@inproceedings{tan-etal-2026-moaq,
title = "{MOAQ}: A Multi-Scale Outlier-Aware Framework for Extreme Quantization of Large Language Models",
author = "Tan, Jiaxi and
Li, Xudong and
Zhang, Yan and
Ji, Jiayi and
Zheng, Xiawu and
Shen, Yunhang and
Li, Ke and
Cao, Liujuan",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.1434/",
pages = "32762--32781",
ISBN = "979-8-89176-459-0",
abstract = "Low-bit quantization is crucial for deploying Large Language Models (LLMs) on resource-constrained hardware. However, existing Post-Training Quantization (PTQ) methods adopt a monolithic view of outliers, overlooking their global spatial patterns within LLM architectures and weights that are quantization-sensitive but not numerically large. To overcome these challenges, we introduce MOAQ, a novel finetuning-free weight-only PTQ framework based on a new perspective for outlier analysis. MOAQ holistically addresses the outlier problem by three core components: (1) Spatial Calibration (SC) uses a clustering-optimized Hadamard matrix to reorganize disordered outliers into both discrete and clustered patterns. (2) Local Awareness and Suppression (LAS) adopts a dual strategy guided by our outlier metric. It assigns higher bit-widths to clustered outliers via Clustered Outlier Bit Search (COBS), enabling flexible fractional-bit quantization, while further refining quantization parameters for local discrete outliers through Local Outlierness-Aware Calibration (LOAC). (3) Global Feedback and Denoising (GFD) introduces a dynamic correction mechanism informed by our outlier metric to mitigate the progressive cross-layer enrichment of outliers. Extensive experiments show that MOAQ achieves state-of-the-art performance, outperforming both finetuning-free and many finetuning-based methods with substantially lower computational cost."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="tan-etal-2026-moaq">
<titleInfo>
<title>MOAQ: A Multi-Scale Outlier-Aware Framework for Extreme Quantization of Large Language Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Jiaxi</namePart>
<namePart type="family">Tan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xudong</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yan</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiayi</namePart>
<namePart type="family">Ji</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiawu</namePart>
<namePart type="family">Zheng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yunhang</namePart>
<namePart type="family">Shen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ke</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Liujuan</namePart>
<namePart type="family">Cao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Low-bit quantization is crucial for deploying Large Language Models (LLMs) on resource-constrained hardware. However, existing Post-Training Quantization (PTQ) methods adopt a monolithic view of outliers, overlooking their global spatial patterns within LLM architectures and weights that are quantization-sensitive but not numerically large. To overcome these challenges, we introduce MOAQ, a novel finetuning-free weight-only PTQ framework based on a new perspective for outlier analysis. MOAQ holistically addresses the outlier problem by three core components: (1) Spatial Calibration (SC) uses a clustering-optimized Hadamard matrix to reorganize disordered outliers into both discrete and clustered patterns. (2) Local Awareness and Suppression (LAS) adopts a dual strategy guided by our outlier metric. It assigns higher bit-widths to clustered outliers via Clustered Outlier Bit Search (COBS), enabling flexible fractional-bit quantization, while further refining quantization parameters for local discrete outliers through Local Outlierness-Aware Calibration (LOAC). (3) Global Feedback and Denoising (GFD) introduces a dynamic correction mechanism informed by our outlier metric to mitigate the progressive cross-layer enrichment of outliers. Extensive experiments show that MOAQ achieves state-of-the-art performance, outperforming both finetuning-free and many finetuning-based methods with substantially lower computational cost.</abstract>
<identifier type="citekey">tan-etal-2026-moaq</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.1434/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>32762</start>
<end>32781</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T MOAQ: A Multi-Scale Outlier-Aware Framework for Extreme Quantization of Large Language Models
%A Tan, Jiaxi
%A Li, Xudong
%A Zhang, Yan
%A Ji, Jiayi
%A Zheng, Xiawu
%A Shen, Yunhang
%A Li, Ke
%A Cao, Liujuan
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F tan-etal-2026-moaq
%X Low-bit quantization is crucial for deploying Large Language Models (LLMs) on resource-constrained hardware. However, existing Post-Training Quantization (PTQ) methods adopt a monolithic view of outliers, overlooking their global spatial patterns within LLM architectures and weights that are quantization-sensitive but not numerically large. To overcome these challenges, we introduce MOAQ, a novel finetuning-free weight-only PTQ framework based on a new perspective for outlier analysis. MOAQ holistically addresses the outlier problem by three core components: (1) Spatial Calibration (SC) uses a clustering-optimized Hadamard matrix to reorganize disordered outliers into both discrete and clustered patterns. (2) Local Awareness and Suppression (LAS) adopts a dual strategy guided by our outlier metric. It assigns higher bit-widths to clustered outliers via Clustered Outlier Bit Search (COBS), enabling flexible fractional-bit quantization, while further refining quantization parameters for local discrete outliers through Local Outlierness-Aware Calibration (LOAC). (3) Global Feedback and Denoising (GFD) introduces a dynamic correction mechanism informed by our outlier metric to mitigate the progressive cross-layer enrichment of outliers. Extensive experiments show that MOAQ achieves state-of-the-art performance, outperforming both finetuning-free and many finetuning-based methods with substantially lower computational cost.
%U https://aclanthology.org/2026.emnlp-main.1434/
%P 32762-32781
Markdown (Informal)
[MOAQ: A Multi-Scale Outlier-Aware Framework for Extreme Quantization of Large Language Models](https://aclanthology.org/2026.emnlp-main.1434/) (Tan et al., EMNLP 2026)
ACL
- Jiaxi Tan, Xudong Li, Yan Zhang, Jiayi Ji, Xiawu Zheng, Yunhang Shen, Ke Li, and Liujuan Cao. 2026. MOAQ: A Multi-Scale Outlier-Aware Framework for Extreme Quantization of Large Language Models. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 32762–32781, Budapest, Hungary. Association for Computational Linguistics.