selfbiorag-7b-dpo-full-wo-kqa_golden-ep3
This model is a fine-tuned version of dmis-lab/selfbiorag_7b on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:
- Loss: 0.6355
- Rewards/chosen: 0.1376
- Rewards/rejected: -0.0007
- Rewards/accuracies: 0.7326
- Rewards/margins: 0.1383
- Logps/rejected: -144.1728
- Logps/chosen: -121.3669
- Logits/rejected: -1.5473
- Logits/chosen: -1.8108
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 | Logits/chosen | Logits/rejected | Logps/chosen | Logps/rejected | Validation Loss | Rewards/accuracies | Rewards/chosen | Rewards/margins | Rewards/rejected |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.6595 | 0.22 | 100 | -1.7720 | -1.4957 | -124.0620 | -139.3042 | 0.6647 | 0.6875 | 0.1107 | 0.0627 | 0.0480 |
| 0.6273 | 0.45 | 200 | -1.7316 | -1.4653 | -119.3853 | -138.4957 | 0.6494 | 0.6979 | 0.1574 | 0.1013 | 0.0561 |
| 0.6009 | 0.67 | 300 | -1.7770 | -1.5098 | -120.2743 | -141.8649 | 0.6398 | 0.7188 | 0.1485 | 0.1262 | 0.0224 |
| 0.6003 | 0.9 | 400 | -1.8108 | -1.5449 | -121.3720 | -144.0564 | 0.6355 | 0.7326 | 0.1376 | 0.1371 | 0.0005 |
Framework versions
- Transformers 4.39.0.dev0
- Pytorch 2.1.2
- Datasets 2.14.6
- Tokenizers 0.15.2