metadata
license: mit
base_model: HuggingFaceH4/mistral-7b-sft-beta
tags:
- trl
- dpo
- generated_from_trainer
model-index:
- name: zephyr-7b-dpo-full
results: []
zephyr-7b-dpo-full
This model is a fine-tuned version of HuggingFaceH4/mistral-7b-sft-beta on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.0632
- Rewards/chosen: -1.3406
- Rewards/rejected: -2.3147
- Rewards/accuracies: 0.7734
- Rewards/margins: 0.9740
- Logps/rejected: -488.8222
- Logps/chosen: -391.1042
- Logits/rejected: -2.0084
- Logits/chosen: -2.0472
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: 1
- 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 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.0717 | 0.21 | 100 | 0.0733 | -0.5920 | -1.1062 | 0.7266 | 0.5142 | -367.9753 | -316.2405 | -2.7130 | -2.7290 |
| 0.0672 | 0.42 | 200 | 0.0662 | -0.9311 | -1.7199 | 0.7422 | 0.7888 | -429.3445 | -350.1472 | -2.2044 | -2.2377 |
| 0.0648 | 0.63 | 300 | 0.0643 | -1.2563 | -2.1377 | 0.7734 | 0.8814 | -471.1217 | -382.6705 | -2.0727 | -2.1098 |
| 0.0636 | 0.84 | 400 | 0.0632 | -1.3406 | -2.3147 | 0.7734 | 0.9740 | -488.8222 | -391.1042 | -2.0084 | -2.0472 |
Framework versions
- Transformers 4.35.2
- Pytorch 2.1.2+cu121
- Datasets 2.14.6
- Tokenizers 0.14.1