{"id":"https://openalex.org/W7165737623","doi":"https://doi.org/10.48550/arxiv.2606.24467","title":"CompressKV: Semantic-Retrieval-Guided KV-Cache Compression for Resource-Efficient Long-Context LLM Inference","display_name":"CompressKV: Semantic-Retrieval-Guided KV-Cache Compression for Resource-Efficient Long-Context LLM Inference","publication_year":2026,"publication_date":"2026-06-23","ids":{"openalex":"https://openalex.org/W7165737623","doi":"https://doi.org/10.48550/arxiv.2606.24467"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.24467","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.24467","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2606.24467","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5101425768","display_name":"Xiaolin Lin","orcid":"https://orcid.org/0000-0002-2484-2814"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lin, Xiaolin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5033598547","display_name":"Jingcun Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Jingcun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139294216","display_name":"Olga Kondrateva","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kondrateva, Olga","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100902783","display_name":"Y Shi","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shi, Yiyu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139284668","display_name":"Bing Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Bing","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5070194930","display_name":"Grace Li Zhang","orcid":"https://orcid.org/0000-0002-8289-9288"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Grace Li","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.17509999871253967,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"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/T10181","display_name":"Natural Language Processing Techniques","score":0.17509999871253967,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"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/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.16990000009536743,"subfield":{"id":"https://openalex.org/subfields/1708","display_name":"Hardware and Architecture"},"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/T14347","display_name":"Big Data and Digital Economy","score":0.07930000126361847,"subfield":{"id":"https://openalex.org/subfields/1710","display_name":"Information Systems"},"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/cache","display_name":"Cache","score":0.7366999983787537},{"id":"https://openalex.org/keywords/security-token","display_name":"Security token","score":0.5916000008583069},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.5863000154495239},{"id":"https://openalex.org/keywords/eviction","display_name":"Eviction","score":0.5220999717712402},{"id":"https://openalex.org/keywords/decoding-methods","display_name":"Decoding methods","score":0.5020999908447266},{"id":"https://openalex.org/keywords/limiting","display_name":"Limiting","score":0.4616999924182892},{"id":"https://openalex.org/keywords/heuristic","display_name":"Heuristic","score":0.4368000030517578},{"id":"https://openalex.org/keywords/margin","display_name":"Margin (machine learning)","score":0.4341999888420105},{"id":"https://openalex.org/keywords/code","display_name":"Code (set theory)","score":0.4250999987125397}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8098999857902527},{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.7366999983787537},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.5916000008583069},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5863000154495239},{"id":"https://openalex.org/C2778227907","wikidata":"https://www.wikidata.org/wiki/Q1893186","display_name":"Eviction","level":2,"score":0.5220999717712402},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.5020999908447266},{"id":"https://openalex.org/C188198153","wikidata":"https://www.wikidata.org/wiki/Q1613840","display_name":"Limiting","level":2,"score":0.4616999924182892},{"id":"https://openalex.org/C173801870","wikidata":"https://www.wikidata.org/wiki/Q201413","display_name":"Heuristic","level":2,"score":0.4368000030517578},{"id":"https://openalex.org/C774472","wikidata":"https://www.wikidata.org/wiki/Q6760393","display_name":"Margin (machine learning)","level":2,"score":0.4341999888420105},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.4250999987125397},{"id":"https://openalex.org/C189783530","wikidata":"https://www.wikidata.org/wiki/Q352090","display_name":"CPU cache","level":3,"score":0.38029998540878296},{"id":"https://openalex.org/C74912251","wikidata":"https://www.wikidata.org/wiki/Q6815727","display_name":"Memory footprint","level":2,"score":0.36980000138282776},{"id":"https://openalex.org/C38556500","wikidata":"https://www.wikidata.org/wiki/Q13404475","display_name":"Cache algorithms","level":4,"score":0.3634999990463257},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3488999903202057},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.3474999964237213},{"id":"https://openalex.org/C184337299","wikidata":"https://www.wikidata.org/wiki/Q1437428","display_name":"Semantics (computer science)","level":2,"score":0.3345000147819519},{"id":"https://openalex.org/C78548338","wikidata":"https://www.wikidata.org/wiki/Q2493","display_name":"Data compression","level":2,"score":0.3319000005722046},{"id":"https://openalex.org/C2983355114","wikidata":"https://www.wikidata.org/wiki/Q161380","display_name":"Credit card","level":3,"score":0.3310999870300293},{"id":"https://openalex.org/C105339364","wikidata":"https://www.wikidata.org/wiki/Q2297740","display_name":"Software deployment","level":2,"score":0.33059999346733093},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.3260999917984009},{"id":"https://openalex.org/C180016635","wikidata":"https://www.wikidata.org/wiki/Q2712821","display_name":"Compression (physics)","level":2,"score":0.30399999022483826},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.2976999878883362},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.28439998626708984},{"id":"https://openalex.org/C79403827","wikidata":"https://www.wikidata.org/wiki/Q3988","display_name":"Real-time computing","level":1,"score":0.28380000591278076},{"id":"https://openalex.org/C77618280","wikidata":"https://www.wikidata.org/wiki/Q1155772","display_name":"Scheme (mathematics)","level":2,"score":0.2777999937534332},{"id":"https://openalex.org/C2779960059","wikidata":"https://www.wikidata.org/wiki/Q7113681","display_name":"Overhead (engineering)","level":2,"score":0.26010000705718994},{"id":"https://openalex.org/C132943942","wikidata":"https://www.wikidata.org/wiki/Q2562511","display_name":"Footprint","level":2,"score":0.2574999928474426},{"id":"https://openalex.org/C142210648","wikidata":"https://www.wikidata.org/wiki/Q1266546","display_name":"Record linkage","level":3,"score":0.25679999589920044}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.24467","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.24467","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2606.24467","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.24467","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[{"id":"https://metadata.un.org/sdg/12","display_name":"Responsible consumption and production","score":0.41175445914268494}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Long-context":[0],"large":[1],"language":[2],"model":[3],"(LLM)":[4],"inference":[5],"is":[6,197],"increasingly":[7],"constrained":[8],"by":[9],"the":[10,45,53,61,97,166],"memory":[11,152],"footprint":[12],"and":[13,58,99,105,110,140,173],"decoding":[14],"cost":[15],"of":[16,48,55,63,81,102,133,159,165],"key-value":[17],"(KV)":[18],"caches,":[19],"limiting":[20],"sustainable":[21],"deployment":[22],"on":[23,138,169,182],"resource-constrained":[24],"hardware.":[25],"Existing":[26],"KV":[27,117,167,180],"cache":[28,125,168],"eviction":[29,54,135,149],"methods":[30,43,150],"typically":[31],"apply":[32],"heuristic":[33],"token":[34],"scoring":[35],"over":[36,157],"all":[37,86],"heads":[38],"in":[39],"GQA-based":[40,78],"LLMs.":[41,64,79],"These":[42,184],"ignore":[44],"different":[46],"functionalities":[47],"attention":[49,83],"heads,":[50,87],"leading":[51],"to":[52,113,130],"critical":[56],"tokens":[57,101,115],"thus":[59],"degrading":[60],"performance":[62,161],"To":[65],"address":[66],"this":[67],"issue,":[68],"we":[69],"propose":[70],"CompressKV,":[71],"a":[72,103],"resource-efficient":[73],"KV-cache":[74,148],"compression":[75],"framework":[76],"for":[77,191],"Instead":[80],"aggregating":[82],"scores":[84],"from":[85],"CompressKV":[88,123,144],"identifies":[89],"Semantic":[90],"Retrieval":[91],"Heads":[92],"(SRHs)":[93],"that":[94,143],"capture":[95],"both":[96],"initial":[98],"final":[100],"prompt":[104],"semantically":[106],"important":[107],"mid-context":[108],"evidence,":[109],"uses":[111],"them":[112],"select":[114],"whose":[116],"pairs":[118],"should":[119],"be":[120],"retained.":[121],"Furthermore,":[122],"allocates":[124],"budgets":[126],"across":[127,151],"layers":[128],"according":[129],"offline":[131],"estimates":[132],"layer-wise":[134],"error.":[136],"Experiments":[137],"LongBench":[139,170],"Needle-in-a-Haystack":[141],"show":[142],"consistently":[145],"outperforms":[146],"existing":[147],"budgets.":[153],"Notably,":[154],"it":[155],"preserves":[156],"97\\%":[158],"full-cache":[160],"using":[162],"only":[163],"3\\%":[164],"question-answering":[171],"tasks":[172],"achieves":[174],"90\\%":[175],"accuracy":[176],"with":[177],"just":[178],"0.7\\%":[179],"storage":[181],"Needle-in-a-Haystack.":[183],"results":[185],"demonstrate":[186],"an":[187],"improved":[188],"resource--performance":[189],"trade-off":[190],"long-context":[192],"LLM":[193],"inference.":[194],"Our":[195],"code":[196],"publicly":[198],"available":[199],"at:":[200],"https://github.com/TUDa-HWAI/CompressKV":[201]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-06-25T00:00:00"}