@inproceedings{wang-etal-2026-primettt,
title = "{P}rime{TTT}: Priming Test-Time Training for {LLM}s via Meta-Learned Initialization",
author = "Wang, Xiangqi and
Aishan, Dilinuer and
Zhou, Yujun and
Huang, Yue and
Zhang, Xiangliang",
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.271/",
pages = "6152--6170",
ISBN = "979-8-89176-459-0",
abstract = "Test-time training (TTT) adapts a frozen large language model (LLM) at inference time by updating prompt-specific parameters before generation. It is especially appealing for long-context reasoning. However, vanilla TTT yields only marginal gains on modern LLMs. We trace this bottleneck to two coupled failures: the inner loop is unprimed, because LLMs are trained as static endpoints rather than initialization points for test-time gradient adaptation; and the proxy objective is misaligned, because prompt-masked next-token prediction (PM-NTP) rewards prompt memorization rather than answer correctness. We propose PrimeTTT, a lightweight TTT method that casts this adaptation problem as meta-learning. A compact hypernetwork emits a prompt-specific LoRA initialization for the inner loop, trained with a MAML-style second-order objective: the inner loop descends the test-time-available PM-NTP loss, while the outer loss is answer-conditioned NTP on the gold answer and is backpropagated through the unrolled inner steps. Across out-of-distribution benchmarks disjoint from training, PrimeTTT consistently outperforms strong TTT baselines and transfers across model families and scales."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="wang-etal-2026-primettt">
<titleInfo>
<title>PrimeTTT: Priming Test-Time Training for LLMs via Meta-Learned Initialization</title>
</titleInfo>
<name type="personal">
<namePart type="given">Xiangqi</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Dilinuer</namePart>
<namePart type="family">Aishan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yujun</namePart>
<namePart type="family">Zhou</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yue</namePart>
<namePart type="family">Huang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiangliang</namePart>
<namePart type="family">Zhang</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>Test-time training (TTT) adapts a frozen large language model (LLM) at inference time by updating prompt-specific parameters before generation. It is especially appealing for long-context reasoning. However, vanilla TTT yields only marginal gains on modern LLMs. We trace this bottleneck to two coupled failures: the inner loop is unprimed, because LLMs are trained as static endpoints rather than initialization points for test-time gradient adaptation; and the proxy objective is misaligned, because prompt-masked next-token prediction (PM-NTP) rewards prompt memorization rather than answer correctness. We propose PrimeTTT, a lightweight TTT method that casts this adaptation problem as meta-learning. A compact hypernetwork emits a prompt-specific LoRA initialization for the inner loop, trained with a MAML-style second-order objective: the inner loop descends the test-time-available PM-NTP loss, while the outer loss is answer-conditioned NTP on the gold answer and is backpropagated through the unrolled inner steps. Across out-of-distribution benchmarks disjoint from training, PrimeTTT consistently outperforms strong TTT baselines and transfers across model families and scales.</abstract>
<identifier type="citekey">wang-etal-2026-primettt</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.271/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>6152</start>
<end>6170</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T PrimeTTT: Priming Test-Time Training for LLMs via Meta-Learned Initialization
%A Wang, Xiangqi
%A Aishan, Dilinuer
%A Zhou, Yujun
%A Huang, Yue
%A Zhang, Xiangliang
%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-primettt
%X Test-time training (TTT) adapts a frozen large language model (LLM) at inference time by updating prompt-specific parameters before generation. It is especially appealing for long-context reasoning. However, vanilla TTT yields only marginal gains on modern LLMs. We trace this bottleneck to two coupled failures: the inner loop is unprimed, because LLMs are trained as static endpoints rather than initialization points for test-time gradient adaptation; and the proxy objective is misaligned, because prompt-masked next-token prediction (PM-NTP) rewards prompt memorization rather than answer correctness. We propose PrimeTTT, a lightweight TTT method that casts this adaptation problem as meta-learning. A compact hypernetwork emits a prompt-specific LoRA initialization for the inner loop, trained with a MAML-style second-order objective: the inner loop descends the test-time-available PM-NTP loss, while the outer loss is answer-conditioned NTP on the gold answer and is backpropagated through the unrolled inner steps. Across out-of-distribution benchmarks disjoint from training, PrimeTTT consistently outperforms strong TTT baselines and transfers across model families and scales.
%U https://aclanthology.org/2026.emnlp-main.271/
%P 6152-6170
Markdown (Informal)
[PrimeTTT: Priming Test-Time Training for LLMs via Meta-Learned Initialization](https://aclanthology.org/2026.emnlp-main.271/) (Wang et al., EMNLP 2026)
ACL