mistral-7b-instruct-v0.2-dpo-multilingual-mix-1st-apr-final

This model is a fine-tuned version of mistralai/Mistral-7B-Instruct-v0.2 on the nthakur/multilingual-truthy-v0.1-dpo-pairs, the nthakur/multilingual-orca-dpo-pairs and the nthakur/multilingual-mistral-neg-orca-dpo-pairs datasets. It achieves the following results on the evaluation set:

• Loss: 0.0524
• Rewards/chosen: -0.1806
• Rewards/rejected: -11.0616
• Rewards/accuracies: 0.9918
• Rewards/margins: 10.8811
• Logps/rejected: -1280.7075
• Logps/chosen: -363.8373
• Logits/rejected: -0.2239
• Logits/chosen: -0.5036

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: 16
• eval_batch_size: 16
• seed: 42
• distributed_type: multi-GPU
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 1

Training results

Framework versions

• PEFT 0.7.1
• Transformers 4.39.0.dev0
• Pytorch 2.2.2
• Datasets 2.14.6
• Tokenizers 0.15.2