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: 0.0675
- Rewards/chosen: -2.4788
- Rewards/rejected: -2.9505
- Rewards/accuracies: 0.6406
- Rewards/margins: 0.4717
- Logps/rejected: -552.4012
- Logps/chosen: -504.9170
- Logits/rejected: -2.1295
- Logits/chosen: -2.1638
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: 5
- 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 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.067 | 0.25 | 100 | 0.1174 | -1.4873 | -1.7314 | 0.6133 | 0.2442 | -430.4969 | -405.7653 | -2.3244 | -2.3408 |
| 0.0435 | 0.49 | 200 | 0.0799 | -2.1802 | -2.5492 | 0.6211 | 0.3690 | -512.2731 | -475.0585 | -2.1421 | -2.1734 |
| 0.0288 | 0.74 | 300 | 0.0710 | -2.4383 | -2.9105 | 0.6172 | 0.4722 | -548.4017 | -500.8697 | -2.1339 | -2.1675 |
| 0.032 | 0.99 | 400 | 0.0675 | -2.4788 | -2.9505 | 0.6406 | 0.4717 | -552.4012 | -504.9170 | -2.1295 | -2.1638 |
Framework versions
- Transformers 4.35.2
- Pytorch 2.1.2+cu121
- Datasets 2.14.6
- Tokenizers 0.14.1