openhermes-mistral-dpo-gptq

This model is a fine-tuned version of TheBloke/OpenHermes-2-Mistral-7B-GPTQ on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.5172
• Rewards/chosen: 0.2719
• Rewards/rejected: 0.0027
• Rewards/accuracies: 0.4375
• Rewards/margins: 0.2692
• Logps/rejected: -523.3978
• Logps/chosen: -559.6996
• Logits/rejected: -1.8608
• Logits/chosen: -1.7791

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: 0.0002
• train_batch_size: 1
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 2
• training_steps: 50
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.8935	0.01	10	0.6254	0.5705	0.2915	0.9375	0.2790	-520.5098	-556.7138	-1.8760	-1.8462
2.7353	0.01	20	1.4708	0.6063	2.5173	0.25	-1.9110	-498.2514	-556.3558	-1.9855	-2.0419
3.4601	0.01	30	0.4799	5.3762	3.1472	0.6875	2.2289	-491.9521	-508.6570	-1.9445	-1.9689
15.4868	0.02	40	0.4790	1.4084	0.7478	0.6875	0.6606	-515.9464	-548.3342	-1.8667	-1.7981
0.4857	0.03	50	0.5172	0.2719	0.0027	0.4375	0.2692	-523.3978	-559.6996	-1.8608	-1.7791

Framework versions

• Transformers 4.35.2
• Pytorch 2.0.1+cu117
• Datasets 2.14.7
• Tokenizers 0.15.0