metadata
license: mit
base_model: HuggingFaceH4/zephyr-7b-beta
tags:
- generated_from_trainer
model-index:
- name: zephyr-7b-dpo-lora
results: []
zephyr-7b-dpo-lora
This model is a fine-tuned version of HuggingFaceH4/zephyr-7b-beta on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.6874
- Rewards/chosen: 0.0803
- Rewards/rejected: 0.0298
- Rewards/accuracies: 1.0
- Rewards/margins: 0.0505
- Logps/rejected: -101.0604
- Logps/chosen: -102.9630
- Logits/rejected: -2.2160
- Logits/chosen: -2.1724
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: 2
- eval_batch_size: 4
- seed: 42
- distributed_type: multi-GPU
- num_devices: 8
- gradient_accumulation_steps: 32
- total_train_batch_size: 512
- total_eval_batch_size: 32
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 5
Training results
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.6931 | 0.8 | 1 | 0.6931 | 0.0 | 0.0 | 0.0 | 0.0 | -101.3584 | -103.7664 | -2.2157 | -2.1717 |
| 0.6931 | 1.6 | 2 | 0.6948 | 0.0296 | 0.0079 | 0.5 | 0.0217 | -101.2790 | -103.4700 | -2.2147 | -2.1715 |
| 0.6931 | 2.4 | 3 | 0.6913 | 0.0277 | 0.0090 | 0.75 | 0.0188 | -101.2689 | -103.4891 | -2.2153 | -2.1709 |
| 0.6931 | 4.0 | 5 | 0.6874 | 0.0803 | 0.0298 | 1.0 | 0.0505 | -101.0604 | -102.9630 | -2.2160 | -2.1724 |
Framework versions
- Transformers 4.35.0
- Pytorch 2.1.0
- Datasets 2.14.6
- Tokenizers 0.14.1