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: 0.7106
• Rewards/chosen: -0.0309
• Rewards/rejected: -0.0145
• Rewards/accuracies: 0.4375
• Rewards/margins: -0.0164
• Logps/rejected: -200.0436
• Logps/chosen: -156.7882
• Logits/rejected: -2.6163
• Logits/chosen: -2.6691

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
0.684	0.01	10	0.7086	-0.0352	-0.0042	0.4375	-0.0310	-199.9411	-156.8310	-2.6026	-2.6661
0.71	0.01	20	0.7081	-0.0095	0.0133	0.125	-0.0228	-199.7660	-156.5740	-2.6117	-2.6719
0.6479	0.01	30	0.7104	-0.0099	0.0071	0.375	-0.0171	-199.8277	-156.5786	-2.6162	-2.6706
0.6523	0.02	40	0.7093	-0.0276	-0.0128	0.4375	-0.0148	-200.0267	-156.7554	-2.6165	-2.6692
0.6697	0.03	50	0.7106	-0.0309	-0.0145	0.4375	-0.0164	-200.0436	-156.7882	-2.6163	-2.6691

Framework versions

• Transformers 4.35.2
• Pytorch 2.0.1+cu117
• Datasets 2.16.1
• Tokenizers 0.15.1