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: 0.5256
• Rewards/chosen: -0.1539
• Rewards/rejected: -0.9025
• Rewards/accuracies: 0.7420
• Rewards/margins: 0.7486
• Logps/rejected: -228.3078
• Logps/chosen: -266.1707
• Logits/rejected: -1.9406
• Logits/chosen: -2.0654

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: 8
• total_train_batch_size: 64
• 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
0.5516	1.0	968	0.5547	-0.1140	-0.6431	0.7160	0.5291	-225.7137	-265.7718	-1.9903	-2.1116
0.5443	2.0	1936	0.5307	-0.1506	-0.8643	0.7420	0.7136	-227.9256	-266.1383	-1.9496	-2.0740
0.5439	3.0	2904	0.5256	-0.1539	-0.9025	0.7420	0.7486	-228.3078	-266.1707	-1.9406	-2.0654

Framework versions

• Transformers 4.35.0
• Pytorch 2.0.1
• Datasets 2.14.6
• Tokenizers 0.14.1