selfbiorag-7b-dpo-full-sft-wo-medication_qa

This model is a fine-tuned version of Minbyul/selfbiorag-7b-wo-medication_qa-sft on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

• Loss: 0.2759
• Rewards/chosen: -1.2305
• Rewards/rejected: -7.1000
• Rewards/accuracies: 0.8920
• Rewards/margins: 5.8695
• Logps/rejected: -1442.5582
• Logps/chosen: -679.8936
• Logits/rejected: -0.3285
• Logits/chosen: -0.3524

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-06
• 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
0.2249	0.32	100	-0.1107	-0.0290	-650.2339	-1190.2701	0.3821	0.8551	-0.9339	3.6432	-4.5771
0.1549	0.65	200	-0.3180	-0.3222	-652.9113	-1308.4048	0.2709	0.8977	-0.9607	4.7978	-5.7585
0.0946	0.97	300	-0.3523	-0.3283	-679.6155	-1442.4718	0.2756	0.8920	-1.2277	5.8714	-7.0991

Framework versions

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