@inproceedings{wang-etal-2026-ttt,
title = "{TTT}-Sparse: Test-Time Training of Sparse Attention for Long-Context Understanding",
author = "Wang, Xuehao and
Deng, Huayu and
Zhong, Linhao and
Wei, Yanbin and
Zhao, Hengyu and
Liu, Mingyu and
Chen, Hao and
Shen, Chunhua",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.1575/",
pages = "35866--35881",
ISBN = "979-8-89176-459-0",
abstract = "Pretrained LLMs excel across diverse tasks, yet standard attention remains a major bottleneck for long-context understanding. Recent test-time training (TTT) methods alleviate this by encoding context into internal model parameters, but they typically require specialized architectures, meta-learned initialization, or continual post-training, preventing plug-and-play use on off-the-shelf LLMs. We propose test-time training sparse attention (TTT-Sparse), a plug-and-play alternative that performs test-time training over external memory while keeping the internal parameters frozen. For each input, TTT-Sparse constructs contextual chunk key states from the KV cache and learns a lightweight query adapter to induce a task-specific sparse attention pattern, which is used during generation to attend only to the most relevant context chunks and suppress interference from irrelevant context. Extensive experiments on long-context benchmarks show that TTT-Sparse improves long-context understanding while reducing generation-time attention cost."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="wang-etal-2026-ttt">
<titleInfo>
<title>TTT-Sparse: Test-Time Training of Sparse Attention for Long-Context Understanding</title>
</titleInfo>
<name type="personal">
<namePart type="given">Xuehao</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Huayu</namePart>
<namePart type="family">Deng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Linhao</namePart>
<namePart type="family">Zhong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yanbin</namePart>
<namePart type="family">Wei</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hengyu</namePart>
<namePart type="family">Zhao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mingyu</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hao</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Chunhua</namePart>
<namePart type="family">Shen</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>Pretrained LLMs excel across diverse tasks, yet standard attention remains a major bottleneck for long-context understanding. Recent test-time training (TTT) methods alleviate this by encoding context into internal model parameters, but they typically require specialized architectures, meta-learned initialization, or continual post-training, preventing plug-and-play use on off-the-shelf LLMs. We propose test-time training sparse attention (TTT-Sparse), a plug-and-play alternative that performs test-time training over external memory while keeping the internal parameters frozen. For each input, TTT-Sparse constructs contextual chunk key states from the KV cache and learns a lightweight query adapter to induce a task-specific sparse attention pattern, which is used during generation to attend only to the most relevant context chunks and suppress interference from irrelevant context. Extensive experiments on long-context benchmarks show that TTT-Sparse improves long-context understanding while reducing generation-time attention cost.</abstract>
<identifier type="citekey">wang-etal-2026-ttt</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.1575/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>35866</start>
<end>35881</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T TTT-Sparse: Test-Time Training of Sparse Attention for Long-Context Understanding
%A Wang, Xuehao
%A Deng, Huayu
%A Zhong, Linhao
%A Wei, Yanbin
%A Zhao, Hengyu
%A Liu, Mingyu
%A Chen, Hao
%A Shen, Chunhua
%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 wang-etal-2026-ttt
%X Pretrained LLMs excel across diverse tasks, yet standard attention remains a major bottleneck for long-context understanding. Recent test-time training (TTT) methods alleviate this by encoding context into internal model parameters, but they typically require specialized architectures, meta-learned initialization, or continual post-training, preventing plug-and-play use on off-the-shelf LLMs. We propose test-time training sparse attention (TTT-Sparse), a plug-and-play alternative that performs test-time training over external memory while keeping the internal parameters frozen. For each input, TTT-Sparse constructs contextual chunk key states from the KV cache and learns a lightweight query adapter to induce a task-specific sparse attention pattern, which is used during generation to attend only to the most relevant context chunks and suppress interference from irrelevant context. Extensive experiments on long-context benchmarks show that TTT-Sparse improves long-context understanding while reducing generation-time attention cost.
%U https://aclanthology.org/2026.emnlp-main.1575/
%P 35866-35881
Markdown (Informal)
[TTT-Sparse: Test-Time Training of Sparse Attention for Long-Context Understanding](https://aclanthology.org/2026.emnlp-main.1575/) (Wang et al., EMNLP 2026)
ACL
- Xuehao Wang, Huayu Deng, Linhao Zhong, Yanbin Wei, Hengyu Zhao, Mingyu Liu, Hao Chen, and Chunhua Shen. 2026. TTT-Sparse: Test-Time Training of Sparse Attention for Long-Context Understanding. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 35866–35881, Budapest, Hungary. Association for Computational Linguistics.