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: 0.6671
• Rewards/chosen: -0.4223
• Rewards/rejected: -1.9854
• Rewards/accuracies: 0.4375
• Rewards/margins: 1.5631
• Logps/rejected: -350.5416
• Logps/chosen: -220.8451
• Logits/rejected: -1.7880
• Logits/chosen: -1.7651

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
0.6968	0.01	10	0.6686	-0.0503	-0.0549	0.5625	0.0046	-331.2371	-217.1252	-1.8320	-1.8255
0.6793	0.01	20	1.7619	0.2100	3.0956	0.3125	-2.8856	-299.7324	-214.5222	-1.9578	-1.9436
0.6789	0.01	30	0.6364	-0.2040	-1.0884	0.4375	0.8843	-341.5715	-218.6622	-1.8060	-1.7894
0.6966	0.02	40	0.6716	-0.2823	-1.4572	0.4375	1.1749	-345.2603	-219.4454	-1.7994	-1.7786
0.8051	0.03	50	0.6671	-0.4223	-1.9854	0.4375	1.5631	-350.5416	-220.8451	-1.7880	-1.7651

Framework versions

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