mistral_dpo2

This model is a fine-tuned version of mistralai/Mistral-7B-v0.1 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.6594
• Rewards/chosen: -0.0073
• Rewards/rejected: -0.0880
• Rewards/accuracies: 0.6532
• Rewards/margins: 0.0807
• Logps/rejected: -47.9164
• Logps/chosen: -70.6164
• Logits/rejected: -0.6606
• Logits/chosen: -0.8711

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: 4
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 40
• training_steps: 200

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.6886	0.1251	40	0.6857	0.0161	0.0006	0.6532	0.0154	-47.0302	-70.3825	-0.6637	-0.8738
0.6729	0.2502	80	0.6731	0.0155	-0.0284	0.6587	0.0438	-47.3201	-70.3884	-0.6620	-0.8723
0.6661	0.3753	120	0.6656	0.0214	-0.0413	0.6532	0.0627	-47.4494	-70.3287	-0.6616	-0.8720
0.6495	0.5004	160	0.6611	-0.0187	-0.0942	0.6532	0.0756	-47.9787	-70.7297	-0.6609	-0.8713
0.6525	0.6255	200	0.6594	-0.0073	-0.0880	0.6532	0.0807	-47.9164	-70.6164	-0.6606	-0.8711

Framework versions

• PEFT 0.12.0
• Transformers 4.43.2
• Pytorch 2.4.0+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1