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: 0.7067
• Rewards/chosen: 0.0088
• Rewards/rejected: -0.0947
• Rewards/accuracies: 0.625
• Rewards/margins: 0.1035
• Logps/rejected: -172.7847
• Logps/chosen: -98.3108
• Logits/rejected: -2.0623
• Logits/chosen: -1.9279

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
0.6744	0.005	10	0.6970	-0.0186	-0.0266	0.6875	0.0080	-172.1035	-98.5849	-2.0765	-1.9425
0.7073	0.01	20	0.7152	-0.0388	-0.0448	0.4375	0.0060	-172.2850	-98.7869	-2.0706	-1.9392
0.7287	0.015	30	0.7197	0.0026	-0.0203	0.625	0.0230	-172.0406	-98.3726	-2.0688	-1.9317
0.701	0.02	40	0.7120	0.0131	-0.0600	0.625	0.0731	-172.4374	-98.2679	-2.0641	-1.9302
0.6726	0.025	50	0.7067	0.0088	-0.0947	0.625	0.1035	-172.7847	-98.3108	-2.0623	-1.9279

Framework versions

• PEFT 0.11.1
• Transformers 4.41.1
• Pytorch 2.0.1+cu117
• Datasets 2.19.2
• Tokenizers 0.19.1