@inproceedings{zuo-etal-2026-rethinking,
title = "Rethinking {L}o{RA} Memory Through the Lens of {KV} Cache Compression",
author = "Zuo, Chunsheng and
Wang, Liaoyaqi and
Jurayj, William and
Fleshman, William and
Van Durme, Benjamin",
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.733/",
pages = "17013--17030",
ISBN = "979-8-89176-459-0",
abstract = "Parametric retrieval augmentation encodes document information into lightweight, document-specific modules such as LoRA adapters, reducing the need to include all evidence as input context. However, it remains unclear how this parameter-side memory interacts with context-side memory stored in the KV cache. We study this interaction in document-level question answering by progressively evicting document key-value states and measuring when a document LoRA contributes beyond the retained context. We find that document LoRA adds little when the KV cache is largely intact, but becomes increasingly useful under aggressive compression, recovering 13{--}21 ROUGE-L points when no document context remains. The gain is largest when the base model encodes the document, and the adapter is applied only during answer generation, suggesting that LoRA serves as decoding-time parametric memory rather than as a document encoder. Finally, QA-style supervision produces substantially stronger adapters than raw-context, continuation, or context-plus-QA objectives. These results position document LoRA as a complementary memory channel whose value emerges precisely when context-side evidence is scarce."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zuo-etal-2026-rethinking">
<titleInfo>
<title>Rethinking LoRA Memory Through the Lens of KV Cache Compression</title>
</titleInfo>
<name type="personal">
<namePart type="given">Chunsheng</namePart>
<namePart type="family">Zuo</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Liaoyaqi</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">William</namePart>
<namePart type="family">Jurayj</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">William</namePart>
<namePart type="family">Fleshman</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Benjamin</namePart>
<namePart type="family">Van Durme</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>Parametric retrieval augmentation encodes document information into lightweight, document-specific modules such as LoRA adapters, reducing the need to include all evidence as input context. However, it remains unclear how this parameter-side memory interacts with context-side memory stored in the KV cache. We study this interaction in document-level question answering by progressively evicting document key-value states and measuring when a document LoRA contributes beyond the retained context. We find that document LoRA adds little when the KV cache is largely intact, but becomes increasingly useful under aggressive compression, recovering 13–21 ROUGE-L points when no document context remains. The gain is largest when the base model encodes the document, and the adapter is applied only during answer generation, suggesting that LoRA serves as decoding-time parametric memory rather than as a document encoder. Finally, QA-style supervision produces substantially stronger adapters than raw-context, continuation, or context-plus-QA objectives. These results position document LoRA as a complementary memory channel whose value emerges precisely when context-side evidence is scarce.</abstract>
<identifier type="citekey">zuo-etal-2026-rethinking</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.733/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>17013</start>
<end>17030</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Rethinking LoRA Memory Through the Lens of KV Cache Compression
%A Zuo, Chunsheng
%A Wang, Liaoyaqi
%A Jurayj, William
%A Fleshman, William
%A Van Durme, Benjamin
%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 zuo-etal-2026-rethinking
%X Parametric retrieval augmentation encodes document information into lightweight, document-specific modules such as LoRA adapters, reducing the need to include all evidence as input context. However, it remains unclear how this parameter-side memory interacts with context-side memory stored in the KV cache. We study this interaction in document-level question answering by progressively evicting document key-value states and measuring when a document LoRA contributes beyond the retained context. We find that document LoRA adds little when the KV cache is largely intact, but becomes increasingly useful under aggressive compression, recovering 13–21 ROUGE-L points when no document context remains. The gain is largest when the base model encodes the document, and the adapter is applied only during answer generation, suggesting that LoRA serves as decoding-time parametric memory rather than as a document encoder. Finally, QA-style supervision produces substantially stronger adapters than raw-context, continuation, or context-plus-QA objectives. These results position document LoRA as a complementary memory channel whose value emerges precisely when context-side evidence is scarce.
%U https://aclanthology.org/2026.emnlp-main.733/
%P 17013-17030
Markdown (Informal)
[Rethinking LoRA Memory Through the Lens of KV Cache Compression](https://aclanthology.org/2026.emnlp-main.733/) (Zuo et al., EMNLP 2026)
ACL
- Chunsheng Zuo, Liaoyaqi Wang, William Jurayj, William Fleshman, and Benjamin Van Durme. 2026. Rethinking LoRA Memory Through the Lens of KV Cache Compression. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 17013–17030, Budapest, Hungary. Association for Computational Linguistics.