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: 0.6173
• Rewards/chosen: 12.3662
• Rewards/rejected: 8.3268
• Rewards/accuracies: 0.875
• Rewards/margins: 4.0394
• Logps/rejected: -284.7693
• Logps/chosen: -270.0549
• Logits/rejected: -2.4409
• Logits/chosen: -2.6955

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
0.6874	0.01	10	0.6396	0.0822	-0.0086	0.5625	0.0908	-368.1237	-392.8953	-2.2951	-2.4730
0.7988	0.01	20	0.5694	1.1019	0.5787	0.875	0.5232	-362.2504	-382.6982	-2.2767	-2.5084
0.6368	0.01	30	0.5572	11.3452	7.2855	0.875	4.0597	-295.1821	-280.2652	-2.4358	-2.6872
1.6793	0.02	40	0.5216	11.9759	7.8540	0.9375	4.1220	-289.4976	-273.9581	-2.4389	-2.6947
4.9001	0.03	50	0.6173	12.3662	8.3268	0.875	4.0394	-284.7693	-270.0549	-2.4409	-2.6955

Framework versions

• PEFT 0.7.1
• Transformers 4.36.2
• Pytorch 2.1.0+cu121
• Datasets 2.15.0
• Tokenizers 0.15.0