zephyr-7b-dpo-lora

This model is a fine-tuned version of mistralai/Mistral-7B-v0.1 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.5874
• Rewards/chosen: 0.0339
• Rewards/rejected: -0.3006
• Rewards/accuracies: 0.6920
• Rewards/margins: 0.3345
• Logps/rejected: -220.4644
• Logps/chosen: -263.6834
• Logits/rejected: -2.2601
• Logits/chosen: -2.3604

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: 5e-07
• train_batch_size: 2
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 32
• total_train_batch_size: 256
• total_eval_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• 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
0.6259	1.0	242	0.6261	0.0706	-0.0990	0.6700	0.1696	-218.4478	-263.3158	-2.2669	-2.3669
0.5991	2.0	484	0.5946	0.0494	-0.2508	0.6780	0.3002	-219.9661	-263.5286	-2.2620	-2.3623
0.5852	3.0	726	0.5874	0.0339	-0.3006	0.6920	0.3345	-220.4644	-263.6834	-2.2601	-2.3604

Framework versions

• Transformers 4.35.0
• Pytorch 2.1.1+cu121
• Datasets 2.14.6
• Tokenizers 0.14.1