<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="shen-etal-2024-pruning">
    <titleInfo>
        <title>Pruning Large Language Models to Intra-module Low-rank Architecture with Transitional Activations</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Bowen</namePart>
        <namePart type="family">Shen</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Zheng</namePart>
        <namePart type="family">Lin</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Daren</namePart>
        <namePart type="family">Zha</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Wei</namePart>
        <namePart type="family">Liu</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Jian</namePart>
        <namePart type="family">Luan</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Bin</namePart>
        <namePart type="family">Wang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Weiping</namePart>
        <namePart type="family">Wang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <originInfo>
        <dateIssued>2024-08</dateIssued>
    </originInfo>
    <typeOfResource>text</typeOfResource>
    <relatedItem type="host">
        <titleInfo>
            <title>Findings of the Association for Computational Linguistics: ACL 2024</title>
        </titleInfo>
        <name type="personal">
            <namePart type="given">Lun-Wei</namePart>
            <namePart type="family">Ku</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Andre</namePart>
            <namePart type="family">Martins</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Vivek</namePart>
            <namePart type="family">Srikumar</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <originInfo>
            <publisher>Association for Computational Linguistics</publisher>
            <place>
                <placeTerm type="text">Bangkok, Thailand</placeTerm>
            </place>
        </originInfo>
        <genre authority="marcgt">conference publication</genre>
    </relatedItem>
    <abstract>Structured pruning fundamentally reduces computational and memory overheads of large language models (LLMs) and offers a feasible solution for end-side LLM deployment. Structurally pruned models remain dense and high-precision, highly compatible with further tuning and compression. However, as the coarse-grained structured pruning poses large damage to the highly interconnected model, achieving a high compression ratio for scaled-up LLMs remains a challenge. In this paper, we introduce a task-agnostic structured pruning approach coupled with a compact Transformer architecture design. The proposed approach, named TransAct, reduces transitional activations inside multi-head attention (MHA) and multi-layer perceptron (MLP) modules, while preserving the inter-module activations that are sensitive to perturbations. Hence, the LLM is pruned into an intra-module low-rank architecture, significantly reducing weights, KV Cache and attention computation. TransAct is implemented on the LLaMA model and evaluated on downstream benchmarks. Results verify the optimality of our approach at high compression with respect to both efficiency and performance. Further, ablation studies reveal the strength of activation-guided iterative pruning and provide experimental analysis on the redundancy of MHA and MLP modules.</abstract>
    <identifier type="citekey">shen-etal-2024-pruning</identifier>
    <identifier type="doi">10.18653/v1/2024.findings-acl.582</identifier>
    <location>
        <url>https://aclanthology.org/2024.findings-acl.582</url>
    </location>
    <part>
        <date>2024-08</date>
        <extent unit="page">
            <start>9781</start>
            <end>9793</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Pruning Large Language Models to Intra-module Low-rank Architecture with Transitional Activations
%A Shen, Bowen
%A Lin, Zheng
%A Zha, Daren
%A Liu, Wei
%A Luan, Jian
%A Wang, Bin
%A Wang, Weiping
%Y Ku, Lun-Wei
%Y Martins, Andre
%Y Srikumar, Vivek
%S Findings of the Association for Computational Linguistics: ACL 2024
%D 2024
%8 August
%I Association for Computational Linguistics
%C Bangkok, Thailand
%F shen-etal-2024-pruning
%X Structured pruning fundamentally reduces computational and memory overheads of large language models (LLMs) and offers a feasible solution for end-side LLM deployment. Structurally pruned models remain dense and high-precision, highly compatible with further tuning and compression. However, as the coarse-grained structured pruning poses large damage to the highly interconnected model, achieving a high compression ratio for scaled-up LLMs remains a challenge. In this paper, we introduce a task-agnostic structured pruning approach coupled with a compact Transformer architecture design. The proposed approach, named TransAct, reduces transitional activations inside multi-head attention (MHA) and multi-layer perceptron (MLP) modules, while preserving the inter-module activations that are sensitive to perturbations. Hence, the LLM is pruned into an intra-module low-rank architecture, significantly reducing weights, KV Cache and attention computation. TransAct is implemented on the LLaMA model and evaluated on downstream benchmarks. Results verify the optimality of our approach at high compression with respect to both efficiency and performance. Further, ablation studies reveal the strength of activation-guided iterative pruning and provide experimental analysis on the redundancy of MHA and MLP modules.
%R 10.18653/v1/2024.findings-acl.582
%U https://aclanthology.org/2024.findings-acl.582
%U https://doi.org/10.18653/v1/2024.findings-acl.582
%P 9781-9793

Markdown (Informal)

[Pruning Large Language Models to Intra-module Low-rank Architecture with Transitional Activations](https://aclanthology.org/2024.findings-acl.582) (Shen et al., Findings 2024)

• Pruning Large Language Models to Intra-module Low-rank Architecture with Transitional Activations (Shen et al., Findings 2024)

ACL

• Bowen Shen, Zheng Lin, Daren Zha, Wei Liu, Jian Luan, Bin Wang, and Weiping Wang. 2024. Pruning Large Language Models to Intra-module Low-rank Architecture with Transitional Activations. In Findings of the Association for Computational Linguistics: ACL 2024, pages 9781–9793, Bangkok, Thailand. Association for Computational Linguistics.