mistral-sft-7b-dpo-qlora

This model is a fine-tuned version of HuggingFaceH4/mistral-7b-sft-beta on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

• Loss: 0.6936
• Rewards/chosen: 0.0005
• Rewards/rejected: 0.0001
• Rewards/accuracies: 0.6875
• Rewards/margins: 0.0003
• Logps/rejected: -122.9776
• Logps/chosen: -86.4464
• Logits/rejected: -3.0453
• Logits/chosen: -2.9824

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-06
• train_batch_size: 4
• eval_batch_size: 8
• seed: 221
• distributed_type: multi-GPU
• num_devices: 2
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• total_eval_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 5

Training results

Framework versions

• PEFT 0.7.1
• Transformers 4.38.2
• Pytorch 2.2.1
• Datasets 2.14.6
• Tokenizers 0.15.2