Sqft Phi 3 Mini 4k 30 Base

🚀 SQFT Base Model: sqft-phi-3-mini-4k-30-base

This is a base model based on SQFT technology, leveraging a specific source model and sparse method to achieve a certain sparsity.

🚀 Quick Start

To get this model, refer to the command in SQFT/run_command/phi-3-mini-4k-instruct/sparse_quantization.sh#11.

✨ Features

• Source Model: microsoft/Phi-3-mini-4k-instruct
• Sparse Method: Wanda
• Sparsity: 30%
• Quantization: No

📚 Documentation

Model Sources

• Repository: https://github.com/IntelLabs/Hardware-Aware-Automated-Machine-Learning/tree/main/SQFT
• Paper:
  • SQFT: Low-cost Model Adaptation in Low-precision Sparse Foundation Models
  • Low-Rank Adapters Meet Neural Architecture Search for LLM Compression

Citation

@inproceedings{munoz-etal-2024-sqft,
    title = "{SQFT}: Low-cost Model Adaptation in Low-precision Sparse Foundation Models",
    author = "Munoz, Juan Pablo  and
      Yuan, Jinjie  and
      Jain, Nilesh",
    editor = "Al-Onaizan, Yaser  and
      Bansal, Mohit  and
      Chen, Yun-Nung",
    booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2024",
    month = nov,
    year = "2024",
    address = "Miami, Florida, USA",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2024.findings-emnlp.749",
    pages = "12817--12832",
}

Acknowledgement

Thanks to the work Wanda (paper, code), which provides a simple but effective pruning approach.

📄 License

This project is licensed under the Apache-2.0 license.