selfbiorag-7b-dpo-full-wo-live_qa-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: 0.6503
• Rewards/chosen: 0.1533
• Rewards/rejected: 0.0496
• Rewards/accuracies: 0.7273
• Rewards/margins: 0.1037
• Logps/rejected: -152.4542
• Logps/chosen: -129.3861
• Logits/rejected: -1.6930
• Logits/chosen: -1.9168

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
0.6586	0.22	100	0.6694	0.1304	0.0785	0.6932	0.0520	-149.5688	-131.6699	-1.5849	-1.7960
0.6342	0.44	200	0.6581	0.1743	0.0934	0.7273	0.0808	-148.0715	-127.2864	-1.5703	-1.7959
0.5967	0.66	300	0.6527	0.1658	0.0685	0.7159	0.0973	-150.5620	-128.1308	-1.6454	-1.8741
0.5979	0.88	400	0.6502	0.1544	0.0500	0.7273	0.1043	-152.4127	-129.2778	-1.6898	-1.9159

Framework versions

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