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: 0.1947
• Rewards/chosen: 1.6152
• Rewards/rejected: -0.7739
• Rewards/accuracies: 1.0
• Rewards/margins: 2.3891
• Logps/rejected: -44.7556
• Logps/chosen: -277.0341
• Logits/rejected: -1.2926
• Logits/chosen: -2.0369

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
0.6696	0.01	10	0.5880	0.1428	-0.0848	0.75	0.2276	-37.8643	-291.7574	-1.2055	-2.0304
0.5209	0.02	20	0.3704	0.7380	-0.2765	1.0	1.0145	-39.7818	-285.8060	-1.2770	-2.0614
0.3319	0.03	30	0.2640	1.1988	-0.4424	1.0	1.6412	-41.4403	-281.1980	-1.3008	-2.0673
0.216	0.04	40	0.2150	1.4326	-0.6569	1.0	2.0896	-43.5858	-278.8598	-1.2972	-2.0492
0.1711	0.06	50	0.1947	1.6152	-0.7739	1.0	2.3891	-44.7556	-277.0341	-1.2926	-2.0369

Framework versions

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