Mistral-7B-Instruct-v0.2-ORPO

This model is a fine-tuned version of mistralai/Mistral-7B-Instruct-v0.2 on the dpo_mix_en dataset. It achieves the following results on the evaluation set:

• Loss: 0.8975
• Rewards/chosen: -0.0835
• Rewards/rejected: -0.1074
• Rewards/accuracies: 0.5900
• Rewards/margins: 0.0238
• Logps/rejected: -1.0737
• Logps/chosen: -0.8352
• Logits/rejected: -2.8721
• Logits/chosen: -2.8461
• Sft Loss: 0.8352
• Odds Ratio Loss: 0.6231

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: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 0.1
• num_epochs: 3.0

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen	Sft Loss	Odds Ratio Loss
1.0001	0.8891	500	0.9318	-0.0869	-0.1112	0.5920	0.0243	-1.1123	-0.8690	-2.8936	-2.8713	0.8690	0.6284
0.906	1.7782	1000	0.9039	-0.0841	-0.1081	0.5780	0.0240	-1.0811	-0.8415	-2.8783	-2.8533	0.8415	0.6243
0.9019	2.6673	1500	0.8975	-0.0835	-0.1074	0.5900	0.0238	-1.0737	-0.8352	-2.8721	-2.8461	0.8352	0.6231

Framework versions

• PEFT 0.10.0
• Transformers 4.40.1
• Pytorch 2.3.0
• Datasets 2.19.0
• Tokenizers 0.19.1