biomistral-7b-dpo-full-wo-medication_qa-ep3

This model is a fine-tuned version of BioMistral/BioMistral-7B on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

• Loss: 0.6016
• Rewards/chosen: -0.9078
• Rewards/rejected: -1.6854
• Rewards/accuracies: 0.7108
• Rewards/margins: 0.7776
• Logps/rejected: -356.8195
• Logps/chosen: -337.4762
• Logits/rejected: 0.8719
• Logits/chosen: -1.4171

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
0.3643	0.36	100	0.5978	-0.3792	-0.7130	0.7593	0.3338	-259.5780	-284.6122	0.0454	-1.7486
0.2041	0.71	200	0.5963	-0.8767	-1.6290	0.7108	0.7523	-351.1816	-334.3655	0.8360	-1.4335

Framework versions

• Transformers 4.39.0.dev0
• Pytorch 2.1.2
• Datasets 2.14.6
• Tokenizers 0.15.2