metadata
base_model: lvwerra/gpt2-imdb
tags:
- generated_from_trainer
model-index:
- name: gpt-imdb-ipo-beta_0.5
results: []
gpt-imdb-ipo-beta_0.5
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: 0.9628
- Rewards/chosen: -0.4934
- Rewards/rejected: -0.8358
- Rewards/accuracies: 0.7812
- Rewards/margins: 0.3424
- Logps/rejected: -265.3568
- Logps/chosen: -236.2520
- Logits/rejected: -32.5835
- Logits/chosen: -32.6621
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 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 10.732 | 0.21 | 500 | 21.6330 | -0.2465 | -0.4751 | 0.5792 | 0.2286 | -264.6355 | -235.7583 | -34.3644 | -34.6229 |
| 11.0252 | 0.42 | 1000 | 17.5281 | 0.3734 | 0.1008 | 0.5437 | 0.2726 | -263.4837 | -234.5185 | -35.1543 | -35.3784 |
| 17.5294 | 0.63 | 1500 | 18.4782 | -0.4521 | -0.6725 | 0.6208 | 0.2203 | -265.0302 | -236.1696 | -33.9319 | -34.0933 |
| 7.8398 | 0.83 | 2000 | 17.4130 | -0.5472 | -0.6406 | 0.6083 | 0.0933 | -264.9664 | -236.3597 | -34.0128 | -34.1803 |
| 6.2214 | 1.04 | 2500 | 9.4072 | -0.5101 | -0.8182 | 0.6292 | 0.3080 | -265.3216 | -236.2855 | -33.2396 | -33.3578 |
| 9.8652 | 1.25 | 3000 | 13.4878 | -0.6413 | -0.8801 | 0.6375 | 0.2388 | -265.4454 | -236.5479 | -32.0018 | -32.1655 |
| 11.4779 | 1.46 | 3500 | 7.5245 | -0.0755 | -0.3944 | 0.6750 | 0.3189 | -264.4740 | -235.4162 | -32.8982 | -33.0074 |
| 3.9833 | 1.67 | 4000 | 4.4888 | -0.7021 | -1.0680 | 0.6729 | 0.3659 | -265.8214 | -236.6695 | -32.9502 | -33.0304 |
| 3.389 | 1.88 | 4500 | 3.9317 | -0.5045 | -0.8887 | 0.7271 | 0.3841 | -265.4626 | -236.2743 | -32.7817 | -32.8828 |
| 3.2338 | 2.08 | 5000 | 2.4116 | -0.5185 | -0.8672 | 0.7146 | 0.3487 | -265.4196 | -236.3022 | -32.5025 | -32.5681 |
| 1.2381 | 2.29 | 5500 | 2.1558 | -0.5066 | -0.8815 | 0.7458 | 0.3749 | -265.4483 | -236.2784 | -32.3108 | -32.3902 |
| 1.6263 | 2.5 | 6000 | 1.1972 | -0.5280 | -0.8664 | 0.7396 | 0.3384 | -265.4182 | -236.3213 | -32.5356 | -32.6104 |
| 1.0882 | 2.71 | 6500 | 1.1163 | -0.5303 | -0.8584 | 0.7562 | 0.3281 | -265.4022 | -236.3259 | -32.5615 | -32.6406 |
| 1.0559 | 2.92 | 7000 | 0.9628 | -0.4934 | -0.8358 | 0.7812 | 0.3424 | -265.3568 | -236.2520 | -32.5835 | -32.6621 |
Framework versions
- Transformers 4.35.2
- Pytorch 2.1.1
- Datasets 2.15.0
- Tokenizers 0.15.0