Mistral-7B-v0.3-dpo-lora_sr_mxm1_lr1e-5_3ep

This model is a fine-tuned version of HachiML/Mistral-7B-v0.3-mxm0.5-lora on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.3042
• Rewards/chosen: -1.2203
• Rewards/rejected: -3.3451
• Rewards/accuracies: 0.8378
• Rewards/margins: 2.1247
• Logps/rejected: -261.0693
• Logps/chosen: -372.8520
• Logits/rejected: 1.1299
• Logits/chosen: 0.4372

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: 1e-05
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.05
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.4588	1.0	103	0.3524	-0.7926	-2.3725	0.7857	1.5799	-251.3439	-368.5752	0.8310	0.2112
0.1934	2.0	206	0.3055	-1.1865	-3.2760	0.8170	2.0896	-260.3791	-372.5136	1.0927	0.4082
0.1254	3.0	309	0.3042	-1.2203	-3.3451	0.8378	2.1247	-261.0693	-372.8520	1.1299	0.4372

Framework versions

• PEFT 0.11.1
• Transformers 4.41.1
• Pytorch 2.3.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1