metadata
license: apache-2.0
base_model: Minbyul/biomistral-7b-wo-kqa_golden-iter-dpo-step2
tags:
- alignment-handbook
- trl
- dpo
- generated_from_trainer
- trl
- dpo
- generated_from_trainer
datasets:
- HuggingFaceH4/ultrafeedback_binarized
model-index:
- name: biomistral-7b-wo-kqa_golden-iter-dpo-step3
results: []
biomistral-7b-wo-kqa_golden-iter-dpo-step3
This model is a fine-tuned version of Minbyul/biomistral-7b-wo-kqa_golden-iter-dpo-step2 on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:
- Loss: 0.6829
- Rewards/chosen: -0.0502
- Rewards/rejected: -0.0800
- Rewards/accuracies: 0.6300
- Rewards/margins: 0.0298
- Logps/rejected: -60.1185
- Logps/chosen: -40.1264
- Logits/rejected: -1.5228
- Logits/chosen: -0.8710
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: 1e-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 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.6794 | 0.37 | 100 | -0.8266 | -1.4757 | -35.5860 | -53.0765 | 0.6906 | 0.5900 | -0.0048 | 0.0048 | -0.0096 |
| 0.6555 | 0.74 | 200 | -0.8589 | -1.5130 | -39.0432 | -58.7210 | 0.6837 | 0.6400 | -0.0394 | 0.0267 | -0.0661 |
Framework versions
- Transformers 4.39.0.dev0
- Pytorch 2.1.2
- Datasets 2.14.6
- Tokenizers 0.15.2