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: 0.6282
• Rewards/chosen: 0.0648
• Rewards/rejected: -0.1682
• Rewards/accuracies: 0.5625
• Rewards/margins: 0.2330
• Logps/rejected: -449.7376
• Logps/chosen: -595.4427
• Logits/rejected: -1.8275
• Logits/chosen: -1.7465

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
0.6839	0.01	10	0.6418	0.1941	0.0444	0.5625	0.1498	-447.6118	-594.1498	-1.8467	-1.7399
0.7038	0.01	20	0.6286	0.2428	0.0265	0.5625	0.2162	-447.7904	-593.6635	-1.8468	-1.7509
0.6388	0.01	30	0.6218	0.2294	-0.0446	0.625	0.2741	-448.5019	-593.7968	-1.8401	-1.7519
0.648	0.02	40	0.6218	0.1284	-0.1299	0.625	0.2582	-449.3542	-594.8075	-1.8332	-1.7487
0.673	0.03	50	0.6282	0.0648	-0.1682	0.5625	0.2330	-449.7376	-595.4427	-1.8275	-1.7465

Framework versions

• PEFT 0.8.2
• Transformers 4.37.0
• Pytorch 2.0.1+cu117
• Datasets 2.17.1
• Tokenizers 0.15.1