<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zheng-etal-2026-learning">
    <titleInfo>
        <title>Learning New Facts with QLoRA: An Acquisition-Retention Frontier</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Estelle</namePart>
        <namePart type="family">Zheng</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Sébastien</namePart>
        <namePart type="family">Warichet</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Emmanuel</namePart>
        <namePart type="family">Helbert</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Christophe</namePart>
        <namePart type="family">Cerisara</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>Findings of the Association for Computational Linguistics: EMNLP 2026</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">979-8-89176-462-0</identifier>
    </relatedItem>
    <abstract>Parameter-efficient fine-tuning (PEFT) is often assumed to preserve pretrained capabilities because it updates only a small number of parameters. We show that this assumption depends strongly on adapter capacity. We study factual acquisition in a controlled OpenStreetMap-derived benchmark where Qwen3-4B must learn anonymized geographic relations while retaining unrelated capabilities. Comparing FFT with QLoRA ranks 8, 16, 32, and 64, we find that rank induces a clear acquisition–retention frontier. Low-rank QLoRA preserves OOD performance but acquires fewer facts, whereas higher ranks improve same-fact paraphrase generalization at increasing cost to unrelated benchmarks. FFT behaves as a conservative baseline: it retains general capabilities well, but does not reach the highest factual-acquisition regime. Distributional, weight-space, and spectral diagnostics mirror this behavioral trade-off, with higher-rank QLoRA moving farther from the pretrained model. A separate math adaptation experiment shows a weaker frontier, suggesting that the effect is most pronounced when adaptation must install new factual associations rather than reinforce skills already supported by pretraining.</abstract>
    <identifier type="citekey">zheng-etal-2026-learning</identifier>
    <location>
        <url>https://aclanthology.org/2026.findings-emnlp.918/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>19397</start>
            <end>19407</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Learning New Facts with QLoRA: An Acquisition-Retention Frontier
%A Zheng, Estelle
%A Warichet, Sébastien
%A Helbert, Emmanuel
%A Cerisara, Christophe
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F zheng-etal-2026-learning
%X Parameter-efficient fine-tuning (PEFT) is often assumed to preserve pretrained capabilities because it updates only a small number of parameters. We show that this assumption depends strongly on adapter capacity. We study factual acquisition in a controlled OpenStreetMap-derived benchmark where Qwen3-4B must learn anonymized geographic relations while retaining unrelated capabilities. Comparing FFT with QLoRA ranks 8, 16, 32, and 64, we find that rank induces a clear acquisition–retention frontier. Low-rank QLoRA preserves OOD performance but acquires fewer facts, whereas higher ranks improve same-fact paraphrase generalization at increasing cost to unrelated benchmarks. FFT behaves as a conservative baseline: it retains general capabilities well, but does not reach the highest factual-acquisition regime. Distributional, weight-space, and spectral diagnostics mirror this behavioral trade-off, with higher-rank QLoRA moving farther from the pretrained model. A separate math adaptation experiment shows a weaker frontier, suggesting that the effect is most pronounced when adaptation must install new factual associations rather than reinforce skills already supported by pretraining.
%U https://aclanthology.org/2026.findings-emnlp.918/
%P 19397-19407

Markdown (Informal)

[Learning New Facts with QLoRA: An Acquisition-Retention Frontier](https://aclanthology.org/2026.findings-emnlp.918/) (Zheng et al., Findings 2026)

• Learning New Facts with QLoRA: An Acquisition-Retention Frontier (Zheng et al., Findings 2026)

ACL

• Estelle Zheng, Sébastien Warichet, Emmanuel Helbert, and Christophe Cerisara. 2026. Learning New Facts with QLoRA: An Acquisition-Retention Frontier. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 19397–19407, Budapest, Hungary. Association for Computational Linguistics.