metadata
license: apache-2.0
base_model: mistralai/Mistral-7B-Instruct-v0.1
tags:
- trl
- dpo
- generated_from_trainer
model-index:
- name: v1_1000_STEPS_1e7_rate_01_beta_DPO
results: []
v1_1000_STEPS_1e7_rate_01_beta_DPO
This model is a fine-tuned version of mistralai/Mistral-7B-Instruct-v0.1 on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 0.6730
- Rewards/chosen: -0.0669
- Rewards/rejected: -0.1113
- Rewards/accuracies: 0.5890
- Rewards/margins: 0.0445
- Logps/rejected: -17.9930
- Logps/chosen: -15.9218
- Logits/rejected: -3.3417
- Logits/chosen: -3.3418
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: 1e-07
- train_batch_size: 2
- eval_batch_size: 1
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 4
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_steps: 100
- training_steps: 1000
Training results
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.6944 | 0.05 | 50 | 0.6930 | -0.0001 | -0.0004 | 0.4791 | 0.0003 | -16.8836 | -15.2543 | -3.3540 | -3.3541 |
| 0.6896 | 0.1 | 100 | 0.6907 | -0.0026 | -0.0076 | 0.5670 | 0.0050 | -16.9551 | -15.2788 | -3.3527 | -3.3528 |
| 0.6879 | 0.15 | 150 | 0.6878 | -0.0076 | -0.0188 | 0.5736 | 0.0112 | -17.0680 | -15.3294 | -3.3516 | -3.3517 |
| 0.6836 | 0.2 | 200 | 0.6849 | -0.0190 | -0.0363 | 0.5670 | 0.0173 | -17.2422 | -15.4426 | -3.3479 | -3.3480 |
| 0.6804 | 0.24 | 250 | 0.6825 | -0.0285 | -0.0510 | 0.5868 | 0.0226 | -17.3899 | -15.5377 | -3.3456 | -3.3457 |
| 0.6753 | 0.29 | 300 | 0.6802 | -0.0411 | -0.0689 | 0.5890 | 0.0277 | -17.5681 | -15.6645 | -3.3452 | -3.3453 |
| 0.6908 | 0.34 | 350 | 0.6788 | -0.0382 | -0.0690 | 0.5956 | 0.0307 | -17.5691 | -15.6352 | -3.3447 | -3.3448 |
| 0.6881 | 0.39 | 400 | 0.6773 | -0.0391 | -0.0735 | 0.5934 | 0.0344 | -17.6147 | -15.6439 | -3.3446 | -3.3447 |
| 0.6519 | 0.44 | 450 | 0.6757 | -0.0500 | -0.0881 | 0.5912 | 0.0381 | -17.7606 | -15.7528 | -3.3434 | -3.3435 |
| 0.6871 | 0.49 | 500 | 0.6751 | -0.0504 | -0.0897 | 0.5978 | 0.0394 | -17.7768 | -15.7565 | -3.3425 | -3.3426 |
| 0.6495 | 0.54 | 550 | 0.6737 | -0.0598 | -0.1025 | 0.5934 | 0.0427 | -17.9043 | -15.8506 | -3.3424 | -3.3425 |
| 0.6756 | 0.59 | 600 | 0.6738 | -0.0611 | -0.1038 | 0.5912 | 0.0427 | -17.9179 | -15.8641 | -3.3420 | -3.3421 |
| 0.6584 | 0.64 | 650 | 0.6735 | -0.0625 | -0.1058 | 0.5890 | 0.0434 | -17.9379 | -15.8778 | -3.3422 | -3.3423 |
| 0.6747 | 0.68 | 700 | 0.6734 | -0.0652 | -0.1089 | 0.5824 | 0.0437 | -17.9690 | -15.9052 | -3.3417 | -3.3418 |
| 0.6735 | 0.73 | 750 | 0.6733 | -0.0662 | -0.1102 | 0.5670 | 0.0440 | -17.9819 | -15.9150 | -3.3417 | -3.3418 |
| 0.6573 | 0.78 | 800 | 0.6732 | -0.0671 | -0.1112 | 0.5868 | 0.0442 | -17.9917 | -15.9236 | -3.3417 | -3.3418 |
| 0.6768 | 0.83 | 850 | 0.6732 | -0.0671 | -0.1112 | 0.5934 | 0.0441 | -17.9912 | -15.9238 | -3.3417 | -3.3418 |
| 0.6745 | 0.88 | 900 | 0.6733 | -0.0671 | -0.1110 | 0.5780 | 0.0439 | -17.9897 | -15.9243 | -3.3416 | -3.3418 |
| 0.6751 | 0.93 | 950 | 0.6730 | -0.0668 | -0.1114 | 0.5868 | 0.0446 | -17.9934 | -15.9211 | -3.3417 | -3.3418 |
| 0.6645 | 0.98 | 1000 | 0.6730 | -0.0669 | -0.1113 | 0.5890 | 0.0445 | -17.9930 | -15.9218 | -3.3417 | -3.3418 |
Framework versions
- Transformers 4.39.1
- Pytorch 2.0.0+cu117
- Datasets 2.18.0
- Tokenizers 0.15.2