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: 0.6183
• Rewards/chosen: 0.0311
• Rewards/rejected: -0.0695
• Rewards/accuracies: 0.625
• Rewards/margins: 0.1006
• Logps/rejected: -143.5769
• Logps/chosen: -125.5347
• Logits/rejected: -2.7201
• Logits/chosen: -2.8454

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
0.6852	0.01	10	0.6765	0.0305	0.0058	0.5625	0.0247	-142.8238	-125.5402	-2.7089	-2.8446
0.7058	0.01	20	0.6604	0.0370	-0.0005	0.5625	0.0375	-142.8867	-125.4757	-2.7121	-2.8454
0.6407	0.01	30	0.6319	0.0537	-0.0265	0.6875	0.0802	-143.1462	-125.3082	-2.7146	-2.8457
0.6445	0.02	40	0.6210	0.0345	-0.0659	0.625	0.1004	-143.5407	-125.5005	-2.7173	-2.8463
0.6847	0.03	50	0.6183	0.0311	-0.0695	0.625	0.1006	-143.5769	-125.5347	-2.7201	-2.8454

Framework versions

• Transformers 4.35.2
• Pytorch 2.0.1+cu117
• Datasets 2.16.1
• Tokenizers 0.15.1