<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="ren-etal-2026-cforce">
    <titleInfo>
        <title>CForce: Boosting Parallel Decoding for dLLMs via Consistency Forcing</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Yuji</namePart>
        <namePart type="family">Ren</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Chenkai</namePart>
        <namePart type="family">Xu</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Zhuocheng</namePart>
        <namePart type="family">Gong</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Jianguo</namePart>
        <namePart type="family">Li</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Zhijie</namePart>
        <namePart type="family">Deng</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>Diffusion large language models (dLLMs) accelerate language generation by predicting multiple masks in a single forward pass.However, existing dLLMs can suffer from unreliable predictions in early denoising stages under aggressive parallelism strategies, leading to errors that can propagate to later stages. To tackle this issue, we present Consistency Forcing (CForce) for dLLMs, a distillation method to force the mask predictions of early stages to align with those of later stages. CForce trains the model on pre-collected self-rollout trajectories, thereby improving training-inference alignment.We introduce Confidence Adaptive KL Divergence as a distillation objective to conjoin the merits of forward and reverse KL.We further provide a theoretical analysis for the consistency objective to explain why CForce can approximately minimize the prediction error of early stages.Critically, the same formulation applies to both mask-to-token decoding and edit-capable decoding; in the edit-capable case, later token-to-token refinements provide additional supervision for earlier masked-state predictions. Experiments on non-edit and edit-capable LLaDA models show improved speed-quality trade-offs, especially under high-parallelism decoding budgets.</abstract>
    <identifier type="citekey">ren-etal-2026-cforce</identifier>
    <location>
        <url>https://aclanthology.org/2026.emnlp-main.2187/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>49718</start>
            <end>49734</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T CForce: Boosting Parallel Decoding for dLLMs via Consistency Forcing
%A Ren, Yuji
%A Xu, Chenkai
%A Gong, Zhuocheng
%A Li, Jianguo
%A Deng, Zhijie
%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 ren-etal-2026-cforce
%X Diffusion large language models (dLLMs) accelerate language generation by predicting multiple masks in a single forward pass.However, existing dLLMs can suffer from unreliable predictions in early denoising stages under aggressive parallelism strategies, leading to errors that can propagate to later stages. To tackle this issue, we present Consistency Forcing (CForce) for dLLMs, a distillation method to force the mask predictions of early stages to align with those of later stages. CForce trains the model on pre-collected self-rollout trajectories, thereby improving training-inference alignment.We introduce Confidence Adaptive KL Divergence as a distillation objective to conjoin the merits of forward and reverse KL.We further provide a theoretical analysis for the consistency objective to explain why CForce can approximately minimize the prediction error of early stages.Critically, the same formulation applies to both mask-to-token decoding and edit-capable decoding; in the edit-capable case, later token-to-token refinements provide additional supervision for earlier masked-state predictions. Experiments on non-edit and edit-capable LLaDA models show improved speed-quality trade-offs, especially under high-parallelism decoding budgets.
%U https://aclanthology.org/2026.emnlp-main.2187/
%P 49718-49734

Markdown (Informal)

[CForce: Boosting Parallel Decoding for dLLMs via Consistency Forcing](https://aclanthology.org/2026.emnlp-main.2187/) (Ren et al., EMNLP 2026)

• CForce: Boosting Parallel Decoding for dLLMs via Consistency Forcing (Ren et al., EMNLP 2026)

ACL

• Yuji Ren, Chenkai Xu, Zhuocheng Gong, Jianguo Li, and Zhijie Deng. 2026. CForce: Boosting Parallel Decoding for dLLMs via Consistency Forcing. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 49718–49734, Budapest, Hungary. Association for Computational Linguistics.