<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="chung-etal-2024-stable">
    <titleInfo>
        <title>Stable Language Model Pre-training by Reducing Embedding Variability</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Woojin</namePart>
        <namePart type="family">Chung</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Jiwoo</namePart>
        <namePart type="family">Hong</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Na</namePart>
        <namePart type="given">Min</namePart>
        <namePart type="family">An</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">James</namePart>
        <namePart type="family">Thorne</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Se-Young</namePart>
        <namePart type="family">Yun</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <originInfo>
        <dateIssued>2024-11</dateIssued>
    </originInfo>
    <typeOfResource>text</typeOfResource>
    <relatedItem type="host">
        <titleInfo>
            <title>Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing</title>
        </titleInfo>
        <name type="personal">
            <namePart type="given">Yaser</namePart>
            <namePart type="family">Al-Onaizan</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Mohit</namePart>
            <namePart type="family">Bansal</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Yun-Nung</namePart>
            <namePart type="family">Chen</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <originInfo>
            <publisher>Association for Computational Linguistics</publisher>
            <place>
                <placeTerm type="text">Miami, Florida, USA</placeTerm>
            </place>
        </originInfo>
        <genre authority="marcgt">conference publication</genre>
    </relatedItem>
    <abstract>Stable pre-training is essential for achieving better-performing language models. However, tracking pre-training stability is impractical due to high computational costs. We study Token Embedding Variability as a simple proxy to estimate pre-training stability. We theoretically and empirically demonstrate that Multi-head Low-Rank Attention acts as a fundamental approach to reducing instability. This is supported by empirical findings on variants on GPT-2, demonstrating improved stability and lower perplexities, even at deeper layer counts.</abstract>
    <identifier type="citekey">chung-etal-2024-stable</identifier>
    <identifier type="doi">10.18653/v1/2024.emnlp-main.606</identifier>
    <location>
        <url>https://aclanthology.org/2024.emnlp-main.606/</url>
    </location>
    <part>
        <date>2024-11</date>
        <extent unit="page">
            <start>10852</start>
            <end>10863</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Stable Language Model Pre-training by Reducing Embedding Variability
%A Chung, Woojin
%A Hong, Jiwoo
%A An, Na Min
%A Thorne, James
%A Yun, Se-Young
%Y Al-Onaizan, Yaser
%Y Bansal, Mohit
%Y Chen, Yun-Nung
%S Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing
%D 2024
%8 November
%I Association for Computational Linguistics
%C Miami, Florida, USA
%F chung-etal-2024-stable
%X Stable pre-training is essential for achieving better-performing language models. However, tracking pre-training stability is impractical due to high computational costs. We study Token Embedding Variability as a simple proxy to estimate pre-training stability. We theoretically and empirically demonstrate that Multi-head Low-Rank Attention acts as a fundamental approach to reducing instability. This is supported by empirical findings on variants on GPT-2, demonstrating improved stability and lower perplexities, even at deeper layer counts.
%R 10.18653/v1/2024.emnlp-main.606
%U https://aclanthology.org/2024.emnlp-main.606/
%U https://doi.org/10.18653/v1/2024.emnlp-main.606
%P 10852-10863

Markdown (Informal)

[Stable Language Model Pre-training by Reducing Embedding Variability](https://aclanthology.org/2024.emnlp-main.606/) (Chung et al., EMNLP 2024)

• Stable Language Model Pre-training by Reducing Embedding Variability (Chung et al., EMNLP 2024)

ACL

• Woojin Chung, Jiwoo Hong, Na Min An, James Thorne, and Se-Young Yun. 2024. Stable Language Model Pre-training by Reducing Embedding Variability. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 10852–10863, Miami, Florida, USA. Association for Computational Linguistics.