<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="shi-etal-2026-stir">
    <titleInfo>
        <title>STIR: Self-Distilled Latent Tools for Dynamic Control of LLM Reasoning</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Zhenning</namePart>
        <namePart type="family">Shi</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Yijia</namePart>
        <namePart type="family">Zhu</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Junhan</namePart>
        <namePart type="family">Shi</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Xun</namePart>
        <namePart type="family">Zhang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Lei</namePart>
        <namePart type="family">Wang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Congcong</namePart>
        <namePart type="family">Miao</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>Chain-of-thought reasoning improves large language models by spending more tokens at test time, but this couples accuracy to generation length. Activation steering offers a lightweight alternative, yet most steering methods apply a fixed vector throughout generation and cannot adapt to different stages of a multi-step solution. We propose STIR (Self-Distilled Tools for Internal Reasoning), a test-time framework that builds a compact library of state-conditioned correction vectors from the model’s own successful and failed rollouts. At inference time, STIR retrieves candidate corrections at structural reasoning checkpoints, validates them with a short lookahead probe, and abstains when the current trajectory already matches a high-quality state. Across six mathematical and knowledge-intensive reasoning benchmarks, STIR improves average accuracy by 1.9% to 7.5% over Vanilla decoding while reducing average token consumption by up to 35%. These results suggest that selective latent interventions can improve the accuracy-cost trade-off of reasoning without forcing every problem to generate longer chains. Our code is available at https://github.com/sznnzs/STIR.</abstract>
    <identifier type="citekey">shi-etal-2026-stir</identifier>
    <location>
        <url>https://aclanthology.org/2026.emnlp-main.2278/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>51814</start>
            <end>51829</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T STIR: Self-Distilled Latent Tools for Dynamic Control of LLM Reasoning
%A Shi, Zhenning
%A Zhu, Yijia
%A Shi, Junhan
%A Zhang, Xun
%A Wang, Lei
%A Miao, Congcong
%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 shi-etal-2026-stir
%X Chain-of-thought reasoning improves large language models by spending more tokens at test time, but this couples accuracy to generation length. Activation steering offers a lightweight alternative, yet most steering methods apply a fixed vector throughout generation and cannot adapt to different stages of a multi-step solution. We propose STIR (Self-Distilled Tools for Internal Reasoning), a test-time framework that builds a compact library of state-conditioned correction vectors from the model’s own successful and failed rollouts. At inference time, STIR retrieves candidate corrections at structural reasoning checkpoints, validates them with a short lookahead probe, and abstains when the current trajectory already matches a high-quality state. Across six mathematical and knowledge-intensive reasoning benchmarks, STIR improves average accuracy by 1.9% to 7.5% over Vanilla decoding while reducing average token consumption by up to 35%. These results suggest that selective latent interventions can improve the accuracy-cost trade-off of reasoning without forcing every problem to generate longer chains. Our code is available at https://github.com/sznnzs/STIR.
%U https://aclanthology.org/2026.emnlp-main.2278/
%P 51814-51829

Markdown (Informal)

[STIR: Self-Distilled Latent Tools for Dynamic Control of LLM Reasoning](https://aclanthology.org/2026.emnlp-main.2278/) (Shi et al., EMNLP 2026)

• STIR: Self-Distilled Latent Tools for Dynamic Control of LLM Reasoning (Shi et al., EMNLP 2026)

ACL

• Zhenning Shi, Yijia Zhu, Junhan Shi, Xun Zhang, Lei Wang, and Congcong Miao. 2026. STIR: Self-Distilled Latent Tools for Dynamic Control of LLM Reasoning. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 51814–51829, Budapest, Hungary. Association for Computational Linguistics.