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: 0.5940
• Rewards/chosen: 0.0155
• Rewards/rejected: -0.3247
• Rewards/accuracies: 0.875
• Rewards/margins: 0.3402
• Logps/rejected: -329.5930
• Logps/chosen: -281.9707
• Logits/rejected: -2.2879
• Logits/chosen: -2.2211

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
0.6766	0.01	10	0.7047	-0.0328	-0.0733	0.5	0.0405	-327.0790	-282.4540	-2.2913	-2.2181
0.7059	0.01	20	0.6603	-0.0641	-0.2367	0.8125	0.1726	-328.7130	-282.7669	-2.2896	-2.2224
0.6913	0.01	30	0.6036	0.0238	-0.2729	0.875	0.2968	-329.0755	-281.8873	-2.2889	-2.2248
0.7003	0.02	40	0.5921	0.0259	-0.3104	0.875	0.3364	-329.4504	-281.8663	-2.2881	-2.2227
0.6585	0.03	50	0.5940	0.0155	-0.3247	0.875	0.3402	-329.5930	-281.9707	-2.2879	-2.2211

Framework versions

• PEFT 0.10.0
• Transformers 4.38.2
• Pytorch 2.0.1+cu117
• Datasets 2.18.0
• Tokenizers 0.15.2