gpt-imdb-dpo_annealing
This model is a fine-tuned version of lvwerra/gpt2-imdb on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 0.3482
- Rewards/chosen: -13.2925
- Rewards/rejected: -37.2767
- Rewards/accuracies: 0.9354
- Rewards/margins: 23.9842
- Logps/rejected: -302.0002
- Logps/chosen: -248.9281
- Logits/rejected: -38.9773
- Logits/chosen: -40.1868
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-05
- train_batch_size: 24
- eval_batch_size: 24
- seed: 42
- optimizer: Adam with betas=(0.9,0.99) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_steps: 150
- training_steps: 7197
Training results
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.2713 | 0.21 | 500 | 0.3576 | -0.9589 | -2.8806 | 0.8417 | 1.9217 | -300.2507 | -247.4370 | -34.9635 | -36.2514 |
| 0.2605 | 0.42 | 1000 | 0.2876 | -1.8668 | -5.2245 | 0.8708 | 3.3577 | -299.0920 | -247.9165 | -39.8673 | -41.1403 |
| 0.134 | 0.63 | 1500 | 0.2827 | -3.3220 | -8.2599 | 0.8833 | 4.9379 | -301.8662 | -250.6212 | -38.4289 | -39.6488 |
| 0.2246 | 0.83 | 2000 | 0.2412 | -3.0672 | -9.5366 | 0.9000 | 6.4694 | -297.1335 | -246.0230 | -36.9979 | -38.2478 |
| 0.0612 | 1.04 | 2500 | 0.2382 | -4.4276 | -12.4767 | 0.9062 | 8.0491 | -298.9408 | -247.7763 | -38.3549 | -39.5684 |
| 0.2336 | 1.25 | 3000 | 0.2628 | -5.5352 | -15.3372 | 0.9042 | 9.8020 | -299.9716 | -248.3611 | -39.0799 | -40.3999 |
| 0.1755 | 1.46 | 3500 | 0.2670 | -6.0750 | -18.0326 | 0.9229 | 11.9576 | -300.3778 | -247.6266 | -38.3635 | -39.7127 |
| 0.34 | 1.67 | 4000 | 0.2499 | -7.2657 | -20.1377 | 0.9208 | 12.8719 | -299.6307 | -248.2345 | -38.0993 | -39.2549 |
| 0.1822 | 1.88 | 4500 | 0.3000 | -7.9584 | -22.7421 | 0.9271 | 14.7838 | -299.8409 | -247.9176 | -38.7806 | -39.9153 |
| 0.153 | 2.08 | 5000 | 0.2972 | -9.4217 | -26.8046 | 0.9333 | 17.3829 | -302.0991 | -248.7675 | -38.2977 | -39.5006 |
| 0.0004 | 2.29 | 5500 | 0.2962 | -9.6704 | -28.5833 | 0.9354 | 18.9129 | -300.9727 | -247.8805 | -38.6801 | -39.9033 |
| 0.0584 | 2.5 | 6000 | 0.3113 | -11.3462 | -31.8850 | 0.9375 | 20.5388 | -301.8552 | -248.8479 | -38.5484 | -39.7563 |
| 0.0304 | 2.71 | 6500 | 0.3441 | -12.4687 | -34.7986 | 0.9354 | 22.3299 | -302.1741 | -249.0562 | -38.8388 | -40.0519 |
| 0.223 | 2.92 | 7000 | 0.3482 | -13.2925 | -37.2767 | 0.9354 | 23.9842 | -302.0002 | -248.9281 | -38.9773 | -40.1868 |
Framework versions
- Transformers 4.35.2
- Pytorch 2.1.1
- Datasets 2.15.0
- Tokenizers 0.15.0
- Downloads last month
- 6