@inproceedings{li-etal-2026-plan,
title = "Plan, Verify and Fill: A Structured Parallel Decoding Approach for Diffusion Language Models",
author = "Li, Miao and
Jiang, Hanyang and
Cheng, Sikai and
Fu, Hengyu and
Cai, Yuhang and
Huang, Baihe and
Chen, Zhongzhu and
Ye, Tinghan and
Chen, Xuanzhou and
Van Hentenryck, Pascal",
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.1119/",
doi = "10.18653/2026.findings-emnlp.1119",
pages = "23694--23717",
ISBN = "979-8-89176-462-0",
abstract = "Diffusion Language Models (DLMs) enable parallel text generation by denoising multiple masked positions in a single pass, but confidence-threshold decoding often commits only the safest predictions, leaving informative uncertainty underused. We propose Plan-Verify-Fill (PVF), a training-free draft-and-verification decoding strategy that preserves the high-confidence decoding backbone while selectively testing lower-confidence candidates with high expected verified utility. PVF uses a calibrated planning-token vocabulary to allocate its limited verification budget, accepts proposed drafts only when they remain consistent with the backbone trajectory, and otherwise defaults to conservative confidence-gated unmasking. Across mathematical reasoning, knowledge, and code-generation benchmarks on LLaDA-8B-Instruct and Dream-7B-Instruct, PVF maintains comparable accuracy while delivering up to $2.3\times$ higher throughput than baseline methods."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="li-etal-2026-plan">
<titleInfo>
<title>Plan, Verify and Fill: A Structured Parallel Decoding Approach for Diffusion Language Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Miao</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hanyang</namePart>
<namePart type="family">Jiang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sikai</namePart>
<namePart type="family">Cheng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hengyu</namePart>
<namePart type="family">Fu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yuhang</namePart>
<namePart type="family">Cai</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Baihe</namePart>
<namePart type="family">Huang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhongzhu</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tinghan</namePart>
<namePart type="family">Ye</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xuanzhou</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Pascal</namePart>
<namePart type="family">Van Hentenryck</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>Diffusion Language Models (DLMs) enable parallel text generation by denoising multiple masked positions in a single pass, but confidence-threshold decoding often commits only the safest predictions, leaving informative uncertainty underused. We propose Plan-Verify-Fill (PVF), a training-free draft-and-verification decoding strategy that preserves the high-confidence decoding backbone while selectively testing lower-confidence candidates with high expected verified utility. PVF uses a calibrated planning-token vocabulary to allocate its limited verification budget, accepts proposed drafts only when they remain consistent with the backbone trajectory, and otherwise defaults to conservative confidence-gated unmasking. Across mathematical reasoning, knowledge, and code-generation benchmarks on LLaDA-8B-Instruct and Dream-7B-Instruct, PVF maintains comparable accuracy while delivering up to 2.3\times higher throughput than baseline methods.</abstract>
<identifier type="citekey">li-etal-2026-plan</identifier>
<identifier type="doi">10.18653/2026.findings-emnlp.1119</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1119/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>23694</start>
<end>23717</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Plan, Verify and Fill: A Structured Parallel Decoding Approach for Diffusion Language Models
%A Li, Miao
%A Jiang, Hanyang
%A Cheng, Sikai
%A Fu, Hengyu
%A Cai, Yuhang
%A Huang, Baihe
%A Chen, Zhongzhu
%A Ye, Tinghan
%A Chen, Xuanzhou
%A Van Hentenryck, Pascal
%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 li-etal-2026-plan
%X Diffusion Language Models (DLMs) enable parallel text generation by denoising multiple masked positions in a single pass, but confidence-threshold decoding often commits only the safest predictions, leaving informative uncertainty underused. We propose Plan-Verify-Fill (PVF), a training-free draft-and-verification decoding strategy that preserves the high-confidence decoding backbone while selectively testing lower-confidence candidates with high expected verified utility. PVF uses a calibrated planning-token vocabulary to allocate its limited verification budget, accepts proposed drafts only when they remain consistent with the backbone trajectory, and otherwise defaults to conservative confidence-gated unmasking. Across mathematical reasoning, knowledge, and code-generation benchmarks on LLaDA-8B-Instruct and Dream-7B-Instruct, PVF maintains comparable accuracy while delivering up to 2.3\times higher throughput than baseline methods.
%R 10.18653/2026.findings-emnlp.1119
%U https://aclanthology.org/2026.findings-emnlp.1119/
%U https://doi.org/10.18653/2026.findings-emnlp.1119
%P 23694-23717
Markdown (Informal)
[Plan, Verify and Fill: A Structured Parallel Decoding Approach for Diffusion Language Models](https://aclanthology.org/2026.findings-emnlp.1119/) (Li et al., Findings 2026)
ACL
- Miao Li, Hanyang Jiang, Sikai Cheng, Hengyu Fu, Yuhang Cai, Baihe Huang, Zhongzhu Chen, Tinghan Ye, Xuanzhou Chen, and Pascal Van Hentenryck. 2026. Plan, Verify and Fill: A Structured Parallel Decoding Approach for Diffusion Language Models. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 23694–23717, Budapest, Hungary. Association for Computational Linguistics.