gpt-imdb-cdpo_0.15-beta_0.1
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: 0.5181
- Rewards/chosen: -0.6104
- Rewards/rejected: -1.9969
- Rewards/accuracies: 0.9271
- Rewards/margins: 1.3866
- Logps/rejected: -283.6544
- Logps/chosen: -241.3688
- Logits/rejected: -36.1797
- Logits/chosen: -37.0193
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
- num_epochs: 3
Training results
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.5541 | 0.21 | 500 | 0.5598 | -0.1801 | -1.1214 | 0.8417 | 0.9413 | -274.8995 | -237.0667 | -33.1267 | -34.0864 |
| 0.5399 | 0.42 | 1000 | 0.5555 | -0.4075 | -1.5309 | 0.8604 | 1.1234 | -278.9942 | -239.3399 | -36.6366 | -37.5032 |
| 0.5379 | 0.63 | 1500 | 0.5445 | -0.5885 | -1.8167 | 0.875 | 1.2282 | -281.8521 | -241.1506 | -34.0236 | -34.9075 |
| 0.5224 | 0.83 | 2000 | 0.5347 | -0.4581 | -1.7693 | 0.8917 | 1.3112 | -281.3783 | -239.8462 | -34.9412 | -35.8186 |
| 0.4992 | 1.04 | 2500 | 0.5318 | -0.5998 | -1.9222 | 0.9000 | 1.3224 | -282.9072 | -241.2631 | -34.8041 | -35.6967 |
| 0.5654 | 1.25 | 3000 | 0.5308 | -0.5502 | -1.9299 | 0.9021 | 1.3797 | -282.9844 | -240.7672 | -35.6718 | -36.5937 |
| 0.5382 | 1.46 | 3500 | 0.5247 | -0.4952 | -1.8522 | 0.9125 | 1.3570 | -282.2072 | -240.2172 | -35.7229 | -36.6547 |
| 0.5409 | 1.67 | 4000 | 0.5220 | -0.5742 | -1.9755 | 0.9292 | 1.4013 | -283.4403 | -241.0072 | -36.4780 | -37.3339 |
| 0.4911 | 1.88 | 4500 | 0.5186 | -0.6281 | -2.0249 | 0.9271 | 1.3967 | -283.9341 | -241.5466 | -36.1014 | -36.8989 |
| 0.5007 | 2.08 | 5000 | 0.5170 | -0.6115 | -2.0085 | 0.9312 | 1.3969 | -283.7699 | -241.3805 | -36.7092 | -37.5360 |
| 0.4714 | 2.29 | 5500 | 0.5166 | -0.5400 | -1.9265 | 0.9229 | 1.3865 | -282.9501 | -240.6650 | -36.1382 | -36.9914 |
| 0.5159 | 2.5 | 6000 | 0.5168 | -0.5925 | -1.9754 | 0.9271 | 1.3829 | -283.4395 | -241.1906 | -35.9587 | -36.8156 |
| 0.5103 | 2.71 | 6500 | 0.5171 | -0.6197 | -2.0190 | 0.9333 | 1.3993 | -283.8753 | -241.4619 | -36.0316 | -36.8825 |
| 0.5049 | 2.92 | 7000 | 0.5181 | -0.6104 | -1.9969 | 0.9271 | 1.3866 | -283.6544 | -241.3688 | -36.1797 | -37.0193 |
Framework versions
- Transformers 4.35.2
- Pytorch 2.1.1
- Datasets 2.15.0
- Tokenizers 0.15.0
- Downloads last month
- 9
Inference API (serverless) is not available, repository is disabled.
Model tree for Myashka/gpt-imdb-cdpo_0.15-beta_0.1
Base model
lvwerra/gpt2-imdb
Finetuned
this model