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: 0.5952
• Rewards/chosen: 0.1959
• Rewards/rejected: -0.1959
• Rewards/accuracies: 0.8125
• Rewards/margins: 0.3918
• Logps/rejected: -225.0688
• Logps/chosen: -260.4842
• Logits/rejected: -2.6980
• Logits/chosen: -2.6353

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
0.6829	0.01	10	0.6352	0.0696	-0.0459	0.5625	0.1156	-223.5694	-261.7468	-2.6848	-2.6255
0.7018	0.01	20	0.6071	0.1534	-0.1323	0.5625	0.2857	-224.4329	-260.9089	-2.6932	-2.6321
0.6391	0.01	30	0.5922	0.2186	-0.1600	0.8125	0.3787	-224.7106	-260.2570	-2.6988	-2.6359
0.5993	0.02	40	0.5937	0.1966	-0.1896	0.8125	0.3861	-225.0059	-260.4774	-2.6987	-2.6355
0.6781	0.03	50	0.5952	0.1959	-0.1959	0.8125	0.3918	-225.0688	-260.4842	-2.6980	-2.6353

Framework versions

• Transformers 4.35.2
• Pytorch 2.0.1+cu117
• Datasets 2.16.1
• Tokenizers 0.15.0