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: 0.8471
• Rewards/chosen: -0.2589
• Rewards/rejected: -0.1510
• Rewards/accuracies: 0.375
• Rewards/margins: -0.1079
• Logps/rejected: -116.0277
• Logps/chosen: -111.7328
• Logits/rejected: -2.2331
• Logits/chosen: -2.3546

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
0.6755	0.1	10	0.7298	-0.0301	-0.0035	0.375	-0.0266	-114.5520	-109.4439	-2.2395	-2.3722
0.6379	0.2	20	0.7804	-0.1600	-0.1132	0.375	-0.0468	-115.6494	-110.7433	-2.2341	-2.3621
0.7061	0.3	30	0.8180	-0.2242	-0.1463	0.375	-0.0779	-115.9803	-111.3849	-2.2357	-2.3577
0.6503	0.4	40	0.8460	-0.2548	-0.1442	0.375	-0.1106	-115.9595	-111.6915	-2.2330	-2.3554
0.9618	0.5	50	0.8471	-0.2589	-0.1510	0.375	-0.1079	-116.0277	-111.7328	-2.2331	-2.3546

Framework versions

• Transformers 4.35.2
• Pytorch 2.0.1+cu117
• Datasets 2.17.0
• Tokenizers 0.15.2