CLAMP-ViT: Contrastive Data-Free Learning for Adaptive Post-Training Quantization of ViTs
We present CLAMP-ViT, a data-free post-training quantization method for vision transformers (ViTs). We identify the limitations of recent techniques, notably their inability to leverage meaningful inter-patch relationships, leading to the generation of simplistic and semantically vague data, impacti...
Gespeichert in:
Veröffentlicht in: | arXiv.org 2024-09 |
---|---|
Hauptverfasser: | , , |
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 | Ramachandran, Akshat Kundu, Souvik Krishna, Tushar |
description | We present CLAMP-ViT, a data-free post-training quantization method for vision transformers (ViTs). We identify the limitations of recent techniques, notably their inability to leverage meaningful inter-patch relationships, leading to the generation of simplistic and semantically vague data, impacting quantization accuracy. CLAMP-ViT employs a two-stage approach, cyclically adapting between data generation and model quantization. Specifically, we incorporate a patch-level contrastive learning scheme to generate richer, semantically meaningful data. Furthermore, we leverage contrastive learning in layer-wise evolutionary search for fixed- and mixed-precision quantization to identify optimal quantization parameters while mitigating the effects of a non-smooth loss landscape. Extensive evaluations across various vision tasks demonstrate the superiority of CLAMP-ViT, with performance improvements of up to 3% in top-1 accuracy for classification, 0.6 mAP for object detection, and 1.5 mIoU for segmentation at similar or better compression ratio over existing alternatives. Code is available at https://github.com/georgia-tech-synergy-lab/CLAMP-ViT.git |
format | Article |
fullrecord | <record><control><sourceid>proquest</sourceid><recordid>TN_cdi_proquest_journals_3077526645</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>3077526645</sourcerecordid><originalsourceid>FETCH-proquest_journals_30775266453</originalsourceid><addsrcrecordid>eNqNzEELgjAYxvERBEn5HQadB2tzGt3Ekg4GBRJ0kheaMYnN9s4OffpE-gCdnsPv4T8jkZByw7aJEAsSI3acc5FmQikZkVtR5aczu5p6RwtngwcM5q3pHgKw0mtNKw3eGvugrfM0v0M_-dlhYLUHM9FlABvMB4JxlrqWjjlckXkLT9Txb5dkXR7q4sh6716DxtB0bvB2pEbyLFMiTRMl_3t9ARERQRA</addsrcrecordid><sourcetype>Aggregation Database</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype><pqid>3077526645</pqid></control><display><type>article</type><title>CLAMP-ViT: Contrastive Data-Free Learning for Adaptive Post-Training Quantization of ViTs</title><source>Free E- Journals</source><creator>Ramachandran, Akshat ; Kundu, Souvik ; Krishna, Tushar</creator><creatorcontrib>Ramachandran, Akshat ; Kundu, Souvik ; Krishna, Tushar</creatorcontrib><description>We present CLAMP-ViT, a data-free post-training quantization method for vision transformers (ViTs). We identify the limitations of recent techniques, notably their inability to leverage meaningful inter-patch relationships, leading to the generation of simplistic and semantically vague data, impacting quantization accuracy. CLAMP-ViT employs a two-stage approach, cyclically adapting between data generation and model quantization. Specifically, we incorporate a patch-level contrastive learning scheme to generate richer, semantically meaningful data. Furthermore, we leverage contrastive learning in layer-wise evolutionary search for fixed- and mixed-precision quantization to identify optimal quantization parameters while mitigating the effects of a non-smooth loss landscape. Extensive evaluations across various vision tasks demonstrate the superiority of CLAMP-ViT, with performance improvements of up to 3% in top-1 accuracy for classification, 0.6 mAP for object detection, and 1.5 mIoU for segmentation at similar or better compression ratio over existing alternatives. Code is available at https://github.com/georgia-tech-synergy-lab/CLAMP-ViT.git</description><identifier>EISSN: 2331-8422</identifier><language>eng</language><publisher>Ithaca: Cornell University Library, arXiv.org</publisher><subject>Compression ratio ; Learning ; Object recognition ; Parameter identification</subject><ispartof>arXiv.org, 2024-09</ispartof><rights>2024. 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>780,784</link.rule.ids></links><search><creatorcontrib>Ramachandran, Akshat</creatorcontrib><creatorcontrib>Kundu, Souvik</creatorcontrib><creatorcontrib>Krishna, Tushar</creatorcontrib><title>CLAMP-ViT: Contrastive Data-Free Learning for Adaptive Post-Training Quantization of ViTs</title><title>arXiv.org</title><description>We present CLAMP-ViT, a data-free post-training quantization method for vision transformers (ViTs). We identify the limitations of recent techniques, notably their inability to leverage meaningful inter-patch relationships, leading to the generation of simplistic and semantically vague data, impacting quantization accuracy. CLAMP-ViT employs a two-stage approach, cyclically adapting between data generation and model quantization. Specifically, we incorporate a patch-level contrastive learning scheme to generate richer, semantically meaningful data. Furthermore, we leverage contrastive learning in layer-wise evolutionary search for fixed- and mixed-precision quantization to identify optimal quantization parameters while mitigating the effects of a non-smooth loss landscape. Extensive evaluations across various vision tasks demonstrate the superiority of CLAMP-ViT, with performance improvements of up to 3% in top-1 accuracy for classification, 0.6 mAP for object detection, and 1.5 mIoU for segmentation at similar or better compression ratio over existing alternatives. Code is available at https://github.com/georgia-tech-synergy-lab/CLAMP-ViT.git</description><subject>Compression ratio</subject><subject>Learning</subject><subject>Object recognition</subject><subject>Parameter identification</subject><issn>2331-8422</issn><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2024</creationdate><recordtype>article</recordtype><sourceid>ABUWG</sourceid><sourceid>AFKRA</sourceid><sourceid>AZQEC</sourceid><sourceid>BENPR</sourceid><sourceid>CCPQU</sourceid><sourceid>DWQXO</sourceid><recordid>eNqNzEELgjAYxvERBEn5HQadB2tzGt3Ekg4GBRJ0kheaMYnN9s4OffpE-gCdnsPv4T8jkZByw7aJEAsSI3acc5FmQikZkVtR5aczu5p6RwtngwcM5q3pHgKw0mtNKw3eGvugrfM0v0M_-dlhYLUHM9FlABvMB4JxlrqWjjlckXkLT9Txb5dkXR7q4sh6716DxtB0bvB2pEbyLFMiTRMl_3t9ARERQRA</recordid><startdate>20240909</startdate><enddate>20240909</enddate><creator>Ramachandran, Akshat</creator><creator>Kundu, Souvik</creator><creator>Krishna, Tushar</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>20240909</creationdate><title>CLAMP-ViT: Contrastive Data-Free Learning for Adaptive Post-Training Quantization of ViTs</title><author>Ramachandran, Akshat ; Kundu, Souvik ; Krishna, Tushar</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-proquest_journals_30775266453</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2024</creationdate><topic>Compression ratio</topic><topic>Learning</topic><topic>Object recognition</topic><topic>Parameter identification</topic><toplevel>online_resources</toplevel><creatorcontrib>Ramachandran, Akshat</creatorcontrib><creatorcontrib>Kundu, Souvik</creatorcontrib><creatorcontrib>Krishna, Tushar</creatorcontrib><collection>ProQuest SciTech Collection</collection><collection>ProQuest Technology Collection</collection><collection>Materials Science & 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>Ramachandran, Akshat</au><au>Kundu, Souvik</au><au>Krishna, Tushar</au><format>book</format><genre>document</genre><ristype>GEN</ristype><atitle>CLAMP-ViT: Contrastive Data-Free Learning for Adaptive Post-Training Quantization of ViTs</atitle><jtitle>arXiv.org</jtitle><date>2024-09-09</date><risdate>2024</risdate><eissn>2331-8422</eissn><abstract>We present CLAMP-ViT, a data-free post-training quantization method for vision transformers (ViTs). We identify the limitations of recent techniques, notably their inability to leverage meaningful inter-patch relationships, leading to the generation of simplistic and semantically vague data, impacting quantization accuracy. CLAMP-ViT employs a two-stage approach, cyclically adapting between data generation and model quantization. Specifically, we incorporate a patch-level contrastive learning scheme to generate richer, semantically meaningful data. Furthermore, we leverage contrastive learning in layer-wise evolutionary search for fixed- and mixed-precision quantization to identify optimal quantization parameters while mitigating the effects of a non-smooth loss landscape. Extensive evaluations across various vision tasks demonstrate the superiority of CLAMP-ViT, with performance improvements of up to 3% in top-1 accuracy for classification, 0.6 mAP for object detection, and 1.5 mIoU for segmentation at similar or better compression ratio over existing alternatives. Code is available at https://github.com/georgia-tech-synergy-lab/CLAMP-ViT.git</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, 2024-09 |
issn | 2331-8422 |
language | eng |
recordid | cdi_proquest_journals_3077526645 |
source | Free E- Journals |
subjects | Compression ratio Learning Object recognition Parameter identification |
title | CLAMP-ViT: Contrastive Data-Free Learning for Adaptive Post-Training Quantization of ViTs |
url | https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-01-07T03%3A52%3A51IST&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=CLAMP-ViT:%20Contrastive%20Data-Free%20Learning%20for%20Adaptive%20Post-Training%20Quantization%20of%20ViTs&rft.jtitle=arXiv.org&rft.au=Ramachandran,%20Akshat&rft.date=2024-09-09&rft.eissn=2331-8422&rft_id=info:doi/&rft_dat=%3Cproquest%3E3077526645%3C/proquest%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_pqid=3077526645&rft_id=info:pmid/&rfr_iscdi=true |