, 0.0322, 0.0512, 0.0161, 0.0531, 0.0158],
       device='cuda:0'), in_proj_covar=tensor([0.0176, 0.0146, 0.0155, 0.0177, 0.0173, 0.0154, 0.0169, 0.0152],
       device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
       device='cuda:0')
2022-11-21 08:58:09,662 INFO [decode.py:485] The transcripts are stored in pruned_transducer_stateless7/exp/v2/greedy_search/recogs-test_sdm-greedy_search-epoch-14-avg-8-context-2-max-sym-per-frame-1.txt
2022-11-21 08:58:09,835 INFO [utils.py:530] [test_sdm-greedy_search] %WER 32.38% [29028 / 89659, 3955 ins, 7736 del, 17337 sub ]
2022-11-21 08:58:10,588 INFO [utils.py:530] [test_sdm-greedy_search] %WER 23.44% [83036 / 354205, 18668 ins, 33128 del, 31240 sub ]
2022-11-21 08:58:11,649 INFO [decode.py:511] Wrote detailed error stats to pruned_transducer_stateless7/exp/v2/greedy_search/wers-test_sdm-greedy_search-epoch-14-avg-8-context-2-max-sym-per-frame-1.txt
2022-11-21 08:58:11,650 INFO [decode.py:531] 
For test_sdm, WER/CER of different settings are:
greedy_search	32.38	23.44	best for test_sdm

2022-11-21 08:58:11,654 INFO [decode.py:726] Decoding dev_gss
2022-11-21 08:58:13,417 INFO [decode.py:469] batch 0/?, cuts processed until now is 71
2022-11-21 08:59:00,020 INFO [decode.py:485] The transcripts are stored in pruned_transducer_stateless7/exp/v2/greedy_search/recogs-dev_gss-greedy_search-epoch-14-avg-8-context-2-max-sym-per-frame-1.txt
2022-11-21 08:59:00,182 INFO [utils.py:530] [dev_gss-greedy_search] %WER 22.05% [20935 / 94940, 2787 ins, 4483 del, 13665 sub ]
2022-11-21 08:59:00,898 INFO [utils.py:530] [dev_gss-greedy_search] %WER 14.27% [52797 / 369873, 11721 ins, 19818 del, 21258 sub ]
2022-11-21 08:59:01,847 INFO [decode.py:511] Wrote detailed error stats to pruned_transducer_stateless7/exp/v2/greedy_search/wers-dev_gss-greedy_search-epoch-14-avg-8-context-2-max-sym-per-frame-1.txt
2022-11-21 08:59:01,848 INFO [decode.py:531] 
For dev_gss, WER/CER of different settings are:
greedy_search	22.05	14.27	best for dev_gss

2022-11-21 08:59:01,853 INFO [decode.py:726] Decoding test_gss
2022-11-21 08:59:03,727 INFO [decode.py:469] batch 0/?, cuts processed until now is 69
2022-11-21 08:59:03,886 INFO [zipformer.py:1414] attn_weights_entropy = tensor([4.4828, 4.7358, 4.2268, 4.6778, 4.7580, 3.9699, 4.3828, 4.1917],
       device='cuda:0'), covar=tensor([0.0160, 0.0243, 0.1101, 0.0280, 0.0272, 0.0336, 0.0256, 0.0375],
       device='cuda:0'), in_proj_covar=tensor([0.0111, 0.0156, 0.0254, 0.0151, 0.0197, 0.0151, 0.0167, 0.0154],
       device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
       device='cuda:0')
2022-11-21 08:59:51,508 INFO [decode.py:485] The transcripts are stored in pruned_transducer_stateless7/exp/v2/greedy_search/recogs-test_gss-greedy_search-epoch-14-avg-8-context-2-max-sym-per-frame-1.txt
2022-11-21 08:59:51,662 INFO [utils.py:530] [test_gss-greedy_search] %WER 22.93% [20560 / 89659, 2221 ins, 5099 del, 13240 sub ]
2022-11-21 08:59:52,458 INFO [utils.py:530] [test_gss-greedy_search] %WER 15.12% [53541 / 354205, 10359 ins, 21954 del, 21228 sub ]
2022-11-21 08:59:53,369 INFO [decode.py:511] Wrote detailed error stats to pruned_transducer_stateless7/exp/v2/greedy_search/wers-test_gss-greedy_search-epoch-14-avg-8-context-2-max-sym-per-frame-1.txt
2022-11-21 08:59:53,370 INFO [decode.py:531] 
For test_gss, WER/CER of different settings are:
greedy_search	22.93	15.12	best for test_gss

2022-11-21 08:59:53,375 INFO [decode.py:743] Done!