@inproceedings{huang-etal-2026-cat,
title = "Cat-{M}o{D}: Accelerating Multimodal Alignment via Caption Token Guided Asymmetric Mixture-of-Depths",
author = "Huang, YiJie and
Yang, Xiaocui and
Feng, Shi and
Zhang, Wen and
Song, Kaisong and
Zhang, Yifei and
Wang, Daling",
editor = "Liakata, Maria and
Moreira, Viviane P. and
Zhang, Jiajun and
Jurgens, David",
booktitle = "Proceedings of the 64th Annual Meeting of the {A}ssociation for {C}omputational {L}inguistics (Volume 1: Long Papers)",
month = jul,
year = "2026",
address = "San Diego, California, United States",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.acl-long.1213/",
pages = "26354--26376",
ISBN = "979-8-89176-390-6",
abstract = "Efficiently aligning visual features with Large Language Models (LLMs) remains a critical bottleneck in Multimodal LLMs. Existing query-based alignment modules (e.g., Q-Former) rely on randomly initialized queries, resulting in an inefficient cold start exploration process. Furthermore, they enforce uniform cross-attention across all layers, leading to computational redundancy. Our empirical analysis reveals that query tokens initialized with language priors can rapidly capture global semantics, leading to early representation convergence after only a few layers. In this paper, we propose **Cat-MoD**, a **Ca**ption **t**oken Guided Asymmetric **M**ixture-**o**f-**D**epths framework. It incorporates a **Hybrid Query Construction** module where Guide Tokens initialized from coarse-grained linguistic priors rapidly anchor global semantic context, and randomly initialized Explorer Tokens remain active to capture fine-grained visual details. Exploiting this early convergence, we introduce an **Asymmetric Mixture-of-Depths** mechanism, where a similarity-aware router dynamically prunes redundant tokens from expensive cross-attention layers while preserving their context in self-attention. Experiments on multiple benchmarks demonstrate that Cat-MoD matches or surpasses baseline performance, while substantially reducing alignment FLOPs by approximately 37{\%} during both training and inference, offering a highly efficient solution for multimodal alignment. Code: https://github.com/JasonOrange0726/Cat-MoD."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="huang-etal-2026-cat">
<titleInfo>
<title>Cat-MoD: Accelerating Multimodal Alignment via Caption Token Guided Asymmetric Mixture-of-Depths</title>
</titleInfo>
<name type="personal">
<namePart type="given">YiJie</namePart>
<namePart type="family">Huang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiaocui</namePart>
<namePart type="family">Yang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shi</namePart>
<namePart type="family">Feng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Wen</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kaisong</namePart>
<namePart type="family">Song</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yifei</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Daling</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Maria</namePart>
<namePart type="family">Liakata</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Viviane</namePart>
<namePart type="given">P</namePart>
<namePart type="family">Moreira</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiajun</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">David</namePart>
<namePart type="family">Jurgens</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">San Diego, California, United States</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-390-6</identifier>
</relatedItem>
<abstract>Efficiently aligning visual features with Large Language Models (LLMs) remains a critical bottleneck in Multimodal LLMs. Existing query-based alignment modules (e.g., Q-Former) rely on randomly initialized queries, resulting in an inefficient cold start exploration process. Furthermore, they enforce uniform cross-attention across all layers, leading to computational redundancy. Our empirical analysis reveals that query tokens initialized with language priors can rapidly capture global semantics, leading to early representation convergence after only a few layers. In this paper, we propose **Cat-MoD**, a **Ca**ption **t**oken Guided Asymmetric **M**ixture-**o**f-**D**epths framework. It incorporates a **Hybrid Query Construction** module where Guide Tokens initialized from coarse-grained linguistic priors rapidly anchor global semantic context, and randomly initialized Explorer Tokens remain active to capture fine-grained visual details. Exploiting this early convergence, we introduce an **Asymmetric Mixture-of-Depths** mechanism, where a similarity-aware router dynamically prunes redundant tokens from expensive cross-attention layers while preserving their context in self-attention. Experiments on multiple benchmarks demonstrate that Cat-MoD matches or surpasses baseline performance, while substantially reducing alignment FLOPs by approximately 37% during both training and inference, offering a highly efficient solution for multimodal alignment. Code: https://github.com/JasonOrange0726/Cat-MoD.</abstract>
<identifier type="citekey">huang-etal-2026-cat</identifier>
<location>
<url>https://aclanthology.org/2026.acl-long.1213/</url>
</location>
<part>
<date>2026-07</date>
<extent unit="page">
<start>26354</start>
<end>26376</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Cat-MoD: Accelerating Multimodal Alignment via Caption Token Guided Asymmetric Mixture-of-Depths
%A Huang, YiJie
%A Yang, Xiaocui
%A Feng, Shi
%A Zhang, Wen
%A Song, Kaisong
%A Zhang, Yifei
%A Wang, Daling
%Y Liakata, Maria
%Y Moreira, Viviane P.
%Y Zhang, Jiajun
%Y Jurgens, David
%S Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)
%D 2026
%8 July
%I Association for Computational Linguistics
%C San Diego, California, United States
%@ 979-8-89176-390-6
%F huang-etal-2026-cat
%X Efficiently aligning visual features with Large Language Models (LLMs) remains a critical bottleneck in Multimodal LLMs. Existing query-based alignment modules (e.g., Q-Former) rely on randomly initialized queries, resulting in an inefficient cold start exploration process. Furthermore, they enforce uniform cross-attention across all layers, leading to computational redundancy. Our empirical analysis reveals that query tokens initialized with language priors can rapidly capture global semantics, leading to early representation convergence after only a few layers. In this paper, we propose **Cat-MoD**, a **Ca**ption **t**oken Guided Asymmetric **M**ixture-**o**f-**D**epths framework. It incorporates a **Hybrid Query Construction** module where Guide Tokens initialized from coarse-grained linguistic priors rapidly anchor global semantic context, and randomly initialized Explorer Tokens remain active to capture fine-grained visual details. Exploiting this early convergence, we introduce an **Asymmetric Mixture-of-Depths** mechanism, where a similarity-aware router dynamically prunes redundant tokens from expensive cross-attention layers while preserving their context in self-attention. Experiments on multiple benchmarks demonstrate that Cat-MoD matches or surpasses baseline performance, while substantially reducing alignment FLOPs by approximately 37% during both training and inference, offering a highly efficient solution for multimodal alignment. Code: https://github.com/JasonOrange0726/Cat-MoD.
%U https://aclanthology.org/2026.acl-long.1213/
%P 26354-26376
Markdown (Informal)
[Cat-MoD: Accelerating Multimodal Alignment via Caption Token Guided Asymmetric Mixture-of-Depths](https://aclanthology.org/2026.acl-long.1213/) (Huang et al., ACL 2026)
ACL
- YiJie Huang, Xiaocui Yang, Shi Feng, Wen Zhang, Kaisong Song, Yifei Zhang, and Daling Wang. 2026. Cat-MoD: Accelerating Multimodal Alignment via Caption Token Guided Asymmetric Mixture-of-Depths. In Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 26354–26376, San Diego, California, United States. Association for Computational Linguistics.