metadata
tags:
- trl
- dpo
- generated_from_trainer
model-index:
- name: zephyr-7b-dpo-full
results: []
zephyr-7b-dpo-full
This model was trained from scratch on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.4801
- Rewards/chosen: -0.7670
- Rewards/rejected: -1.9120
- Rewards/accuracies: 0.7617
- Rewards/margins: 1.1450
- Logps/rejected: -508.7245
- Logps/chosen: -388.4799
- Logits/rejected: 0.7285
- Logits/chosen: 0.2273
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 |
---|---|---|---|---|---|---|---|---|---|---|---|
0.539 | 0.21 | 100 | 0.5452 | -0.5988 | -1.3188 | 0.7422 | 0.7200 | -449.4053 | -371.6547 | -0.7240 | -0.8812 |
0.5294 | 0.42 | 200 | 0.5000 | -0.6654 | -1.5376 | 0.7617 | 0.8722 | -471.2849 | -378.3161 | 0.8080 | 0.4502 |
0.4704 | 0.63 | 300 | 0.4878 | -0.8441 | -1.9900 | 0.7539 | 1.1459 | -516.5240 | -396.1895 | 0.8998 | 0.3977 |
0.4856 | 0.84 | 400 | 0.4801 | -0.7670 | -1.9120 | 0.7617 | 1.1450 | -508.7245 | -388.4799 | 0.7285 | 0.2273 |
Framework versions
- Transformers 4.38.2
- Pytorch 2.1.2+cu118
- Datasets 2.16.1
- Tokenizers 0.15.2