Qwen2.5-7B-Instruct-Ara240

<div align="center"> <div> <a href="https://huggingface.co/nxp/Qwen2.5-7B-Instruct-Ara240/blob/main/LICENSE"><img src="https://img.shields.io/badge/License-Apache%202.0-blue" alt="License Apache-2.0"></a> <a href="https://huggingface.co/Qwen/Qwen2.5-7B-Instruct"><img alt="Static Badge" src="https://img.shields.io/badge/Qwen2.5--7B--Instruct-blue?logo=huggingface&label=Origin&link=https%3A%2F%2Fhuggingface.co%2FQwen%2FQwen2.5-7B-Instruct"></a> <img src="https://img.shields.io/badge/Ara%20SDK-r3.0-green.svg" alt="Ara SDK r3.0"> </div> </div>

Overview

This repository hosts the instruction-tuned Qwen2.5-7B-Instruct model. Compared to earlier versions, Qwen2.5 offers stronger performance in areas such as coding, mathematics, long-form generation, structured data handling, and JSON-based outputs. It also responds more reliably to a wide range of system prompts, useful for scenarios like role-playing agents or chatbot configuration. This specific repository provides the 7.61B-parameter (exact non-rounded count) instruction-tuned model quantized and compiled for Ara240 DNPU acceleration.

Model Description

This is a quantized and compiled version of Qwen/Qwen2.5-7B-Instruct optimized for Ara240 DNPU.

• Base Model: Qwen/Qwen2.5-7B-Instruct
• Original Model Authors: Qwen Team, Alibaba Cloud
• Original License: Apache-2.0
• Modified by: NXP

Modifications

This model is a derivative work with the following changes from the original:

• Quantization: W4A8 - QuaRot (wikitext-2-raw-v1) + EQAT (timdettmers/openassistant-guanaco)
• Compilation: Compiled for Ara240 DNPU
• Format: Converted to DVM format for NPU deployment via the Ara SDK

Original model available at: Qwen/Qwen2.5-7B-Instruct.

Limitations and Biases

This model inherits all limitations from the original Qwen2.5-7B-Instruct model. Additional limitations:

• Hardware-specific: Only runs on hardware with an Ara240 DNPU; not compatible with standard CPU/GPU inference frameworks.
• Language: This deployment is validated for English only (en). The base model supports additional languages, but multilingual use on this quantized build is untested.
• Context length: Maximum supported context length is 4096 tokens.
• Quantization effects: W4A8 quantization may introduce accuracy differences relative to the FP16 baseline.

License

This model is released under the Apache License 2.0, the same license as the original Qwen/Qwen2.5-7B-Instruct model.

Citation

If you use this model, please cite both this work and the original model:

• Qwen Team, "Qwen2.5: A Party of Foundation Models." Sep. 2024. [Online]. Available: https://qwenlm.github.io/blog/qwen2.5/
• A. Yang et al., "Qwen2 Technical Report," arXiv preprint arXiv:2407.10671, 2024.

@misc{qwen2.5,
    title = {Qwen2.5: A Party of Foundation Models},
    url = {https://qwenlm.github.io/blog/qwen2.5/},
    author = {Qwen Team},
    month = {September},
    year = {2024}
}

@article{qwen2,
      title={Qwen2 Technical Report}, 
      author={An Yang and Baosong Yang and Binyuan Hui and Bo Zheng and Bowen Yu and Chang Zhou and Chengpeng Li and Chengyuan Li and Dayiheng Liu and Fei Huang and Guanting Dong and Haoran Wei and Huan Lin and Jialong Tang and Jialin Wang and Jian Yang and Jianhong Tu and Jianwei Zhang and Jianxin Ma and Jin Xu and Jingren Zhou and Jinze Bai and Jinzheng He and Junyang Lin and Kai Dang and Keming Lu and Keqin Chen and Kexin Yang and Mei Li and Mingfeng Xue and Na Ni and Pei Zhang and Peng Wang and Ru Peng and Rui Men and Ruize Gao and Runji Lin and Shijie Wang and Shuai Bai and Sinan Tan and Tianhang Zhu and Tianhao Li and Tianyu Liu and Wenbin Ge and Xiaodong Deng and Xiaohuan Zhou and Xingzhang Ren and Xinyu Zhang and Xipin Wei and Xuancheng Ren and Yang Fan and Yang Yao and Yichang Zhang and Yu Wan and Yunfei Chu and Yuqiong Liu and Zeyu Cui and Zhenru Zhang and Zhihao Fan},
      journal={arXiv preprint arXiv:2407.10671},
      year={2024}
}