mistral-7b-dpo-full-sft-wo-kqa_golden
This model is a fine-tuned version of Minbyul/mistral-7b-wo-kqa_golden-sft on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:
- Loss: 0.0018
- Rewards/chosen: -0.4458
- Rewards/rejected: -10.1099
- Rewards/accuracies: 1.0
- Rewards/margins: 9.6641
- Logps/rejected: -1564.3792
- Logps/chosen: -241.2112
- Logits/rejected: -2.0516
- Logits/chosen: -1.3414
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: 4
- gradient_accumulation_steps: 2
- total_train_batch_size: 64
- total_eval_batch_size: 32
- 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 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.2478 | 0.31 | 100 | 0.0352 | -0.1739 | -4.4264 | 1.0 | 4.2525 | -996.0294 | -214.0196 | -2.9200 | -2.1162 |
| 0.1385 | 0.61 | 200 | 0.0041 | -0.3360 | -8.1997 | 1.0 | 7.8637 | -1373.3590 | -230.2282 | -2.3336 | -1.6287 |
| 0.0899 | 0.92 | 300 | 0.0019 | -0.4479 | -10.0624 | 1.0 | 9.6145 | -1559.6263 | -241.4165 | -2.0553 | -1.3416 |
Framework versions
- Transformers 4.39.0.dev0
- Pytorch 2.1.2
- Datasets 2.14.6
- Tokenizers 0.15.2