doplhin-2.1-mistral-7b-orpo-ultrafeedback-binarized-preferences

This model is a fine-tuned version of cognitivecomputations/dolphin-2.1-mistral-7b on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.8506
• Rewards/chosen: -0.0852
• Rewards/rejected: -0.1166
• Rewards/accuracies: 0.6457
• Rewards/margins: 0.0314
• Logps/rejected: -1.1665
• Logps/chosen: -0.8525
• Logits/rejected: -2.6517
• Logits/chosen: -2.7250
• Nll Loss: 0.7896
• Log Odds Ratio: -0.6110
• Log Odds Chosen: 0.4581

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-05
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• 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	Nll Loss	Log Odds Ratio	Log Odds Chosen
0.9101	0.25	700	0.8845	-0.0869	-0.1106	0.6428	0.0237	-1.1059	-0.8694	-2.6631	-2.7431	0.8224	-0.6225	0.3631
0.8554	0.51	1400	0.8609	-0.0877	-0.1233	0.6555	0.0357	-1.2332	-0.8766	-2.6169	-2.6996	0.8007	-0.6040	0.5048
0.9011	0.76	2100	0.8506	-0.0852	-0.1166	0.6457	0.0314	-1.1665	-0.8525	-2.6517	-2.7250	0.7896	-0.6110	0.4581

Framework versions

• PEFT 0.10.1.dev0
• Transformers 4.40.0.dev0
• Pytorch 2.1.2+cu121
• Datasets 2.18.1.dev0
• Tokenizers 0.15.2