SQFT Base Model: sqft-mistral-7b-v0.3-50-base-gptq

• Source Model: mistralai/Mistral-7B-v0.3
• Sparse Method: Wanda
• Sparsity: 50%
• Quantization: GPTQ-INT4

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

How to get this model

Refer to the commands in SQFT/run_command/mistral-7b-v0.3/sparse_quantization.sh.

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 sparse algorithm Wanda and the quantization method GPTQ.

License

Apache-2.0