deita-7b-v1.0-sft

<img src="https://huggingface.co/datasets/hkust-nlp/deita-images/resolve/main/logo-final.png" alt="Deita banner" width="800" style="margin-left:'auto' margin-right:'auto' display:'block'"/>

Model Card for Deita 7B V1.0 SFT

GitHub | Paper

Deita is an open-sourced project designed to facilitate Automatic Data Selection for instruction tuning in Large Language Models (LLMs). Deita 7B V1.0 SFT is a fine-tuned version of Mistral-7B-v0.1 that was trained on 6k automatically selected lightweight, high-quality alignment SFT data: Deita 6K V0.

Model description

• Model type: Model fine tuned on automatically selected lightweight, high-quality alignment SFT data.
• Language(s) (NLP): Primarily English
• Finetuned from model: Mistral-7B-v0.1

Model Sources

• Repository: https://github.com/hkust-nlp/deita
• Model Family: Other models and the dataset are found in the Deita collection.

Performance

Input Format

The model is trained using the vicuna_v1.1 template

A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions. USER: Hello! ASSISTANT: Hi!</s>USER: How are you? ASSISTANT:

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 128
• total_train_batch_size: 512
• total_eval_batch_size: 4
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 6.0

Framework versions

• Transformers 4.34.1
• Pytorch 2.1.0+cu121
• Datasets 2.14.6
• Tokenizers 0.14.1

Citation

If you find the content of this project helpful, please cite our paper as follows:

@misc{liu2023what,
      title={What Makes Good Data for Alignment? A Comprehensive Study of Automatic Data Selection in Instruction Tuning}, 
      author={Wei Liu and Weihao Zeng and Keqing He and Yong Jiang and Junxian He},
      year={2023},
      eprint={2312.15685},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}