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: 0.5327
- Rewards/chosen: -1.3058
- Rewards/rejected: -2.0790
- Rewards/accuracies: 0.7695
- Rewards/margins: 0.7731
- Logps/rejected: -465.2492
- Logps/chosen: -387.6228
- Logits/rejected: -2.0418
- Logits/chosen: -2.0914
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 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.645 | 0.21 | 100 | 0.6138 | -0.3445 | -0.6635 | 0.7109 | 0.3190 | -323.7003 | -291.4871 | -2.5877 | -2.6051 |
| 0.6088 | 0.42 | 200 | 0.5601 | -1.2478 | -1.8506 | 0.7383 | 0.6028 | -442.4111 | -381.8215 | -2.2506 | -2.2970 |
| 0.5713 | 0.63 | 300 | 0.5404 | -1.1777 | -1.9907 | 0.7656 | 0.8130 | -456.4196 | -374.8101 | -2.1131 | -2.1610 |
| 0.585 | 0.84 | 400 | 0.5327 | -1.3058 | -2.0790 | 0.7695 | 0.7731 | -465.2492 | -387.6228 | -2.0418 | -2.0914 |
Framework versions
- Transformers 4.35.2
- Pytorch 2.1.2+cu121
- Datasets 2.14.6
- Tokenizers 0.14.1