@inproceedings{zhao-etal-2026-self-distillation,
title = "Self-Distillation for Multi-Token Prediction",
author = "Zhao, Guoliang and
Xie, Ruobing and
Wang, An and
Zhang, Kai and
Li, Shuaipeng and
Xie, Huaibing and
Sun, Samm",
editor = "Zhang, Yue and
Xu, Ruifeng and
Purwarianti, Ayu and
Rong, Tong and
Suzuki, Jun",
booktitle = "Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the {A}sia-Pacific Chapter of the {A}ssociation for {C}omputational {L}inguistics (Volume 1: Long Papers)",
month = nov,
year = "2026",
address = "Hengqin, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.aacl-long.189/",
pages = "4225--4243",
ISBN = "979-8-89176-439-2",
abstract = "As Large Language Models (LLMs) continue to scale, inference efficiency has emerged as a critical bottleneck. Multi-Token Prediction (MTP) offers a promising avenue to accelerate LLM inference by predicting multiple future tokens simultaneously. Wherein, DeepSeek MTP has gradually become an indispensable component in industrial LLMs. However, it still faces two challenges: limited accept rates of MTP heads, and difficulties in jointly training multiple MTP heads. To address these challenges, we propose \textbf{MTP-D}, a simple yet effective self-distillation method that substantially boosts MTP head accept rates while maximally preserving main-head performance. We further introduce a \textbf{looped extension strategy} for MTP-D, enabling effective and economical extension to multiple MTP heads and yielding significant additional inference speedup (up to $4.55\times$). Beyond these contributions, we conduct systematic experiments across seven benchmarks to explore and validate key insights into distillation strategies and the scalability potential of DeepSeek MTP. Results demonstrate that MTP-D and the grouped extension strategy consistently enhance MTP head performance and inference efficiency, facilitating the practical usage of MTP in LLMs."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zhao-etal-2026-self-distillation">
<titleInfo>
<title>Self-Distillation for Multi-Token Prediction</title>
</titleInfo>
<name type="personal">
<namePart type="given">Guoliang</namePart>
<namePart type="family">Zhao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ruobing</namePart>
<namePart type="family">Xie</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">An</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kai</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shuaipeng</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Huaibing</namePart>
<namePart type="family">Xie</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Samm</namePart>
<namePart type="family">Sun</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-11</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (Volume 1: Long Papers)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yue</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ruifeng</namePart>
<namePart type="family">Xu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ayu</namePart>
<namePart type="family">Purwarianti</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tong</namePart>
<namePart type="family">Rong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jun</namePart>
<namePart type="family">Suzuki</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Hengqin, China</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-439-2</identifier>
</relatedItem>
<abstract>As Large Language Models (LLMs) continue to scale, inference efficiency has emerged as a critical bottleneck. Multi-Token Prediction (MTP) offers a promising avenue to accelerate LLM inference by predicting multiple future tokens simultaneously. Wherein, DeepSeek MTP has gradually become an indispensable component in industrial LLMs. However, it still faces two challenges: limited accept rates of MTP heads, and difficulties in jointly training multiple MTP heads. To address these challenges, we propose MTP-D, a simple yet effective self-distillation method that substantially boosts MTP head accept rates while maximally preserving main-head performance. We further introduce a looped extension strategy for MTP-D, enabling effective and economical extension to multiple MTP heads and yielding significant additional inference speedup (up to 4.55\times). Beyond these contributions, we conduct systematic experiments across seven benchmarks to explore and validate key insights into distillation strategies and the scalability potential of DeepSeek MTP. Results demonstrate that MTP-D and the grouped extension strategy consistently enhance MTP head performance and inference efficiency, facilitating the practical usage of MTP in LLMs.</abstract>
<identifier type="citekey">zhao-etal-2026-self-distillation</identifier>
<location>
<url>https://aclanthology.org/2026.aacl-long.189/</url>
</location>
<part>
<date>2026-11</date>
<extent unit="page">
<start>4225</start>
<end>4243</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Self-Distillation for Multi-Token Prediction
%A Zhao, Guoliang
%A Xie, Ruobing
%A Wang, An
%A Zhang, Kai
%A Li, Shuaipeng
%A Xie, Huaibing
%A Sun, Samm
%Y Zhang, Yue
%Y Xu, Ruifeng
%Y Purwarianti, Ayu
%Y Rong, Tong
%Y Suzuki, Jun
%S Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (Volume 1: Long Papers)
%D 2026
%8 November
%I Association for Computational Linguistics
%C Hengqin, China
%@ 979-8-89176-439-2
%F zhao-etal-2026-self-distillation
%X As Large Language Models (LLMs) continue to scale, inference efficiency has emerged as a critical bottleneck. Multi-Token Prediction (MTP) offers a promising avenue to accelerate LLM inference by predicting multiple future tokens simultaneously. Wherein, DeepSeek MTP has gradually become an indispensable component in industrial LLMs. However, it still faces two challenges: limited accept rates of MTP heads, and difficulties in jointly training multiple MTP heads. To address these challenges, we propose MTP-D, a simple yet effective self-distillation method that substantially boosts MTP head accept rates while maximally preserving main-head performance. We further introduce a looped extension strategy for MTP-D, enabling effective and economical extension to multiple MTP heads and yielding significant additional inference speedup (up to 4.55\times). Beyond these contributions, we conduct systematic experiments across seven benchmarks to explore and validate key insights into distillation strategies and the scalability potential of DeepSeek MTP. Results demonstrate that MTP-D and the grouped extension strategy consistently enhance MTP head performance and inference efficiency, facilitating the practical usage of MTP in LLMs.
%U https://aclanthology.org/2026.aacl-long.189/
%P 4225-4243
Markdown (Informal)
[Self-Distillation for Multi-Token Prediction](https://aclanthology.org/2026.aacl-long.189/) (Zhao et al., AACL-IJCNLP 2026)
ACL
- Guoliang Zhao, Ruobing Xie, An Wang, Kai Zhang, Shuaipeng Li, Huaibing Xie, and Samm Sun. 2026. Self-Distillation for Multi-Token Prediction. In Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (Volume 1: Long Papers), pages 4225–4243, Hengqin, China. Association for Computational Linguistics.