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: 0.5254
• Rewards/chosen: -0.1284
• Rewards/rejected: -0.9165
• Rewards/accuracies: 0.7400
• Rewards/margins: 0.7881
• Logps/rejected: -229.5277
• Logps/chosen: -266.8034
• Logits/rejected: -2.8774
• Logits/chosen: -2.9668

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: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• gradient_accumulation_steps: 32
• total_train_batch_size: 64
• total_eval_batch_size: 2
• 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
0.545	1.0	968	0.5543	-0.0951	-0.6704	0.7270	0.5752	-227.0664	-266.4709	-2.8955	-2.9826
0.5144	2.0	1936	0.5310	-0.1339	-0.8883	0.7360	0.7544	-229.2458	-266.8581	-2.8813	-2.9701
0.518	3.0	2904	0.5254	-0.1284	-0.9165	0.7400	0.7881	-229.5277	-266.8034	-2.8774	-2.9668

Framework versions

• Transformers 4.36.0.dev0
• Pytorch 2.0.1+cu117
• Datasets 2.14.5
• Tokenizers 0.15.0