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: 1.6114
• Rewards/chosen: -2.4430
• Rewards/rejected: -1.9061
• Rewards/accuracies: 0.6875
• Rewards/margins: -0.5369
• Logps/rejected: -480.5268
• Logps/chosen: -519.2752
• Logits/rejected: -2.5308
• Logits/chosen: -2.6586

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
0.7579	0.01	10	0.6551	0.1429	0.0811	0.625	0.0618	-460.6556	-493.4166	-2.4845	-2.6464
0.7842	0.01	20	0.6336	0.2270	0.1016	0.875	0.1254	-460.4503	-492.5753	-2.5075	-2.6588
1.224	0.01	30	1.4870	-2.1416	-1.6612	0.75	-0.4804	-478.0778	-516.2612	-2.5269	-2.6581
9.9606	0.02	40	19.8907	-23.9174	-11.8255	0.5625	-12.0919	-579.7216	-734.0198	-2.2514	-2.3830
2.3399	0.03	50	1.6114	-2.4430	-1.9061	0.6875	-0.5369	-480.5268	-519.2752	-2.5308	-2.6586

Framework versions

• Transformers 4.35.0
• Pytorch 2.0.1+cu117
• Datasets 2.14.6
• Tokenizers 0.14.1