Just-in-time Quantization with Processing-In-Memory for Efficient ML Training

Data format innovations have been critical for machine learning (ML) scaling, which in turn fuels ground-breaking ML capabilities. However, even in the presence of low-precision formats, model weights are often stored in both high-precision and low-precision during training. Furthermore, with emergi...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:arXiv.org 2023-11
Hauptverfasser: Mohamed Assem Ibrahim, Shaizeen Aga, Li, Ada, Pati, Suchita, Islam, Mahzabeen
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 Mohamed Assem Ibrahim
Shaizeen Aga
Li, Ada
Pati, Suchita
Islam, Mahzabeen
description Data format innovations have been critical for machine learning (ML) scaling, which in turn fuels ground-breaking ML capabilities. However, even in the presence of low-precision formats, model weights are often stored in both high-precision and low-precision during training. Furthermore, with emerging directional data formats (e.g., MX9, MX6, etc.) multiple low-precision weight copies can be required. To lower memory capacity needs of weights, we explore just-in-time quantization (JIT-Q) where we only store high-precision weights in memory and generate low-precision weights only when needed. To perform JIT-Q efficiently, in this work, we evaluate emerging processing-in-memory (PIM) technology to execute quantization. With PIM, we can offload quantization to in-memory compute units enabling quantization to be performed without incurring costly data movement while allowing quantization to be concurrent with accelerator computation. Our proposed PIM-offloaded quantization keeps up with GPU compute and delivers considerable capacity savings (up to 24\%) at marginal throughput loss (up to 2.4\%). Said memory capacity savings can unlock several benefits such as fitting larger model in the same system, reducing model parallelism requirement, and improving overall ML training efficiency.
format Article
fullrecord <record><control><sourceid>proquest</sourceid><recordid>TN_cdi_proquest_journals_2888461152</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2888461152</sourcerecordid><originalsourceid>FETCH-proquest_journals_28884611523</originalsourceid><addsrcrecordid>eNqNyr0KwjAUQOEgCBbtOwScA23S1uxSUbGg0L2Ekugt9kbzg-jT28EHcDrD-WYk4ULkTBacL0jq_ZBlGa82vCxFQppj9IEBsgCjppeoMMBHBbBIXxBu9Oxsr70HvLIDskaP1r2psY7WxkAPGgNtTrR1CnAyKzI36u51-uuSrHd1u92zh7PPqH3oBhsdTqvjUsqiyvOSi__UF4JzPaU</addsrcrecordid><sourcetype>Aggregation Database</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype><pqid>2888461152</pqid></control><display><type>article</type><title>Just-in-time Quantization with Processing-In-Memory for Efficient ML Training</title><source>Free E- Journals</source><creator>Mohamed Assem Ibrahim ; Shaizeen Aga ; Li, Ada ; Pati, Suchita ; Islam, Mahzabeen</creator><creatorcontrib>Mohamed Assem Ibrahim ; Shaizeen Aga ; Li, Ada ; Pati, Suchita ; Islam, Mahzabeen</creatorcontrib><description>Data format innovations have been critical for machine learning (ML) scaling, which in turn fuels ground-breaking ML capabilities. However, even in the presence of low-precision formats, model weights are often stored in both high-precision and low-precision during training. Furthermore, with emerging directional data formats (e.g., MX9, MX6, etc.) multiple low-precision weight copies can be required. To lower memory capacity needs of weights, we explore just-in-time quantization (JIT-Q) where we only store high-precision weights in memory and generate low-precision weights only when needed. To perform JIT-Q efficiently, in this work, we evaluate emerging processing-in-memory (PIM) technology to execute quantization. With PIM, we can offload quantization to in-memory compute units enabling quantization to be performed without incurring costly data movement while allowing quantization to be concurrent with accelerator computation. Our proposed PIM-offloaded quantization keeps up with GPU compute and delivers considerable capacity savings (up to 24\%) at marginal throughput loss (up to 2.4\%). Said memory capacity savings can unlock several benefits such as fitting larger model in the same system, reducing model parallelism requirement, and improving overall ML training efficiency.</description><identifier>EISSN: 2331-8422</identifier><language>eng</language><publisher>Ithaca: Cornell University Library, arXiv.org</publisher><subject>Just in time ; Machine learning</subject><ispartof>arXiv.org, 2023-11</ispartof><rights>2023. This work is published under http://arxiv.org/licenses/nonexclusive-distrib/1.0/ (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>780,784</link.rule.ids></links><search><creatorcontrib>Mohamed Assem Ibrahim</creatorcontrib><creatorcontrib>Shaizeen Aga</creatorcontrib><creatorcontrib>Li, Ada</creatorcontrib><creatorcontrib>Pati, Suchita</creatorcontrib><creatorcontrib>Islam, Mahzabeen</creatorcontrib><title>Just-in-time Quantization with Processing-In-Memory for Efficient ML Training</title><title>arXiv.org</title><description>Data format innovations have been critical for machine learning (ML) scaling, which in turn fuels ground-breaking ML capabilities. However, even in the presence of low-precision formats, model weights are often stored in both high-precision and low-precision during training. Furthermore, with emerging directional data formats (e.g., MX9, MX6, etc.) multiple low-precision weight copies can be required. To lower memory capacity needs of weights, we explore just-in-time quantization (JIT-Q) where we only store high-precision weights in memory and generate low-precision weights only when needed. To perform JIT-Q efficiently, in this work, we evaluate emerging processing-in-memory (PIM) technology to execute quantization. With PIM, we can offload quantization to in-memory compute units enabling quantization to be performed without incurring costly data movement while allowing quantization to be concurrent with accelerator computation. Our proposed PIM-offloaded quantization keeps up with GPU compute and delivers considerable capacity savings (up to 24\%) at marginal throughput loss (up to 2.4\%). Said memory capacity savings can unlock several benefits such as fitting larger model in the same system, reducing model parallelism requirement, and improving overall ML training efficiency.</description><subject>Just in time</subject><subject>Machine learning</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>eNqNyr0KwjAUQOEgCBbtOwScA23S1uxSUbGg0L2Ekugt9kbzg-jT28EHcDrD-WYk4ULkTBacL0jq_ZBlGa82vCxFQppj9IEBsgCjppeoMMBHBbBIXxBu9Oxsr70HvLIDskaP1r2psY7WxkAPGgNtTrR1CnAyKzI36u51-uuSrHd1u92zh7PPqH3oBhsdTqvjUsqiyvOSi__UF4JzPaU</recordid><startdate>20231108</startdate><enddate>20231108</enddate><creator>Mohamed Assem Ibrahim</creator><creator>Shaizeen Aga</creator><creator>Li, Ada</creator><creator>Pati, Suchita</creator><creator>Islam, Mahzabeen</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>20231108</creationdate><title>Just-in-time Quantization with Processing-In-Memory for Efficient ML Training</title><author>Mohamed Assem Ibrahim ; Shaizeen Aga ; Li, Ada ; Pati, Suchita ; Islam, Mahzabeen</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-proquest_journals_28884611523</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2023</creationdate><topic>Just in time</topic><topic>Machine learning</topic><toplevel>online_resources</toplevel><creatorcontrib>Mohamed Assem Ibrahim</creatorcontrib><creatorcontrib>Shaizeen Aga</creatorcontrib><creatorcontrib>Li, Ada</creatorcontrib><creatorcontrib>Pati, Suchita</creatorcontrib><creatorcontrib>Islam, Mahzabeen</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>Mohamed Assem Ibrahim</au><au>Shaizeen Aga</au><au>Li, Ada</au><au>Pati, Suchita</au><au>Islam, Mahzabeen</au><format>book</format><genre>document</genre><ristype>GEN</ristype><atitle>Just-in-time Quantization with Processing-In-Memory for Efficient ML Training</atitle><jtitle>arXiv.org</jtitle><date>2023-11-08</date><risdate>2023</risdate><eissn>2331-8422</eissn><abstract>Data format innovations have been critical for machine learning (ML) scaling, which in turn fuels ground-breaking ML capabilities. However, even in the presence of low-precision formats, model weights are often stored in both high-precision and low-precision during training. Furthermore, with emerging directional data formats (e.g., MX9, MX6, etc.) multiple low-precision weight copies can be required. To lower memory capacity needs of weights, we explore just-in-time quantization (JIT-Q) where we only store high-precision weights in memory and generate low-precision weights only when needed. To perform JIT-Q efficiently, in this work, we evaluate emerging processing-in-memory (PIM) technology to execute quantization. With PIM, we can offload quantization to in-memory compute units enabling quantization to be performed without incurring costly data movement while allowing quantization to be concurrent with accelerator computation. Our proposed PIM-offloaded quantization keeps up with GPU compute and delivers considerable capacity savings (up to 24\%) at marginal throughput loss (up to 2.4\%). Said memory capacity savings can unlock several benefits such as fitting larger model in the same system, reducing model parallelism requirement, and improving overall ML training efficiency.</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, 2023-11
issn 2331-8422
language eng
recordid cdi_proquest_journals_2888461152
source Free E- Journals
subjects Just in time
Machine learning
title Just-in-time Quantization with Processing-In-Memory for Efficient ML Training
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2024-12-25T12%3A13%3A42IST&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=Just-in-time%20Quantization%20with%20Processing-In-Memory%20for%20Efficient%20ML%20Training&rft.jtitle=arXiv.org&rft.au=Mohamed%20Assem%20Ibrahim&rft.date=2023-11-08&rft.eissn=2331-8422&rft_id=info:doi/&rft_dat=%3Cproquest%3E2888461152%3C/proquest%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_pqid=2888461152&rft_id=info:pmid/&rfr_iscdi=true