@inproceedings{chen-etal-2026-distilling,
title = "Distilling the Essence: Efficient Reasoning Distillation via Sequence Truncation",
author = "Chen, Wei-Rui and
Kothapalli, Vignesh and
Fatahibaarzi, Ata and
Sang, Hejian and
Tang, Shao and
Song, Qingquan and
Wang, Zhipeng and
Abdul-Mageed, Muhammad",
editor = "Liakata, Maria and
Moreira, Viviane P. and
Zhang, Jiajun and
Jurgens, David",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {ACL} 2026",
month = jul,
year = "2026",
address = "San Diego, California, United States",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-acl.587/",
pages = "12092--12122",
ISBN = "979-8-89176-395-1",
abstract = "Distilling the capabilities from a large reasoning model (LRM) to a smaller student model often involves training on substantial amounts of reasoning data. However, knowledge distillation (KD) over lengthy sequences with prompt (P), chain-of-thought (CoT), and answer (A) sections makes the process computationally expensive. In this work, we investigate how the allocation of supervision across different sections (P, CoT, A) affects student performance. Our analysis shows that selective KD over only the CoT tokens can be effective when the prompt and answer information is encompassed by it. Building on this insight, we establish a truncation protocol to quantify computation-quality tradeoffs as a function of sequence length. We observe that beyond a specific length, longer training sequences provide marginal returns for downstream performance but require substantially higher memory and FLOPs. To this end, training on only the first 50{\%} of tokens of every training sequence can retain, on average, {\ensuremath{\approx}}91{\%} of full-sequence performance on math benchmarks while reducing training time, memory usage, and FLOPs by about 50{\%} each. Codes are available at https://github.com/weiruichen01/distilling-the-essence."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="chen-etal-2026-distilling">
<titleInfo>
<title>Distilling the Essence: Efficient Reasoning Distillation via Sequence Truncation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Wei-Rui</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Vignesh</namePart>
<namePart type="family">Kothapalli</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ata</namePart>
<namePart type="family">Fatahibaarzi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hejian</namePart>
<namePart type="family">Sang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shao</namePart>
<namePart type="family">Tang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Qingquan</namePart>
<namePart type="family">Song</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhipeng</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Muhammad</namePart>
<namePart type="family">Abdul-Mageed</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>Findings of the Association for Computational Linguistics: ACL 2026</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">979-8-89176-395-1</identifier>
</relatedItem>
<abstract>Distilling the capabilities from a large reasoning model (LRM) to a smaller student model often involves training on substantial amounts of reasoning data. However, knowledge distillation (KD) over lengthy sequences with prompt (P), chain-of-thought (CoT), and answer (A) sections makes the process computationally expensive. In this work, we investigate how the allocation of supervision across different sections (P, CoT, A) affects student performance. Our analysis shows that selective KD over only the CoT tokens can be effective when the prompt and answer information is encompassed by it. Building on this insight, we establish a truncation protocol to quantify computation-quality tradeoffs as a function of sequence length. We observe that beyond a specific length, longer training sequences provide marginal returns for downstream performance but require substantially higher memory and FLOPs. To this end, training on only the first 50% of tokens of every training sequence can retain, on average, \ensuremath\approx91% of full-sequence performance on math benchmarks while reducing training time, memory usage, and FLOPs by about 50% each. Codes are available at https://github.com/weiruichen01/distilling-the-essence.</abstract>
<identifier type="citekey">chen-etal-2026-distilling</identifier>
<location>
<url>https://aclanthology.org/2026.findings-acl.587/</url>
</location>
<part>
<date>2026-07</date>
<extent unit="page">
<start>12092</start>
<end>12122</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Distilling the Essence: Efficient Reasoning Distillation via Sequence Truncation
%A Chen, Wei-Rui
%A Kothapalli, Vignesh
%A Fatahibaarzi, Ata
%A Sang, Hejian
%A Tang, Shao
%A Song, Qingquan
%A Wang, Zhipeng
%A Abdul-Mageed, Muhammad
%Y Liakata, Maria
%Y Moreira, Viviane P.
%Y Zhang, Jiajun
%Y Jurgens, David
%S Findings of the Association for Computational Linguistics: ACL 2026
%D 2026
%8 July
%I Association for Computational Linguistics
%C San Diego, California, United States
%@ 979-8-89176-395-1
%F chen-etal-2026-distilling
%X Distilling the capabilities from a large reasoning model (LRM) to a smaller student model often involves training on substantial amounts of reasoning data. However, knowledge distillation (KD) over lengthy sequences with prompt (P), chain-of-thought (CoT), and answer (A) sections makes the process computationally expensive. In this work, we investigate how the allocation of supervision across different sections (P, CoT, A) affects student performance. Our analysis shows that selective KD over only the CoT tokens can be effective when the prompt and answer information is encompassed by it. Building on this insight, we establish a truncation protocol to quantify computation-quality tradeoffs as a function of sequence length. We observe that beyond a specific length, longer training sequences provide marginal returns for downstream performance but require substantially higher memory and FLOPs. To this end, training on only the first 50% of tokens of every training sequence can retain, on average, \ensuremath\approx91% of full-sequence performance on math benchmarks while reducing training time, memory usage, and FLOPs by about 50% each. Codes are available at https://github.com/weiruichen01/distilling-the-essence.
%U https://aclanthology.org/2026.findings-acl.587/
%P 12092-12122
Markdown (Informal)
[Distilling the Essence: Efficient Reasoning Distillation via Sequence Truncation](https://aclanthology.org/2026.findings-acl.587/) (Chen et al., Findings 2026)
ACL
- Wei-Rui Chen, Vignesh Kothapalli, Ata Fatahibaarzi, Hejian Sang, Shao Tang, Qingquan Song, Zhipeng Wang, and Muhammad Abdul-Mageed. 2026. Distilling the Essence: Efficient Reasoning Distillation via Sequence Truncation. In Findings of the Association for Computational Linguistics: ACL 2026, pages 12092–12122, San Diego, California, United States. Association for Computational Linguistics.