@inproceedings{hu-etal-2026-transferring,
title = "Transferring Math Reasoning Abilities from {LLM}s to {MLLM}s via Causal-Aware Model Merging",
author = "Hu, Yijie and
Yang, Guanyu and
Zhou, Zihao and
Dong, Bin and
Huang, Xiaowei and
Huang, Kaizhu and
Wang, Qiufeng",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.1943/",
pages = "41432--41447",
ISBN = "979-8-89176-462-0",
abstract = "To build a math MLLM, current research efforts predominantly rely on elaborate training schemes or high-quality multi-modal math reasoning data. On the other hand, model merging has attracted much attention recently due to its low-cost and tune-free property, which has shown great potential for constructing math MLLM. However, most recent attempts either assume all parameters are task-relevant or adopt a less interpretable parameter selection strategy, thus merging parameters on a coarse granularity, leading to limited or even degrading performance. In this paper, we propose a causal-analysis{--}guided parameter merging framework, which identifies the math reasoning-related parameters in a finer granularity and interpretable manner, then merges the math LLMs with MLLM. Specifically, we first construct targeted arithmetic and logical tasks, generate reference and counterfactual samples, and then apply intervention-based causal attribution to localize parameters that are causally responsible for math reasoning. Once we identify the crucial reasoning parameters in the math LLM that exhibit higher causal effects, we can more precisely align and merge only the corresponding parameters from the math LLM into the base MLLM. The procedure is tuning-free and enhances both math reasoning and general reasoning abilities. Experimental results of three model families on four math reasoning benchmarks and two general reasoning benchmarks validate the effectiveness of the proposed approach."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="hu-etal-2026-transferring">
<titleInfo>
<title>Transferring Math Reasoning Abilities from LLMs to MLLMs via Causal-Aware Model Merging</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yijie</namePart>
<namePart type="family">Hu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Guanyu</namePart>
<namePart type="family">Yang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zihao</namePart>
<namePart type="family">Zhou</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Bin</namePart>
<namePart type="family">Dong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiaowei</namePart>
<namePart type="family">Huang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kaizhu</namePart>
<namePart type="family">Huang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Qiufeng</namePart>
<namePart type="family">Wang</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>Findings of the Association for Computational Linguistics: EMNLP 2026</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">979-8-89176-462-0</identifier>
</relatedItem>
<abstract>To build a math MLLM, current research efforts predominantly rely on elaborate training schemes or high-quality multi-modal math reasoning data. On the other hand, model merging has attracted much attention recently due to its low-cost and tune-free property, which has shown great potential for constructing math MLLM. However, most recent attempts either assume all parameters are task-relevant or adopt a less interpretable parameter selection strategy, thus merging parameters on a coarse granularity, leading to limited or even degrading performance. In this paper, we propose a causal-analysis–guided parameter merging framework, which identifies the math reasoning-related parameters in a finer granularity and interpretable manner, then merges the math LLMs with MLLM. Specifically, we first construct targeted arithmetic and logical tasks, generate reference and counterfactual samples, and then apply intervention-based causal attribution to localize parameters that are causally responsible for math reasoning. Once we identify the crucial reasoning parameters in the math LLM that exhibit higher causal effects, we can more precisely align and merge only the corresponding parameters from the math LLM into the base MLLM. The procedure is tuning-free and enhances both math reasoning and general reasoning abilities. Experimental results of three model families on four math reasoning benchmarks and two general reasoning benchmarks validate the effectiveness of the proposed approach.</abstract>
<identifier type="citekey">hu-etal-2026-transferring</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1943/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>41432</start>
<end>41447</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Transferring Math Reasoning Abilities from LLMs to MLLMs via Causal-Aware Model Merging
%A Hu, Yijie
%A Yang, Guanyu
%A Zhou, Zihao
%A Dong, Bin
%A Huang, Xiaowei
%A Huang, Kaizhu
%A Wang, Qiufeng
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F hu-etal-2026-transferring
%X To build a math MLLM, current research efforts predominantly rely on elaborate training schemes or high-quality multi-modal math reasoning data. On the other hand, model merging has attracted much attention recently due to its low-cost and tune-free property, which has shown great potential for constructing math MLLM. However, most recent attempts either assume all parameters are task-relevant or adopt a less interpretable parameter selection strategy, thus merging parameters on a coarse granularity, leading to limited or even degrading performance. In this paper, we propose a causal-analysis–guided parameter merging framework, which identifies the math reasoning-related parameters in a finer granularity and interpretable manner, then merges the math LLMs with MLLM. Specifically, we first construct targeted arithmetic and logical tasks, generate reference and counterfactual samples, and then apply intervention-based causal attribution to localize parameters that are causally responsible for math reasoning. Once we identify the crucial reasoning parameters in the math LLM that exhibit higher causal effects, we can more precisely align and merge only the corresponding parameters from the math LLM into the base MLLM. The procedure is tuning-free and enhances both math reasoning and general reasoning abilities. Experimental results of three model families on four math reasoning benchmarks and two general reasoning benchmarks validate the effectiveness of the proposed approach.
%U https://aclanthology.org/2026.findings-emnlp.1943/
%P 41432-41447
Markdown (Informal)
[Transferring Math Reasoning Abilities from LLMs to MLLMs via Causal-Aware Model Merging](https://aclanthology.org/2026.findings-emnlp.1943/) (Hu et al., Findings 2026)
ACL