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 an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 0.7111
- Rewards/chosen: -0.8296
- Rewards/rejected: -0.9542
- Rewards/accuracies: 0.5625
- Rewards/margins: 0.1246
- Logps/rejected: -613.8080
- Logps/chosen: -473.4393
- Logits/rejected: -5.2824
- Logits/chosen: -5.0285
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-08
- 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 |
---|---|---|---|---|---|---|---|---|---|---|---|
0.5044 | 0.2558 | 100 | 0.7105 | -0.1129 | -0.0877 | 0.4727 | -0.0252 | -527.1570 | -401.7669 | -4.8307 | -4.6554 |
0.3343 | 0.5115 | 200 | 0.6982 | -0.5200 | -0.6117 | 0.5586 | 0.0918 | -579.5609 | -442.4707 | -5.1101 | -4.8657 |
0.2972 | 0.7673 | 300 | 0.7111 | -0.8296 | -0.9542 | 0.5625 | 0.1246 | -613.8080 | -473.4393 | -5.2824 | -5.0285 |
Framework versions
- Transformers 4.40.2
- Pytorch 2.1.2+cu118
- Datasets 2.19.1
- Tokenizers 0.19.1