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: 1.8652
- Rewards/chosen: 125.1359
- Rewards/rejected: -204.8868
- Rewards/accuracies: 0.9141
- Rewards/margins: 330.0227
- Logps/rejected: -113.7294
- Logps/chosen: -125.2587
- Logits/rejected: -1.8708
- Logits/chosen: -1.8725
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 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 1.1504 | 0.21 | 100 | 0.9784 | 103.9547 | -116.3378 | 0.9102 | 220.2925 | -104.8745 | -127.3768 | -1.6942 | -1.7242 |
| 2.8553 | 0.42 | 200 | 1.8849 | 118.2614 | -180.4101 | 0.9102 | 298.6714 | -111.2817 | -125.9461 | -1.8462 | -1.8519 |
| 2.2897 | 0.63 | 300 | 2.1029 | 127.9046 | -196.9116 | 0.9141 | 324.8163 | -112.9319 | -124.9818 | -1.8642 | -1.8683 |
| 2.2714 | 0.84 | 400 | 1.8652 | 125.1359 | -204.8868 | 0.9141 | 330.0227 | -113.7294 | -125.2587 | -1.8708 | -1.8725 |
Framework versions
- Transformers 4.38.2
- Pytorch 2.1.2+cu118
- Datasets 2.16.1
- Tokenizers 0.15.2