@inproceedings{liu-etal-2026-mctq,
title = "{MCTQ}: Modality-Coordinated Transformation Quantization for Multimodal Large Language Models",
author = "Liu, Yingen and
Wu, Lanfeng and
Wu, Fan and
Liyan, Wu and
Li, Ruihui and
Tang, Zhuo and
Chng, Eng Siong and
Li, Kenli",
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.1997/",
pages = "45477--45494",
ISBN = "979-8-89176-459-0",
abstract = "Post-training quantization (PTQ) for multimodal large language models (MLLMs) faces a distinct optimization challenge: text and vision tokens can induce conflicting gradients when learning a shared channel-wise transformation. We identify this issue as modality gradient conflict, characterized by opposing update directions and severe magnitude imbalance between modality-specific gradients, and propose Modality-Coordinated Transformation Quantization (MCTQ) to address it. MCTQ combines whitening-based preconditioning, modality-decomposed reconstruction, and modality-coordinated optimization to learn a single shared transformation that serves both modalities without introducing modality-specific parameters or inference overhead. Across multiple MLLMs, benchmarks, and precision settings, MCTQ consistently outperforms existing channel-wise PTQ methods, with the largest gains under aggressive low-bit quantization. These results show that coordinating modality-specific gradients is critical for learning effective shared transformations in multimodal PTQ. Code and reproduction materials are available at https://github.com/jtsdw/UniMQ."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="liu-etal-2026-mctq">
<titleInfo>
<title>MCTQ: Modality-Coordinated Transformation Quantization for Multimodal Large Language Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yingen</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Lanfeng</namePart>
<namePart type="family">Wu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Fan</namePart>
<namePart type="family">Wu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Wu</namePart>
<namePart type="family">Liyan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ruihui</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhuo</namePart>
<namePart type="family">Tang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Eng</namePart>
<namePart type="given">Siong</namePart>
<namePart type="family">Chng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kenli</namePart>
<namePart type="family">Li</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>Post-training quantization (PTQ) for multimodal large language models (MLLMs) faces a distinct optimization challenge: text and vision tokens can induce conflicting gradients when learning a shared channel-wise transformation. We identify this issue as modality gradient conflict, characterized by opposing update directions and severe magnitude imbalance between modality-specific gradients, and propose Modality-Coordinated Transformation Quantization (MCTQ) to address it. MCTQ combines whitening-based preconditioning, modality-decomposed reconstruction, and modality-coordinated optimization to learn a single shared transformation that serves both modalities without introducing modality-specific parameters or inference overhead. Across multiple MLLMs, benchmarks, and precision settings, MCTQ consistently outperforms existing channel-wise PTQ methods, with the largest gains under aggressive low-bit quantization. These results show that coordinating modality-specific gradients is critical for learning effective shared transformations in multimodal PTQ. Code and reproduction materials are available at https://github.com/jtsdw/UniMQ.</abstract>
<identifier type="citekey">liu-etal-2026-mctq</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.1997/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>45477</start>
<end>45494</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T MCTQ: Modality-Coordinated Transformation Quantization for Multimodal Large Language Models
%A Liu, Yingen
%A Wu, Lanfeng
%A Wu, Fan
%A Liyan, Wu
%A Li, Ruihui
%A Tang, Zhuo
%A Chng, Eng Siong
%A Li, Kenli
%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 liu-etal-2026-mctq
%X Post-training quantization (PTQ) for multimodal large language models (MLLMs) faces a distinct optimization challenge: text and vision tokens can induce conflicting gradients when learning a shared channel-wise transformation. We identify this issue as modality gradient conflict, characterized by opposing update directions and severe magnitude imbalance between modality-specific gradients, and propose Modality-Coordinated Transformation Quantization (MCTQ) to address it. MCTQ combines whitening-based preconditioning, modality-decomposed reconstruction, and modality-coordinated optimization to learn a single shared transformation that serves both modalities without introducing modality-specific parameters or inference overhead. Across multiple MLLMs, benchmarks, and precision settings, MCTQ consistently outperforms existing channel-wise PTQ methods, with the largest gains under aggressive low-bit quantization. These results show that coordinating modality-specific gradients is critical for learning effective shared transformations in multimodal PTQ. Code and reproduction materials are available at https://github.com/jtsdw/UniMQ.
%U https://aclanthology.org/2026.emnlp-main.1997/
%P 45477-45494
Markdown (Informal)
[MCTQ: Modality-Coordinated Transformation Quantization for Multimodal Large Language Models](https://aclanthology.org/2026.emnlp-main.1997/) (Liu et al., EMNLP 2026)
ACL
- Yingen Liu, Lanfeng Wu, Fan Wu, Wu Liyan, Ruihui Li, Zhuo Tang, Eng Siong Chng, and Kenli Li. 2026. MCTQ: Modality-Coordinated Transformation Quantization for Multimodal Large Language Models. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 45477–45494, Budapest, Hungary. Association for Computational Linguistics.