mistral-dpo

This model is a fine-tuned version of mistralai/Mistral-7B-v0.1 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.5536
• Rewards/chosen: 0.2363
• Rewards/rejected: -0.3821
• Rewards/accuracies: 0.7095
• Rewards/margins: 0.6183
• Logps/rejected: -278.2731
• Logps/chosen: -292.0563
• Logits/rejected: -2.6610
• Logits/chosen: -2.7000

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: 4
• seed: 100
• 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
• num_epochs: 1.0

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.5093	1.0	625	0.5536	0.2363	-0.3821	0.7095	0.6183	-278.2731	-292.0563	-2.6610	-2.7000

Framework versions

• PEFT 0.7.2.dev0
• Transformers 4.38.0.dev0
• Pytorch 2.1.2+cu121
• Datasets 2.15.0
• Tokenizers 0.15.1