metadata
license: apache-2.0
base_model: mistralai/Mistral-7B-Instruct-v0.2
tags:
- trl
- dpo
- generated_from_trainer
model-index:
- name: mistralit2_1000_STEPS_1e6_rate_05_beta_DPO
results: []
mistralit2_1000_STEPS_1e6_rate_05_beta_DPO
This model is a fine-tuned version of mistralai/Mistral-7B-Instruct-v0.2 on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 0.6223
- Rewards/chosen: -1.9087
- Rewards/rejected: -2.8966
- Rewards/accuracies: 0.6593
- Rewards/margins: 0.9879
- Logps/rejected: -34.3656
- Logps/chosen: -27.2032
- Logits/rejected: -2.8455
- Logits/chosen: -2.8459
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: 4
- eval_batch_size: 1
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 8
- 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 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.6684 | 0.1 | 50 | 0.6660 | -0.2264 | -0.2957 | 0.5934 | 0.0693 | -29.1637 | -23.8386 | -2.8636 | -2.8639 |
| 0.5945 | 0.2 | 100 | 0.6396 | -1.5064 | -1.9635 | 0.6044 | 0.4572 | -32.4994 | -26.3985 | -2.8444 | -2.8447 |
| 0.4899 | 0.29 | 150 | 0.6602 | -2.2474 | -2.9308 | 0.6022 | 0.6835 | -34.4341 | -27.8806 | -2.8445 | -2.8448 |
| 0.5517 | 0.39 | 200 | 0.6024 | -0.7758 | -1.2571 | 0.6418 | 0.4813 | -31.0867 | -24.9374 | -2.8613 | -2.8616 |
| 0.6385 | 0.49 | 250 | 0.5703 | -0.5516 | -1.1264 | 0.6703 | 0.5749 | -30.8253 | -24.4890 | -2.8571 | -2.8574 |
| 0.5653 | 0.59 | 300 | 0.5989 | -1.4256 | -2.1727 | 0.6440 | 0.7471 | -32.9178 | -26.2370 | -2.8464 | -2.8467 |
| 0.5255 | 0.68 | 350 | 0.6054 | -1.6264 | -2.4443 | 0.6484 | 0.8179 | -33.4610 | -26.6386 | -2.8533 | -2.8536 |
| 0.6612 | 0.78 | 400 | 0.6157 | -1.7163 | -2.5329 | 0.6418 | 0.8166 | -33.6383 | -26.8185 | -2.8530 | -2.8533 |
| 0.646 | 0.88 | 450 | 0.6016 | -1.1753 | -1.8651 | 0.6440 | 0.6898 | -32.3026 | -25.7364 | -2.8525 | -2.8529 |
| 0.5146 | 0.98 | 500 | 0.5957 | -1.1531 | -1.8752 | 0.6484 | 0.7221 | -32.3227 | -25.6920 | -2.8553 | -2.8556 |
| 0.297 | 1.07 | 550 | 0.5863 | -1.2310 | -2.0319 | 0.6571 | 0.8009 | -32.6362 | -25.8478 | -2.8539 | -2.8542 |
| 0.2709 | 1.17 | 600 | 0.6234 | -1.7413 | -2.6395 | 0.6527 | 0.8982 | -33.8514 | -26.8684 | -2.8489 | -2.8493 |
| 0.4008 | 1.27 | 650 | 0.6173 | -1.8482 | -2.8001 | 0.6549 | 0.9519 | -34.1726 | -27.0823 | -2.8472 | -2.8476 |
| 0.2846 | 1.37 | 700 | 0.6222 | -1.8576 | -2.8175 | 0.6505 | 0.9599 | -34.2075 | -27.1011 | -2.8466 | -2.8470 |
| 0.2129 | 1.46 | 750 | 0.6233 | -1.8931 | -2.8716 | 0.6571 | 0.9785 | -34.3156 | -27.1720 | -2.8458 | -2.8462 |
| 0.3026 | 1.56 | 800 | 0.6224 | -1.9044 | -2.8881 | 0.6593 | 0.9837 | -34.3486 | -27.1947 | -2.8455 | -2.8458 |
| 0.3361 | 1.66 | 850 | 0.6242 | -1.9113 | -2.9007 | 0.6659 | 0.9894 | -34.3738 | -27.2085 | -2.8456 | -2.8460 |
| 0.2965 | 1.76 | 900 | 0.6223 | -1.9123 | -2.8982 | 0.6615 | 0.9859 | -34.3687 | -27.2103 | -2.8456 | -2.8460 |
| 0.2779 | 1.86 | 950 | 0.6213 | -1.9078 | -2.8977 | 0.6593 | 0.9900 | -34.3678 | -27.2013 | -2.8455 | -2.8459 |
| 0.2334 | 1.95 | 1000 | 0.6223 | -1.9087 | -2.8966 | 0.6593 | 0.9879 | -34.3656 | -27.2032 | -2.8455 | -2.8459 |
Framework versions
- Transformers 4.38.2
- Pytorch 2.0.0+cu117
- Datasets 2.18.0
- Tokenizers 0.15.2