{"id":"https://openalex.org/W4416034352","doi":"https://doi.org/10.18653/v1/2025.findings-emnlp.1009","title":"CARVQ: Corrective Adaptor with Group Residual Vector Quantization for LLM Embedding Compression","display_name":"CARVQ: Corrective Adaptor with Group Residual Vector Quantization for LLM Embedding Compression","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4416034352","doi":"https://doi.org/10.18653/v1/2025.findings-emnlp.1009"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2025.findings-emnlp.1009","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-emnlp.1009","pdf_url":"https://aclanthology.org/2025.findings-emnlp.1009.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":"article","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2025.findings-emnlp.1009.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5120308562","display_name":"Dayin Gou","orcid":null},"institutions":[{"id":"https://openalex.org/I2818286","display_name":"LG (United States)","ror":"https://ror.org/02b948n83","country_code":"US","type":"company","lineage":["https://openalex.org/I2818286","https://openalex.org/I4210131320"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Dayin Gou","raw_affiliation_strings":["LG Electronics USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"LG Electronics USA","institution_ids":["https://openalex.org/I2818286"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5030744163","display_name":"Sanghyun Byun","orcid":null},"institutions":[{"id":"https://openalex.org/I2818286","display_name":"LG (United States)","ror":"https://ror.org/02b948n83","country_code":"US","type":"company","lineage":["https://openalex.org/I2818286","https://openalex.org/I4210131320"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Sanghyun Byun","raw_affiliation_strings":["LG Electronics USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"LG Electronics USA","institution_ids":["https://openalex.org/I2818286"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5120308563","display_name":"Nilesh Malpeddi","orcid":null},"institutions":[{"id":"https://openalex.org/I2818286","display_name":"LG (United States)","ror":"https://ror.org/02b948n83","country_code":"US","type":"company","lineage":["https://openalex.org/I2818286","https://openalex.org/I4210131320"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Nilesh Malpeddi","raw_affiliation_strings":["LG Electronics USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"LG Electronics USA","institution_ids":["https://openalex.org/I2818286"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5120308564","display_name":"Gabrielle De Micheli","orcid":null},"institutions":[{"id":"https://openalex.org/I2818286","display_name":"LG (United States)","ror":"https://ror.org/02b948n83","country_code":"US","type":"company","lineage":["https://openalex.org/I2818286","https://openalex.org/I4210131320"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Gabrielle De Micheli","raw_affiliation_strings":["LG Electronics USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"LG Electronics USA","institution_ids":["https://openalex.org/I2818286"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5120308565","display_name":"Prathamesh Vaste","orcid":null},"institutions":[{"id":"https://openalex.org/I2818286","display_name":"LG (United States)","ror":"https://ror.org/02b948n83","country_code":"US","type":"company","lineage":["https://openalex.org/I2818286","https://openalex.org/I4210131320"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Prathamesh Vaste","raw_affiliation_strings":["LG Electronics USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"LG Electronics USA","institution_ids":["https://openalex.org/I2818286"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5110418450","display_name":"Jilai Song","orcid":"https://orcid.org/0009-0009-4069-7897"},"institutions":[{"id":"https://openalex.org/I2818286","display_name":"LG (United States)","ror":"https://ror.org/02b948n83","country_code":"US","type":"company","lineage":["https://openalex.org/I2818286","https://openalex.org/I4210131320"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Jacob Song","raw_affiliation_strings":["LG Electronics USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"LG Electronics USA","institution_ids":["https://openalex.org/I2818286"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5111961290","display_name":"W.C. Chung","orcid":"https://orcid.org/0000-0002-6287-8025"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Woo Seong Chung","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":7,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.28865918,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"18594","last_page":"18604"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10901","display_name":"Advanced Data Compression Techniques","score":0.5187000036239624,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T10901","display_name":"Advanced Data Compression Techniques","score":0.5187000036239624,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T11233","display_name":"Advanced Adaptive Filtering Techniques","score":0.06809999793767929,"subfield":{"id":"https://openalex.org/subfields/2206","display_name":"Computational Mechanics"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10860","display_name":"Speech and Audio Processing","score":0.03739999979734421,"subfield":{"id":"https://openalex.org/subfields/1711","display_name":"Signal Processing"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/embedding","display_name":"Embedding","score":0.5460000038146973},{"id":"https://openalex.org/keywords/residual","display_name":"Residual","score":0.5008999705314636},{"id":"https://openalex.org/keywords/vector-quantization","display_name":"Vector quantization","score":0.4925000071525574},{"id":"https://openalex.org/keywords/compression","display_name":"Compression (physics)","score":0.4674000144004822},{"id":"https://openalex.org/keywords/quantization","display_name":"Quantization (signal processing)","score":0.44040000438690186},{"id":"https://openalex.org/keywords/group","display_name":"Group (periodic table)","score":0.39809998869895935}],"concepts":[{"id":"https://openalex.org/C41608201","wikidata":"https://www.wikidata.org/wiki/Q980509","display_name":"Embedding","level":2,"score":0.5460000038146973},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.5333999991416931},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5072000026702881},{"id":"https://openalex.org/C155512373","wikidata":"https://www.wikidata.org/wiki/Q287450","display_name":"Residual","level":2,"score":0.5008999705314636},{"id":"https://openalex.org/C199833920","wikidata":"https://www.wikidata.org/wiki/Q612536","display_name":"Vector quantization","level":2,"score":0.4925000071525574},{"id":"https://openalex.org/C180016635","wikidata":"https://www.wikidata.org/wiki/Q2712821","display_name":"Compression (physics)","level":2,"score":0.4674000144004822},{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.44040000438690186},{"id":"https://openalex.org/C2781311116","wikidata":"https://www.wikidata.org/wiki/Q83306","display_name":"Group (periodic table)","level":2,"score":0.39809998869895935},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.3953999876976013},{"id":"https://openalex.org/C78548338","wikidata":"https://www.wikidata.org/wiki/Q2493","display_name":"Data compression","level":2,"score":0.353300005197525},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3212999999523163},{"id":"https://openalex.org/C47446073","wikidata":"https://www.wikidata.org/wiki/Q5165890","display_name":"Control theory (sociology)","level":3,"score":0.2824999988079071},{"id":"https://openalex.org/C184720557","wikidata":"https://www.wikidata.org/wiki/Q7825049","display_name":"Topology (electrical circuits)","level":2,"score":0.2615000009536743}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2025.findings-emnlp.1009","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-emnlp.1009","pdf_url":"https://aclanthology.org/2025.findings-emnlp.1009.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.1009","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-emnlp.1009","pdf_url":"https://aclanthology.org/2025.findings-emnlp.1009.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":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4416034352.pdf","grobid_xml":"https://content.openalex.org/works/W4416034352.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Large":[0],"Language":[1],"Models":[2],"(LLMs)":[3],"typically":[4],"rely":[5],"on":[6,26,70,102,111,185],"a":[7,58,142,176],"large":[8],"number":[9],"of":[10,73,183],"parameters":[11],"for":[12],"token":[13],"embedding,":[14],"leading":[15],"to":[16,85,87,95,137,165],"substantial":[17],"storage":[18],"requirements":[19],"and":[20,31,76,79,108,116,134,154],"memory":[21,34,46,164,169],"footprints.In":[22],"particular,":[23],"LLMs":[24,104,184],"deployed":[25],"edge":[27,186],"devices":[28],"are":[29],"memory-bound,":[30],"reducing":[32],"the":[33,38,45,71,81,167,180],"footprint":[35,170],"by":[36],"compressing":[37],"embedding":[39,84],"layer":[40],"not":[41],"only":[42],"frees":[43],"up":[44,51],"bandwidth":[47],"but":[48],"also":[49],"speeds":[50],"inference.To":[52],"address":[53],"this,":[54],"we":[55],"introduce":[56],"CARVQ,":[57],"post-training":[59],"novel":[60,143],"Corrective":[61],"Adaptor":[62],"combined":[63],"with":[64,149],"group":[65],"Residual":[66],"Vector":[67],"Quantization.CARVQ":[68],"relies":[69],"composition":[72],"both":[74],"linear":[75],"non-linear":[77],"maps":[78],"mimics":[80],"original":[82],"model":[83],"compress":[86],"approximately":[88],"1.6":[89],"bits":[90],"without":[91],"requiring":[92],"specialized":[93],"hardware":[94,161],"support":[96],"lower-bit":[97],"storage.We":[98],"test":[99],"our":[100],"method":[101],"pre-trained":[103],"such":[105],"as":[106],"LLaMA-3.2-1B,LLaMA-3.2-3B,LLaMA-3.2-3B-Instruct,LLaMA-3.1-8B,Qwen2.5-7B,Qwen2.5-Math-7B":[107],"Phi-4,":[109],"evaluating":[110],"common":[112],"generative,":[113],"discriminative,":[114],"math":[115],"reasoning":[117],"tasks.We":[118],"show":[119],"that":[120,146],"in":[121,171],"most":[122],"cases,":[123],"CARVQ":[124],"can":[125,155],"achieve":[126],"lower":[127],"average":[128],"bitwidth-per-parameter":[129],"while":[130],"maintaining":[131],"reasonable":[132],"perplexity":[133],"accuracy":[135],"compared":[136],"scalar":[138],"quantization.Our":[139],"contributions":[140],"include":[141],"compression":[144],"technique":[145],"is":[147],"compatible":[148],"state-of-the-art":[150],"transformer":[151],"quantization":[152],"methods":[153],"be":[156],"seamlessly":[157],"integrated":[158],"into":[159],"any":[160],"supporting":[162],"4-bit":[163],"reduce":[166],"model's":[168],"memoryconstrained":[172],"devices.This":[173],"work":[174],"demonstrates":[175],"crucial":[177],"step":[178],"toward":[179],"efficient":[181],"deployment":[182],"devices.":[187]},"counts_by_year":[],"updated_date":"2026-06-11T09:08:48.828518","created_date":"2025-11-08T00:00:00"}