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: 0.4545
• Rewards/chosen: -0.0587
• Rewards/rejected: -1.0907
• Rewards/accuracies: 0.875
• Rewards/margins: 1.0320
• Logps/rejected: -312.2487
• Logps/chosen: -273.6681
• Logits/rejected: -1.8614
• Logits/chosen: -1.7936

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
0.6989	0.01	10	0.6566	-0.0830	-0.1482	0.75	0.0652	-302.8232	-273.9107	-1.8738	-1.7954
0.6578	0.01	20	0.5787	0.0468	-0.2201	0.8125	0.2669	-303.5421	-272.6130	-1.8707	-1.7965
0.715	0.01	30	0.5021	0.2256	-0.3134	0.8125	0.5391	-304.4756	-270.8246	-1.8729	-1.8014
0.6847	0.02	40	0.4673	0.2097	-0.6320	0.875	0.8417	-307.6610	-270.9843	-1.8682	-1.7996
0.7869	0.03	50	0.4545	-0.0587	-1.0907	0.875	1.0320	-312.2487	-273.6681	-1.8614	-1.7936

Framework versions

• Transformers 4.35.2
• Pytorch 2.0.1+cu117
• Datasets 2.15.0
• Tokenizers 0.15.0