lmind_nq_train6000_eval6489_v1_reciteonly_qa_v3_3e-4_lora2
This model is a fine-tuned version of meta-llama/Llama-2-7b-hf on the tyzhu/lmind_nq_train6000_eval6489_v1_reciteonly_qa_v3 dataset. It achieves the following results on the evaluation set:
- Loss: 2.4500
- Accuracy: 0.6438
Model description
More information needed
Intended uses & limitations
More information needed
Training and evaluation data
More information needed
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 0.0003
- train_batch_size: 2
- eval_batch_size: 2
- seed: 42
- distributed_type: multi-GPU
- num_devices: 4
- gradient_accumulation_steps: 4
- total_train_batch_size: 32
- total_eval_batch_size: 8
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: constant
- lr_scheduler_warmup_ratio: 0.05
- num_epochs: 50.0
Training results
Training Loss | Epoch | Step | Accuracy | Validation Loss |
---|---|---|---|---|
1.2608 | 1.0 | 187 | 0.6692 | 1.1864 |
1.0703 | 2.0 | 375 | 0.6702 | 1.1833 |
0.8377 | 3.0 | 562 | 0.6683 | 1.2356 |
0.5925 | 4.0 | 750 | 0.6631 | 1.3472 |
0.3927 | 5.0 | 937 | 0.6587 | 1.4937 |
0.2434 | 6.0 | 1125 | 0.6541 | 1.6856 |
0.1579 | 7.0 | 1312 | 0.6532 | 1.8412 |
0.1164 | 8.0 | 1500 | 0.6522 | 1.9470 |
0.0915 | 9.0 | 1687 | 0.6519 | 2.0593 |
0.084 | 10.0 | 1875 | 0.6507 | 2.1807 |
0.0787 | 11.0 | 2062 | 0.6511 | 2.1614 |
0.078 | 12.0 | 2250 | 0.6499 | 2.2384 |
0.0778 | 13.0 | 2437 | 0.6497 | 2.2370 |
0.0783 | 14.0 | 2625 | 0.6507 | 2.2405 |
0.0819 | 15.0 | 2812 | 0.6492 | 2.1894 |
0.085 | 16.0 | 3000 | 0.6494 | 2.1976 |
0.0801 | 17.0 | 3187 | 0.6487 | 2.1847 |
0.0812 | 18.0 | 3375 | 0.6487 | 2.2310 |
0.0785 | 19.0 | 3562 | 0.6492 | 2.2172 |
0.0771 | 20.0 | 3750 | 0.6490 | 2.2268 |
0.0747 | 21.0 | 3937 | 0.6479 | 2.2741 |
0.0849 | 22.0 | 4125 | 0.6489 | 2.2643 |
0.0799 | 23.0 | 4312 | 0.6481 | 2.2515 |
0.0815 | 24.0 | 4500 | 0.6488 | 2.2567 |
0.0728 | 25.0 | 4687 | 0.6485 | 2.2895 |
0.0717 | 26.0 | 4875 | 0.6483 | 2.3063 |
0.0706 | 27.0 | 5062 | 0.6479 | 2.3848 |
0.0722 | 28.0 | 5250 | 0.6469 | 2.3759 |
0.0728 | 29.0 | 5437 | 0.6479 | 2.3156 |
0.0731 | 30.0 | 5625 | 0.6476 | 2.2922 |
0.0752 | 31.0 | 5812 | 0.6464 | 2.3409 |
0.0768 | 32.0 | 6000 | 0.6458 | 2.2928 |
0.0709 | 33.0 | 6187 | 0.6462 | 2.3183 |
0.072 | 34.0 | 6375 | 0.6456 | 2.3278 |
0.0711 | 35.0 | 6562 | 0.6462 | 2.3750 |
0.0718 | 36.0 | 6750 | 0.6463 | 2.4077 |
0.0724 | 37.0 | 6937 | 0.6450 | 2.4183 |
0.0718 | 38.0 | 7125 | 0.6449 | 2.3822 |
0.0729 | 39.0 | 7312 | 0.6461 | 2.4076 |
0.0729 | 40.0 | 7500 | 0.6444 | 2.3906 |
0.069 | 41.0 | 7687 | 0.6453 | 2.4055 |
0.0706 | 42.0 | 7875 | 0.6442 | 2.3809 |
0.0702 | 43.0 | 8062 | 0.6451 | 2.3951 |
0.0708 | 44.0 | 8250 | 0.6440 | 2.3934 |
0.0708 | 45.0 | 8437 | 0.6448 | 2.4016 |
0.07 | 46.0 | 8625 | 0.6447 | 2.4118 |
0.0723 | 47.0 | 8812 | 0.6453 | 2.4068 |
0.0735 | 48.0 | 9000 | 0.6444 | 2.4201 |
0.0683 | 49.0 | 9187 | 0.6440 | 2.4153 |
0.0696 | 49.87 | 9350 | 0.6438 | 2.4500 |
Framework versions
- Transformers 4.34.0
- Pytorch 2.1.0+cu121
- Datasets 2.18.0
- Tokenizers 0.14.1
Model tree for tyzhu/lmind_nq_train6000_eval6489_v1_reciteonly_qa_v3_3e-4_lora2
Base model
meta-llama/Llama-2-7b-hfDataset used to train tyzhu/lmind_nq_train6000_eval6489_v1_reciteonly_qa_v3_3e-4_lora2
Evaluation results
- Accuracy on tyzhu/lmind_nq_train6000_eval6489_v1_reciteonly_qa_v3self-reported0.644