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: 0.0680
- Rewards/chosen: -1.6802
- Rewards/rejected: -2.4505
- Rewards/accuracies: 0.7109
- Rewards/margins: 0.7703
- Logps/rejected: -502.4064
- Logps/chosen: -425.0607
- Logits/rejected: -2.2693
- Logits/chosen: -2.2870
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 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.1201 | 0.21 | 100 | 0.1358 | -0.5060 | -0.9026 | 0.6992 | 0.3966 | -347.6168 | -307.6421 | -2.7272 | -2.7404 |
| 0.0885 | 0.42 | 200 | 0.0939 | -0.9340 | -1.6072 | 0.7383 | 0.6732 | -418.0752 | -350.4443 | -2.5184 | -2.5309 |
| 0.0652 | 0.63 | 300 | 0.0711 | -1.5440 | -2.2912 | 0.7266 | 0.7471 | -486.4697 | -411.4413 | -2.3324 | -2.3504 |
| 0.0725 | 0.84 | 400 | 0.0680 | -1.6802 | -2.4505 | 0.7109 | 0.7703 | -502.4064 | -425.0607 | -2.2693 | -2.2870 |
Framework versions
- Transformers 4.35.2
- Pytorch 2.1.2+cu121
- Datasets 2.14.6
- Tokenizers 0.14.1