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: 0.6697
• Rewards/chosen: 0.0061
• Rewards/rejected: -0.0288
• Rewards/accuracies: 0.625
• Rewards/margins: 0.0350
• Logps/rejected: -127.8806
• Logps/chosen: -193.3892
• Logits/rejected: -2.4394
• Logits/chosen: -2.6044

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: 20
• 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
0.6762	0.01	10	0.6833	-0.0083	-0.0216	0.625	0.0133	-127.8089	-193.5340	-2.4411	-2.6076
0.7039	0.01	20	0.6697	0.0061	-0.0288	0.625	0.0350	-127.8806	-193.3892	-2.4394	-2.6044

Framework versions

• PEFT 0.9.0
• Transformers 4.38.2
• Pytorch 2.2.1+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2