openhermes-mistral-dpo-gptq

This model is a fine-tuned version of teknium/OpenHermes-2.5-Mistral-7B on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.5200
• Rewards/chosen: 0.4462
• Rewards/rejected: -0.0495
• Rewards/accuracies: 0.6875
• Rewards/margins: 0.4958
• Logps/rejected: -235.5496
• Logps/chosen: -196.8896
• Logits/rejected: -3.1546
• Logits/chosen: -3.2002

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

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.701	0.01	10	0.6011	0.2373	0.0723	0.375	0.1650	-234.3308	-198.9786	-3.1345	-3.1669
0.6827	0.01	20	0.5701	0.3490	-0.0051	0.6875	0.3541	-235.1054	-197.8619	-3.1513	-3.1894
0.7041	0.01	30	0.5571	0.3683	-0.0432	0.6875	0.4115	-235.4866	-197.6691	-3.1811	-3.2466
0.7273	0.02	40	0.5295	0.4324	-0.0346	0.6875	0.4671	-235.4007	-197.0275	-3.1709	-3.2343
0.6139	0.03	50	0.5200	0.4462	-0.0495	0.6875	0.4958	-235.5496	-196.8896	-3.1546	-3.2002

Framework versions

• Transformers 4.34.1
• Pytorch 2.1.0+cu121
• Datasets 2.14.6
• Tokenizers 0.14.1