<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="khodabandehlou-krishnamachari-2026-compression">
    <titleInfo>
        <title>Compression-Aware Abstention: Teaching LLMs to Refuse When KV-Compression Masks Remove Answer Evidence</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Mohammadali</namePart>
        <namePart type="family">Khodabandehlou</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Bhaskar</namePart>
        <namePart type="family">Krishnamachari</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 1st Workshop on Grounding Language Models: Learning Faithfully and Efficiently (GroundLM 2026)</title>
        </titleInfo>
        <name type="personal">
            <namePart type="given">Yimu</namePart>
            <namePart type="family">Wang</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Yee</namePart>
            <namePart type="given">Man</namePart>
            <namePart type="family">Choi</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Di</namePart>
            <namePart type="family">Wu</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Siqi</namePart>
            <namePart type="family">Ouyang</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Mozhgan</namePart>
            <namePart type="given">Nasr</namePart>
            <namePart type="family">Azadani</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Yi</namePart>
            <namePart type="given">R</namePart>
            <namePart type="given">(May)</namePart>
            <namePart type="family">Fung</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">979-8-89176-478-1</identifier>
    </relatedItem>
    <abstract>KV-cache compression reduces LLM inference memory by evicting context tokens, but when the evicted tokens contain answer-bearing evidence, the model may hallucinate instead of recognizing that the compressed context is insufficient. We address this failure from a behavioral perspective: to our knowledge, this is the first work to formulate compression-aware abstention as a learning problem, in which a model learns to answer when supporting evidence survives compression and abstain when it does not. We construct supervision from compressor survival masks and tight answer-bearing spans, labeling examples as Confident when evidence survives and Abstain when it is removed. A 10.1M-parameter LoRA adapter trained on \ensuremath\sim2.6K MuSiQue 2-hop QA examples reduces base-model hallucinations by 97% under prompt-style truncation while preserving correct answering on evidence-retaining examples. Unlike prompt-only abstention baselines, which over-abstain on many answerable high-retention examples, the trained adapter learns a conditional policy. We also evaluate the method under actual compressed-cache decoding, where multi-compressor training yields a 6–22× relative lift over the unaided base on evidence-retaining examples. Controlled-deletion experiments show that the learned behavior is driven by evidence content rather than input length alone.</abstract>
    <identifier type="citekey">khodabandehlou-krishnamachari-2026-compression</identifier>
    <location>
        <url>https://aclanthology.org/2026.groundlm-1.42/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>531</start>
            <end>549</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Compression-Aware Abstention: Teaching LLMs to Refuse When KV-Compression Masks Remove Answer Evidence
%A Khodabandehlou, Mohammadali
%A Krishnamachari, Bhaskar
%Y Wang, Yimu
%Y Choi, Yee Man
%Y Wu, Di
%Y Ouyang, Siqi
%Y Azadani, Mozhgan Nasr
%Y Fung, Yi R. (May)
%S Proceedings of the 1st Workshop on Grounding Language Models: Learning Faithfully and Efficiently (GroundLM 2026)
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-478-1
%F khodabandehlou-krishnamachari-2026-compression
%X KV-cache compression reduces LLM inference memory by evicting context tokens, but when the evicted tokens contain answer-bearing evidence, the model may hallucinate instead of recognizing that the compressed context is insufficient. We address this failure from a behavioral perspective: to our knowledge, this is the first work to formulate compression-aware abstention as a learning problem, in which a model learns to answer when supporting evidence survives compression and abstain when it does not. We construct supervision from compressor survival masks and tight answer-bearing spans, labeling examples as Confident when evidence survives and Abstain when it is removed. A 10.1M-parameter LoRA adapter trained on \ensuremath\sim2.6K MuSiQue 2-hop QA examples reduces base-model hallucinations by 97% under prompt-style truncation while preserving correct answering on evidence-retaining examples. Unlike prompt-only abstention baselines, which over-abstain on many answerable high-retention examples, the trained adapter learns a conditional policy. We also evaluate the method under actual compressed-cache decoding, where multi-compressor training yields a 6–22× relative lift over the unaided base on evidence-retaining examples. Controlled-deletion experiments show that the learned behavior is driven by evidence content rather than input length alone.
%U https://aclanthology.org/2026.groundlm-1.42/
%P 531-549

Markdown (Informal)

[Compression-Aware Abstention: Teaching LLMs to Refuse When KV-Compression Masks Remove Answer Evidence](https://aclanthology.org/2026.groundlm-1.42/) (Khodabandehlou & Krishnamachari, GroundLM 2026)

• Compression-Aware Abstention: Teaching LLMs to Refuse When KV-Compression Masks Remove Answer Evidence (Khodabandehlou & Krishnamachari, GroundLM 2026)

ACL

• Mohammadali Khodabandehlou and Bhaskar Krishnamachari. 2026. Compression-Aware Abstention: Teaching LLMs to Refuse When KV-Compression Masks Remove Answer Evidence. In Proceedings of the 1st Workshop on Grounding Language Models: Learning Faithfully and Efficiently (GroundLM 2026), pages 531–549, Budapest, Hungary. Association for Computational Linguistics.