metadata
tags:
- trl
- dpo
- generated_from_trainer
model-index:
- name: zephyr-7b-dpo-full
results: []
zephyr-7b-dpo-full
This model was trained from scratch on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.5366
- Rewards/chosen: -2.9738
- Rewards/rejected: -4.4991
- Rewards/accuracies: 0.7617
- Rewards/margins: 1.5252
- Logps/rejected: -767.4317
- Logps/chosen: -609.1594
- Logits/rejected: 1.6095
- Logits/chosen: 0.9559
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: 5e-07
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- distributed_type: multi-GPU
- num_devices: 8
- gradient_accumulation_steps: 2
- total_train_batch_size: 128
- total_eval_batch_size: 64
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 1
Training results
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.5905 | 0.07 | 100 | 0.6429 | -0.1380 | -0.3441 | 0.6719 | 0.2061 | -351.9318 | -325.5744 | -1.7244 | -1.7878 |
| 0.4495 | 0.15 | 200 | 0.5600 | -0.4940 | -1.0973 | 0.7461 | 0.6032 | -427.2510 | -361.1815 | -1.3665 | -1.4371 |
| 0.3963 | 0.22 | 300 | 0.5291 | -1.1123 | -2.0359 | 0.7422 | 0.9236 | -521.1155 | -423.0034 | -1.2770 | -1.4609 |
| 0.4012 | 0.3 | 400 | 0.5315 | -1.0588 | -1.9923 | 0.7734 | 0.9334 | -516.7505 | -417.6586 | -1.1223 | -1.3373 |
| 0.3559 | 0.37 | 500 | 0.5276 | -1.4423 | -2.5146 | 0.7578 | 1.0723 | -568.9822 | -456.0086 | -0.6834 | -1.0067 |
| 0.3291 | 0.45 | 600 | 0.5103 | -1.6617 | -2.7811 | 0.7695 | 1.1194 | -595.6332 | -477.9445 | 0.1886 | -0.2334 |
| 0.2735 | 0.52 | 700 | 0.5289 | -2.2950 | -3.7006 | 0.7617 | 1.4056 | -687.5872 | -541.2795 | 0.6722 | 0.1870 |
| 0.2752 | 0.59 | 800 | 0.5229 | -2.2134 | -3.5070 | 0.7656 | 1.2935 | -668.2236 | -533.1202 | 0.2752 | -0.1628 |
| 0.2492 | 0.67 | 900 | 0.5152 | -2.0646 | -3.3529 | 0.7734 | 1.2882 | -652.8116 | -518.2382 | 1.0726 | 0.5184 |
| 0.262 | 0.74 | 1000 | 0.5241 | -2.4505 | -3.8564 | 0.7617 | 1.4059 | -703.1603 | -556.8265 | 1.3124 | 0.6805 |
| 0.2299 | 0.82 | 1100 | 0.5313 | -2.7647 | -4.2433 | 0.7578 | 1.4786 | -741.8574 | -588.2495 | 1.4834 | 0.8391 |
| 0.1974 | 0.89 | 1200 | 0.5367 | -2.9484 | -4.4713 | 0.7617 | 1.5229 | -764.6512 | -606.6174 | 1.5458 | 0.8964 |
| 0.1842 | 0.97 | 1300 | 0.5366 | -2.9738 | -4.4991 | 0.7617 | 1.5252 | -767.4317 | -609.1594 | 1.6095 | 0.9559 |
Framework versions
- Transformers 4.38.2
- Pytorch 2.1.2+cu118
- Datasets 2.16.1
- Tokenizers 0.15.2