{"id":"https://openalex.org/W4416034963","doi":"https://doi.org/10.18653/v1/2025.findings-emnlp.400","title":"Rotate, Clip, and Partition: Towards W2A4KV4 Quantization by Integrating Rotation and Learnable Non-uniform Quantizer","display_name":"Rotate, Clip, and Partition: Towards W2A4KV4 Quantization by Integrating Rotation and Learnable Non-uniform Quantizer","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4416034963","doi":"https://doi.org/10.18653/v1/2025.findings-emnlp.400"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2025.findings-emnlp.400","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-emnlp.400","pdf_url":"https://aclanthology.org/2025.findings-emnlp.400.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: EMNLP 2025","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2025.findings-emnlp.400.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5020621473","display_name":"Euntae Choi","orcid":null},"institutions":[{"id":"https://openalex.org/I139264467","display_name":"Seoul National University","ror":"https://ror.org/04h9pn542","country_code":"KR","type":"education","lineage":["https://openalex.org/I139264467"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Euntae Choi","raw_affiliation_strings":["Seoul National University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Seoul National University","institution_ids":["https://openalex.org/I139264467"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5111112619","display_name":"Sumin Song","orcid":null},"institutions":[{"id":"https://openalex.org/I139264467","display_name":"Seoul National University","ror":"https://ror.org/04h9pn542","country_code":"KR","type":"education","lineage":["https://openalex.org/I139264467"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Sumin Song","raw_affiliation_strings":["Seoul National University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Seoul National University","institution_ids":["https://openalex.org/I139264467"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5089920562","display_name":"Woosang Lim","orcid":"https://orcid.org/0000-0002-6615-5801"},"institutions":[{"id":"https://openalex.org/I139264467","display_name":"Seoul National University","ror":"https://ror.org/04h9pn542","country_code":"KR","type":"education","lineage":["https://openalex.org/I139264467"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Woosang Lim","raw_affiliation_strings":["Seoul National University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Seoul National University","institution_ids":["https://openalex.org/I139264467"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5063521444","display_name":"Sungjoo Yoo","orcid":"https://orcid.org/0000-0002-5853-0675"},"institutions":[{"id":"https://openalex.org/I139264467","display_name":"Seoul National University","ror":"https://ror.org/04h9pn542","country_code":"KR","type":"education","lineage":["https://openalex.org/I139264467"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Sungjoo Yoo","raw_affiliation_strings":["Seoul National University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Seoul National University","institution_ids":["https://openalex.org/I139264467"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I139264467"],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":1,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"7568","last_page":"7590"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10857","display_name":"Advanced Electron Microscopy Techniques and Applications","score":0.21439999341964722,"subfield":{"id":"https://openalex.org/subfields/1315","display_name":"Structural Biology"},"field":{"id":"https://openalex.org/fields/13","display_name":"Biochemistry, Genetics and Molecular Biology"},"domain":{"id":"https://openalex.org/domains/1","display_name":"Life Sciences"}},"topics":[{"id":"https://openalex.org/T10857","display_name":"Advanced Electron Microscopy Techniques and Applications","score":0.21439999341964722,"subfield":{"id":"https://openalex.org/subfields/1315","display_name":"Structural Biology"},"field":{"id":"https://openalex.org/fields/13","display_name":"Biochemistry, Genetics and Molecular Biology"},"domain":{"id":"https://openalex.org/domains/1","display_name":"Life Sciences"}},{"id":"https://openalex.org/T11948","display_name":"Machine Learning in Materials Science","score":0.13979999721050262,"subfield":{"id":"https://openalex.org/subfields/2505","display_name":"Materials Chemistry"},"field":{"id":"https://openalex.org/fields/25","display_name":"Materials Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10657","display_name":"Topological Materials and Phenomena","score":0.0746999979019165,"subfield":{"id":"https://openalex.org/subfields/3107","display_name":"Atomic and Molecular Physics, and Optics"},"field":{"id":"https://openalex.org/fields/31","display_name":"Physics and Astronomy"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/quantization","display_name":"Quantization (signal processing)","score":0.5393999814987183},{"id":"https://openalex.org/keywords/rotation","display_name":"Rotation (mathematics)","score":0.3937000036239624},{"id":"https://openalex.org/keywords/signal-processing","display_name":"Signal processing","score":0.260699987411499},{"id":"https://openalex.org/keywords/context","display_name":"Context (archaeology)","score":0.24580000340938568}],"concepts":[{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.5393999814987183},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.45410001277923584},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.4345000088214874},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.42800000309944153},{"id":"https://openalex.org/C74050887","wikidata":"https://www.wikidata.org/wiki/Q848368","display_name":"Rotation (mathematics)","level":2,"score":0.3937000036239624},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3151000142097473},{"id":"https://openalex.org/C104267543","wikidata":"https://www.wikidata.org/wiki/Q208163","display_name":"Signal processing","level":3,"score":0.260699987411499},{"id":"https://openalex.org/C28826006","wikidata":"https://www.wikidata.org/wiki/Q33521","display_name":"Applied mathematics","level":1,"score":0.25060001015663147},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.24580000340938568},{"id":"https://openalex.org/C78548338","wikidata":"https://www.wikidata.org/wiki/Q2493","display_name":"Data compression","level":2,"score":0.23880000412464142}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2025.findings-emnlp.400","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-emnlp.400","pdf_url":"https://aclanthology.org/2025.findings-emnlp.400.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: EMNLP 2025","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2025.findings-emnlp.400","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-emnlp.400","pdf_url":"https://aclanthology.org/2025.findings-emnlp.400.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: EMNLP 2025","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[{"id":"https://openalex.org/G269467657","display_name":null,"funder_award_id":"NO.RS-2021-II211343","funder_id":"https://openalex.org/F4320321292","funder_display_name":"Seoul National University"},{"id":"https://openalex.org/G3598079552","display_name":null,"funder_award_id":"RS-2021-II211343","funder_id":"https://openalex.org/F4320335489","funder_display_name":"Institute for Information and Communications Technology Promotion"},{"id":"https://openalex.org/G7288789209","display_name":null,"funder_award_id":"RS-2021-II211343","funder_id":"https://openalex.org/F4320328359","funder_display_name":"Ministry of Science and ICT, South Korea"},{"id":"https://openalex.org/G7798134731","display_name":null,"funder_award_id":"RS-2021-II211343","funder_id":"https://openalex.org/F4320321292","funder_display_name":"Seoul National University"}],"funders":[{"id":"https://openalex.org/F4320315121","display_name":"Samsung Advanced Institute of Technology","ror":null},{"id":"https://openalex.org/F4320321292","display_name":"Seoul National University","ror":"https://ror.org/04h9pn542"},{"id":"https://openalex.org/F4320328359","display_name":"Ministry of Science and ICT, South Korea","ror":"https://ror.org/01wpjm123"},{"id":"https://openalex.org/F4320332195","display_name":"Samsung","ror":"https://ror.org/04w3jy968"},{"id":"https://openalex.org/F4320335489","display_name":"Institute for Information and Communications Technology Promotion","ror":"https://ror.org/01g0hqq23"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4416034963.pdf","grobid_xml":"https://content.openalex.org/works/W4416034963.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"We":[0,144,219,265],"propose":[1],"Rotate,":[2],"Clip,":[3],"and":[4,25,42,103,116,119,129,150,169,210,222,232,255,275],"Partition":[5],"(RCP),":[6],"a":[7,34,75,96,184,199,246,259],"Quantization-Aware":[8],"Training":[9],"(QAT)":[10],"approach":[11],"that":[12,88,161,249],"first":[13,280],"realizes":[14],"extreme":[15,236],"compression":[16],"of":[17,47,52,85,98,186],"LLMs":[18],"with":[19,33,70,95,121,198,229],"W2A4KV4":[20,94,237,274],"(2-bit":[21],"weight,":[22,147],"4-bit":[23,26,173],"activation,":[24,149,174],"KV-cache)":[27],"configuration.RCP":[28],"integrates":[29],"recent":[30],"rotation":[31,48,227,254],"techniques":[32],"novel":[35,260],"non-uniform":[36,67,81,167,263],"weight":[37,53,55,230],"quantizer":[38],"design":[39,176],"by":[40],"theoretically":[41],"empirically":[43],"analyzing":[44],"the":[45,50,191,203,251,279],"impact":[46],"on":[49,80,225,235],"non-uniformity":[51],"distribution.Our":[54],"quantizer,":[56],"Learnable":[57],"Direct":[58],"Partitioning":[59],"(LDP),":[60],"introduces":[61],"learnable":[62,262],"parameters":[63],"to":[64,93,195,269],"directly":[65],"learn":[66],"intervals":[68],"jointly":[69],"LLM":[71,286],"weights.We":[72],"also":[73],"present":[74],"GPU":[76],"kernel":[77,180,188],"supporting":[78],"GEMV":[79,179],"W2A4":[82],"as":[83,126,183,217],"proof":[84,185],"concept.Experiments":[86],"show":[87,270],"RCP":[89,109,271],"can":[90,110,189],"compress":[91],"LLaMA-2-7B":[92],"loss":[97],"only":[99],"2.84":[100],"WikiText2":[101],"PPL":[102],"5.29":[104,196],"times":[105,197],"reduced":[106],"memory":[107,192],"footprint.Furthermore,":[108],"quantize":[111],"challenging":[112],"mobile-targeted":[113],"LLaMA-3.2":[114],"models":[115],"domain-specific":[117],"WizardCoder-7B":[118],"MetaMath-7B":[120],"no":[122,158],"critical":[123],"problems":[124],"such":[125],"convergence":[127],"failure":[128],"repetition.Code":[130],"is":[131,157],"available":[132,159],"at":[133],"https://github.com/":[134],"songsm921/RCP.":[135],"*":[136],"indicates":[137,140],"equal":[138],"contribution":[139],"corresponding":[141],"author":[142],"1":[143],"call":[145],"l-bit":[146],"m-bit":[148],"n-bit":[151],"KV-cache":[152],"WlAmKVn":[153],"like":[154],"W2A4KV4.Since":[155],"there":[156],"hardware":[160],"supports":[162],"both":[163],"LUT":[164],"inference":[165],"for":[166,172,278,285],"quantization":[168,247,277],"specialized":[170],"acceleration":[171],"we":[175,243],"an":[177],"accelerated":[178],"in":[181],"CUDA":[182],"concept.Our":[187],"reduce":[190],"footprint":[193],"up":[194],"latency":[200],"lower":[201],"than":[202],"FP16":[204],"PyTorch":[205],"(Paszke":[206],"et":[207],"al.,":[208],"2019)":[209],"INT4":[211],"QuaRot":[212],"implementation.Our":[213],"contributions":[214],"are":[215],"summarized":[216],"follows:":[218],"provide":[220,266],"empirical":[221],"theoretical":[223],"analysis":[224],"how":[226],"interacts":[228],"distribution":[231],"poses":[233],"difficulty":[234],"quantization.":[238],"To":[239],"address":[240],"this":[241],"issue,":[242],"introduce":[244],"RCP,":[245],"algorithm":[248],"takes":[250],"best":[252],"from":[253],"QAT":[256],"via":[257],"LDP,":[258],"fully":[261],"quantizer.":[264],"extensive":[267],"experiments":[268],"achieves":[272],"state-of-the-art":[273],"W3A4KV4":[276],"time.":[281],"Preliminaries2.1":[282],"Random":[283],"Rotation":[284],"Quantization":[287]},"counts_by_year":[{"year":2026,"cited_by_count":1}],"updated_date":"2026-07-14T23:27:15.235271","created_date":"2025-11-08T00:00:00"}
