Efficient Storage of Fine-Tuned Models via Low-Rank Approximation of Weight Residuals

In this paper, we present an efficient method for storing fine-tuned models by leveraging the low-rank properties of weight residuals. Our key observation is that weight residuals in large overparameterized models exhibit even stronger low-rank characteristics. Based on this insight, we propose Effi...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:arXiv.org 2023-05
Hauptverfasser: Ryu, Simo, Seo, Seunghyun, Yoo, Jaejun
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
container_end_page
container_issue
container_start_page
container_title arXiv.org
container_volume
creator Ryu, Simo
Seo, Seunghyun
Yoo, Jaejun
description In this paper, we present an efficient method for storing fine-tuned models by leveraging the low-rank properties of weight residuals. Our key observation is that weight residuals in large overparameterized models exhibit even stronger low-rank characteristics. Based on this insight, we propose Efficient Residual Encoding (ERE), a novel approach that achieves efficient storage of fine-tuned model weights by approximating the low-rank weight residuals. Furthermore, we analyze the robustness of weight residuals and push the limit of storage efficiency by utilizing additional quantization and layer-wise rank allocation. Our experimental results demonstrate that our method significantly reduces memory footprint while preserving performance in various tasks and modalities. We release our code.
format Article
fullrecord <record><control><sourceid>proquest</sourceid><recordid>TN_cdi_proquest_journals_2821113524</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2821113524</sourcerecordid><originalsourceid>FETCH-proquest_journals_28211135243</originalsourceid><addsrcrecordid>eNqNi0ELgjAYQEcQJOV_GHQeuE3La4TSoS5mdJSR32xmm7mt-vkZ9AM6vcN7b4ICxjklaczYDIXWtlEUsdWaJQkP0CmTUl0UaIePzgyiAWwkzpUGUnoNNT6YGjqLn0rgvXmRQugb3vT9YN7qLpwy-tufQTVXhwuwqvaisws0lSMg_HGOlnlWbndk_B4erKta4wc9qoqljFLKExbz_6oPYVxAHQ</addsrcrecordid><sourcetype>Aggregation Database</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype><pqid>2821113524</pqid></control><display><type>article</type><title>Efficient Storage of Fine-Tuned Models via Low-Rank Approximation of Weight Residuals</title><source>Freely Accessible Journals</source><creator>Ryu, Simo ; Seo, Seunghyun ; Yoo, Jaejun</creator><creatorcontrib>Ryu, Simo ; Seo, Seunghyun ; Yoo, Jaejun</creatorcontrib><description>In this paper, we present an efficient method for storing fine-tuned models by leveraging the low-rank properties of weight residuals. Our key observation is that weight residuals in large overparameterized models exhibit even stronger low-rank characteristics. Based on this insight, we propose Efficient Residual Encoding (ERE), a novel approach that achieves efficient storage of fine-tuned model weights by approximating the low-rank weight residuals. Furthermore, we analyze the robustness of weight residuals and push the limit of storage efficiency by utilizing additional quantization and layer-wise rank allocation. Our experimental results demonstrate that our method significantly reduces memory footprint while preserving performance in various tasks and modalities. We release our code.</description><identifier>EISSN: 2331-8422</identifier><language>eng</language><publisher>Ithaca: Cornell University Library, arXiv.org</publisher><subject>Approximation ; Storage</subject><ispartof>arXiv.org, 2023-05</ispartof><rights>2023. This work is published under http://creativecommons.org/licenses/by/4.0/ (the “License”). Notwithstanding the ProQuest Terms and Conditions, you may use this content in accordance with the terms of the License.</rights><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>776,780</link.rule.ids></links><search><creatorcontrib>Ryu, Simo</creatorcontrib><creatorcontrib>Seo, Seunghyun</creatorcontrib><creatorcontrib>Yoo, Jaejun</creatorcontrib><title>Efficient Storage of Fine-Tuned Models via Low-Rank Approximation of Weight Residuals</title><title>arXiv.org</title><description>In this paper, we present an efficient method for storing fine-tuned models by leveraging the low-rank properties of weight residuals. Our key observation is that weight residuals in large overparameterized models exhibit even stronger low-rank characteristics. Based on this insight, we propose Efficient Residual Encoding (ERE), a novel approach that achieves efficient storage of fine-tuned model weights by approximating the low-rank weight residuals. Furthermore, we analyze the robustness of weight residuals and push the limit of storage efficiency by utilizing additional quantization and layer-wise rank allocation. Our experimental results demonstrate that our method significantly reduces memory footprint while preserving performance in various tasks and modalities. We release our code.</description><subject>Approximation</subject><subject>Storage</subject><issn>2331-8422</issn><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2023</creationdate><recordtype>article</recordtype><sourceid>ABUWG</sourceid><sourceid>AFKRA</sourceid><sourceid>AZQEC</sourceid><sourceid>BENPR</sourceid><sourceid>CCPQU</sourceid><sourceid>DWQXO</sourceid><recordid>eNqNi0ELgjAYQEcQJOV_GHQeuE3La4TSoS5mdJSR32xmm7mt-vkZ9AM6vcN7b4ICxjklaczYDIXWtlEUsdWaJQkP0CmTUl0UaIePzgyiAWwkzpUGUnoNNT6YGjqLn0rgvXmRQugb3vT9YN7qLpwy-tufQTVXhwuwqvaisws0lSMg_HGOlnlWbndk_B4erKta4wc9qoqljFLKExbz_6oPYVxAHQ</recordid><startdate>20230528</startdate><enddate>20230528</enddate><creator>Ryu, Simo</creator><creator>Seo, Seunghyun</creator><creator>Yoo, Jaejun</creator><general>Cornell University Library, arXiv.org</general><scope>8FE</scope><scope>8FG</scope><scope>ABJCF</scope><scope>ABUWG</scope><scope>AFKRA</scope><scope>AZQEC</scope><scope>BENPR</scope><scope>BGLVJ</scope><scope>CCPQU</scope><scope>DWQXO</scope><scope>HCIFZ</scope><scope>L6V</scope><scope>M7S</scope><scope>PIMPY</scope><scope>PQEST</scope><scope>PQQKQ</scope><scope>PQUKI</scope><scope>PRINS</scope><scope>PTHSS</scope></search><sort><creationdate>20230528</creationdate><title>Efficient Storage of Fine-Tuned Models via Low-Rank Approximation of Weight Residuals</title><author>Ryu, Simo ; Seo, Seunghyun ; Yoo, Jaejun</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-proquest_journals_28211135243</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2023</creationdate><topic>Approximation</topic><topic>Storage</topic><toplevel>online_resources</toplevel><creatorcontrib>Ryu, Simo</creatorcontrib><creatorcontrib>Seo, Seunghyun</creatorcontrib><creatorcontrib>Yoo, Jaejun</creatorcontrib><collection>ProQuest SciTech Collection</collection><collection>ProQuest Technology Collection</collection><collection>Materials Science &amp; Engineering Collection</collection><collection>ProQuest Central (Alumni Edition)</collection><collection>ProQuest Central UK/Ireland</collection><collection>ProQuest Central Essentials</collection><collection>ProQuest Central</collection><collection>Technology Collection</collection><collection>ProQuest One Community College</collection><collection>ProQuest Central Korea</collection><collection>SciTech Premium Collection</collection><collection>ProQuest Engineering Collection</collection><collection>Engineering Database</collection><collection>Publicly Available Content Database</collection><collection>ProQuest One Academic Eastern Edition (DO NOT USE)</collection><collection>ProQuest One Academic</collection><collection>ProQuest One Academic UKI Edition</collection><collection>ProQuest Central China</collection><collection>Engineering Collection</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext</fulltext></delivery><addata><au>Ryu, Simo</au><au>Seo, Seunghyun</au><au>Yoo, Jaejun</au><format>book</format><genre>document</genre><ristype>GEN</ristype><atitle>Efficient Storage of Fine-Tuned Models via Low-Rank Approximation of Weight Residuals</atitle><jtitle>arXiv.org</jtitle><date>2023-05-28</date><risdate>2023</risdate><eissn>2331-8422</eissn><abstract>In this paper, we present an efficient method for storing fine-tuned models by leveraging the low-rank properties of weight residuals. Our key observation is that weight residuals in large overparameterized models exhibit even stronger low-rank characteristics. Based on this insight, we propose Efficient Residual Encoding (ERE), a novel approach that achieves efficient storage of fine-tuned model weights by approximating the low-rank weight residuals. Furthermore, we analyze the robustness of weight residuals and push the limit of storage efficiency by utilizing additional quantization and layer-wise rank allocation. Our experimental results demonstrate that our method significantly reduces memory footprint while preserving performance in various tasks and modalities. We release our code.</abstract><cop>Ithaca</cop><pub>Cornell University Library, arXiv.org</pub><oa>free_for_read</oa></addata></record>
fulltext fulltext
identifier EISSN: 2331-8422
ispartof arXiv.org, 2023-05
issn 2331-8422
language eng
recordid cdi_proquest_journals_2821113524
source Freely Accessible Journals
subjects Approximation
Storage
title Efficient Storage of Fine-Tuned Models via Low-Rank Approximation of Weight Residuals
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-01-23T03%3A07%3A18IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-proquest&rft_val_fmt=info:ofi/fmt:kev:mtx:book&rft.genre=document&rft.atitle=Efficient%20Storage%20of%20Fine-Tuned%20Models%20via%20Low-Rank%20Approximation%20of%20Weight%20Residuals&rft.jtitle=arXiv.org&rft.au=Ryu,%20Simo&rft.date=2023-05-28&rft.eissn=2331-8422&rft_id=info:doi/&rft_dat=%3Cproquest%3E2821113524%3C/proquest%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_pqid=2821113524&rft_id=info:pmid/&rfr_iscdi=true