<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="huang-etal-2024-rolora">
    <titleInfo>
        <title>RoLoRA: Fine-tuning Rotated Outlier-free LLMs for Effective Weight-Activation Quantization</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Xijie</namePart>
        <namePart type="family">Huang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Zechun</namePart>
        <namePart type="family">Liu</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Shih-Yang</namePart>
        <namePart type="family">Liu</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Kwang-Ting</namePart>
        <namePart type="family">Cheng</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>Findings of the Association for Computational Linguistics: EMNLP 2024</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>Low-Rank Adaptation (LoRA), as a representative Parameter-Efficient Fine-Tuning (PEFT) method, significantly enhances the training efficiency by updating only a small portion of the weights in Large Language Models (LLMs). Recently, weight-only quantization techniques have also been applied to LoRA methods to reduce the memory footprint of fine-tuning. However, applying weight-activation quantization to the LoRA pipeline is under-explored, and we observe substantial performance degradation primarily due to the presence of activation outliers. In this work, we propose RoLoRA, the first LoRA-based scheme to apply rotation for outlier elimination, and then fine-tune rotated outlier-free LLMs for effective weight-activation quantization. Different from previous work tackling the outlier challenges from a post-training perspective, we propose rotation-aware fine-tuning to eliminate and preserve the outlier-free characteristics brought by rotation operations. RoLoRA can improve low-bit LoRA convergence and post-training quantization robustness in weight-activation settings. RoLoRA is evaluated across various LLM series (LLaMA2, LLaMA3, LLaVA-1.5), tasks, and quantization settings, achieving up to 29.5% absolute accuracy gain of 4-bit weight-activation quantized LLaMA2-13B on commonsense reasoning tasks compared to LoRA baseline. We further demonstrate its effectiveness on Large Multimodal Models (LMMs) and prove the compatibility with advanced LoRA variants.</abstract>
    <identifier type="citekey">huang-etal-2024-rolora</identifier>
    <identifier type="doi">10.18653/v1/2024.findings-emnlp.444</identifier>
    <location>
        <url>https://aclanthology.org/2024.findings-emnlp.444/</url>
    </location>
    <part>
        <date>2024-11</date>
        <extent unit="page">
            <start>7563</start>
            <end>7576</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T RoLoRA: Fine-tuning Rotated Outlier-free LLMs for Effective Weight-Activation Quantization
%A Huang, Xijie
%A Liu, Zechun
%A Liu, Shih-Yang
%A Cheng, Kwang-Ting
%Y Al-Onaizan, Yaser
%Y Bansal, Mohit
%Y Chen, Yun-Nung
%S Findings of the Association for Computational Linguistics: EMNLP 2024
%D 2024
%8 November
%I Association for Computational Linguistics
%C Miami, Florida, USA
%F huang-etal-2024-rolora
%X Low-Rank Adaptation (LoRA), as a representative Parameter-Efficient Fine-Tuning (PEFT) method, significantly enhances the training efficiency by updating only a small portion of the weights in Large Language Models (LLMs). Recently, weight-only quantization techniques have also been applied to LoRA methods to reduce the memory footprint of fine-tuning. However, applying weight-activation quantization to the LoRA pipeline is under-explored, and we observe substantial performance degradation primarily due to the presence of activation outliers. In this work, we propose RoLoRA, the first LoRA-based scheme to apply rotation for outlier elimination, and then fine-tune rotated outlier-free LLMs for effective weight-activation quantization. Different from previous work tackling the outlier challenges from a post-training perspective, we propose rotation-aware fine-tuning to eliminate and preserve the outlier-free characteristics brought by rotation operations. RoLoRA can improve low-bit LoRA convergence and post-training quantization robustness in weight-activation settings. RoLoRA is evaluated across various LLM series (LLaMA2, LLaMA3, LLaVA-1.5), tasks, and quantization settings, achieving up to 29.5% absolute accuracy gain of 4-bit weight-activation quantized LLaMA2-13B on commonsense reasoning tasks compared to LoRA baseline. We further demonstrate its effectiveness on Large Multimodal Models (LMMs) and prove the compatibility with advanced LoRA variants.
%R 10.18653/v1/2024.findings-emnlp.444
%U https://aclanthology.org/2024.findings-emnlp.444/
%U https://doi.org/10.18653/v1/2024.findings-emnlp.444
%P 7563-7576

Markdown (Informal)

[RoLoRA: Fine-tuning Rotated Outlier-free LLMs for Effective Weight-Activation Quantization](https://aclanthology.org/2024.findings-emnlp.444/) (Huang et al., Findings 2024)

• RoLoRA: Fine-tuning Rotated Outlier-free LLMs for Effective Weight-Activation Quantization (Huang et al., Findings 2024)

ACL

• Xijie Huang, Zechun Liu, Shih-Yang Liu, and Kwang-Ting Cheng. 2024. RoLoRA: Fine-tuning Rotated Outlier-free LLMs for Effective Weight-Activation Quantization. In Findings of the Association for Computational Linguistics: EMNLP 2024, pages 7563–7576, Miami, Florida, USA. Association for Computational Linguistics.