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: 0.6930
• Rewards/chosen: 0.0033
• Rewards/rejected: 0.0030
• Rewards/accuracies: 0.4000
• Rewards/margins: 0.0003
• Logps/rejected: -82.3736
• Logps/chosen: -154.2426
• Logits/rejected: -2.5763
• Logits/chosen: -2.5915

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: 1
• eval_batch_size: 2
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• gradient_accumulation_steps: 16
• total_train_batch_size: 32
• total_eval_batch_size: 4
• 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
0.6931	0.96	3	0.6925	-0.0015	-0.0013	0.4200	-0.0002	-82.4166	-154.2906	-2.5763	-2.5917
0.6931	1.92	6	0.6938	0.0059	0.0041	0.4000	0.0018	-82.3626	-154.2166	-2.5766	-2.5914
0.6931	2.88	9	0.6930	0.0033	0.0030	0.4000	0.0003	-82.3736	-154.2426	-2.5763	-2.5915

Framework versions

• Transformers 4.35.2
• Pytorch 2.1.1+cu118
• Datasets 2.14.4
• Tokenizers 0.15.0