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: 1.3947
- Rewards/chosen: -2.4314
- Rewards/rejected: -2.0023
- Rewards/accuracies: 0.3867
- Rewards/margins: -0.4292
- Logps/rejected: -517.7516
- Logps/chosen: -554.9180
- Logits/rejected: -1.0823
- Logits/chosen: -1.1239
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 | Logits/chosen | Logits/rejected | Logps/chosen | Logps/rejected | Validation Loss | Rewards/accuracies | Rewards/chosen | Rewards/margins | Rewards/rejected |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.3047 | 0.1 | 100 | -2.4405 | -2.3863 | -361.0801 | -337.7748 | 0.8551 | 0.3203 | -0.4930 | -0.2905 | -0.2025 |
| 0.1861 | 0.21 | 200 | -1.5418 | -1.5107 | -450.2716 | -421.0934 | 1.0495 | 0.3867 | -1.3850 | -0.3493 | -1.0357 |
| 0.1608 | 0.31 | 300 | -1.4367 | -1.4022 | -454.9446 | -422.9684 | 1.0910 | 0.3945 | -1.4317 | -0.3772 | -1.0544 |
| 0.1368 | 0.42 | 400 | -1.0538 | -1.0131 | -520.1699 | -479.6456 | 1.3010 | 0.4102 | -2.0839 | -0.4627 | -1.6212 |
| 0.1364 | 0.52 | 500 | -1.6466 | -1.6090 | -470.0934 | -430.8614 | 1.1773 | 0.3711 | -1.5832 | -0.4498 | -1.1334 |
| 0.1223 | 0.63 | 600 | 1.3206 | -2.2971 | -1.8297 | 0.4141 | -0.4674 | -500.4930 | -541.4883 | -1.1541 | -1.1880 |
| 0.0971 | 0.73 | 700 | 1.4638 | -2.6554 | -2.1594 | 0.3906 | -0.4959 | -533.4667 | -577.3128 | -0.9392 | -0.9712 |
| 0.1035 | 0.84 | 800 | 1.4475 | -2.5761 | -2.1538 | 0.3945 | -0.4222 | -532.9068 | -569.3817 | -0.8902 | -0.9232 |
| 0.088 | 0.94 | 900 | 1.3947 | -2.4314 | -2.0023 | 0.3867 | -0.4292 | -517.7516 | -554.9180 | -1.0823 | -1.1239 |
Framework versions
- Transformers 4.38.2
- Pytorch 2.1.2+cu118
- Datasets 2.16.1
- Tokenizers 0.15.2