一句话：提出元数据增强微缩放格式 M²XFP 及算法-硬件协同设计，大幅降低低比特量化精度损失并提升加速器效率。
现有低比特微缩放（MX）格式如 MXFP4 使用共享的 2 的幂缩放因子，在大语言模型量化中常导致明显精度下降。本文提出 M²XFP：一种元数据增强的微缩放数据格式，通过引入少量灵活元数据和简单编码的在线量化，在广泛模型范围内恢复量化精度，同时保持高比特效率。作者采用算法-硬件协同设计，实现轻量硬件单元并集成到加速器中。在 LLM 基准上，该方法将精度损失较 MXFP4 平均降低 70.63%，较最新 NVFP4 降低 37.30%；相较当前最优加速器，最高获得 1.91 倍加速和 1.75 倍能耗节省。
Existing low-bit Microscaling (MX) formats, such as MXFP4, often suffer from substantial accuracy degradation due to the use of a shared scaling factor with the Power-of-Two format. In this work, we explore strategies that introduce minimal metadata to recover accuracy lost during quantization while maintaining high bit efficiency across a wide range of large language models. We propose a complete algorithm-hardware co-design based on flexible metadata, featuring an online quantization with simple encoding. To support the proposed method efficiently, we implement a lightweight hardware unit and integrate it into the accelerator. Evaluation results demonstrate that our method substantially narrows the accuracy gap, achieving on average a 70.63% reduction in accuracy loss compared to MXFP4 and a 37.30% reduction relative to the latest NVFP4 on LLM benchmarks. Furthermore, our design delivers up to 1.91× speedup and 1.75× energy savings over state-of-the-art accelerators.
DOI 原文 · arXiv · PDF（开放获取） · DBLP
@inproceedings{0005ZZ000H0HWL26,
title = {M
<sup>2</sup>
XFP: A Metadata-Augmented Microscaling Data Format for Efficient Low-bit Quantization},
author = {Weiming Hu and Zihan Zhang and Haoyan Zhang and Chen Zhang and Cong Guo and Yu Feng and Tianchi Hu and Guanglin Li and Guipeng Hu and Junsong Wang and Jingwen Leng},
booktitle = {ASPLOS},
year = {2026},
doi = {10.1145/3779212.3790185}
}