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: 0.5261
- Rewards/chosen: -2.4591
- Rewards/rejected: -3.9221
- Rewards/accuracies: 0.7773
- Rewards/margins: 1.4631
- Logps/rejected: -703.8400
- Logps/chosen: -549.4910
- Logits/rejected: 0.0289
- Logits/chosen: 0.0663
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-06
- train_batch_size: 4
- eval_batch_size: 8
- seed: 2
- distributed_type: multi-GPU
- num_devices: 8
- gradient_accumulation_steps: 4
- 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: 2
Training results
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.6201 | 0.21 | 100 | 0.6253 | -0.2753 | -0.6662 | 0.7031 | 0.3909 | -378.2405 | -331.1124 | 0.4172 | 0.3706 |
| 0.5547 | 0.42 | 200 | 0.5549 | -0.6988 | -1.4726 | 0.7656 | 0.7738 | -458.8863 | -373.4661 | 0.4261 | 0.3909 |
| 0.5343 | 0.63 | 300 | 0.5316 | -0.8044 | -1.6474 | 0.7656 | 0.8430 | -476.3628 | -384.0199 | 0.2851 | 0.2449 |
| 0.5323 | 0.84 | 400 | 0.5211 | -0.9068 | -1.8283 | 0.7812 | 0.9216 | -494.4600 | -394.2621 | 0.2834 | 0.2514 |
| 0.352 | 1.05 | 500 | 0.5258 | -1.9533 | -3.4166 | 0.7969 | 1.4634 | -653.2899 | -498.9117 | -0.0846 | -0.0654 |
| 0.3342 | 1.26 | 600 | 0.5268 | -2.3123 | -3.7246 | 0.7930 | 1.4124 | -684.0857 | -534.8101 | 0.1128 | 0.1344 |
| 0.337 | 1.47 | 700 | 0.5290 | -2.3753 | -3.8837 | 0.7773 | 1.5084 | -699.9910 | -541.1116 | 0.0099 | 0.0414 |
| 0.3398 | 1.67 | 800 | 0.5297 | -2.5097 | -4.0133 | 0.7734 | 1.5036 | -712.9506 | -554.5546 | 0.0381 | 0.0750 |
| 0.307 | 1.88 | 900 | 0.5261 | -2.4591 | -3.9221 | 0.7773 | 1.4631 | -703.8400 | -549.4910 | 0.0289 | 0.0663 |
Framework versions
- Transformers 4.35.2
- Pytorch 2.1.2+cu121
- Datasets 2.14.6
- Tokenizers 0.14.1