{"id":"https://openalex.org/W4413014966","doi":"https://doi.org/10.1145/3759441.3759448","title":"EMPIRIC: Exploring Missing Pieces in KV Cache Compression for Reducing Computation, Storage, and Latency in Long-Context LLM Inference","display_name":"EMPIRIC: Exploring Missing Pieces in KV Cache Compression for Reducing Computation, Storage, and Latency in Long-Context LLM Inference","publication_year":2025,"publication_date":"2025-08-04","ids":{"openalex":"https://openalex.org/W4413014966","doi":"https://doi.org/10.1145/3759441.3759448"},"language":"en","primary_location":{"id":"doi:10.1145/3759441.3759448","is_oa":false,"landing_page_url":"https://doi.org/10.1145/3759441.3759448","pdf_url":null,"source":{"id":"https://openalex.org/S50071195","display_name":"ACM SIGOPS Operating Systems Review","issn_l":"0163-5980","issn":["0163-5980","1943-586X"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":"https://openalex.org/P4310319798","host_organization_name":"Association for Computing Machinery","host_organization_lineage":["https://openalex.org/P4310319798"],"host_organization_lineage_names":["Association for Computing Machinery"],"type":"journal"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"ACM SIGOPS Operating Systems Review","raw_type":"journal-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":false,"oa_status":"closed","oa_url":null,"any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5032487759","display_name":"Payman Behnam","orcid":"https://orcid.org/0000-0002-3826-9123"},"institutions":[{"id":"https://openalex.org/I1304085615","display_name":"Nvidia (United Kingdom)","ror":"https://ror.org/02kr42612","country_code":"GB","type":"company","lineage":["https://openalex.org/I1304085615","https://openalex.org/I4210127875"]}],"countries":["GB"],"is_corresponding":true,"raw_author_name":"Payman Behnam","raw_affiliation_strings":["NVIDIA, Georgia Tech"],"affiliations":[{"raw_affiliation_string":"NVIDIA, Georgia Tech","institution_ids":["https://openalex.org/I1304085615"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5010195820","display_name":"Yaosheng Fu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yaosheng Fu","raw_affiliation_strings":["NVIDIA"],"affiliations":[{"raw_affiliation_string":"NVIDIA","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5033038201","display_name":"Ritchie Zhao","orcid":"https://orcid.org/0000-0003-1656-9165"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ritchie Zhao","raw_affiliation_strings":["NVIDIA"],"affiliations":[{"raw_affiliation_string":"NVIDIA","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5080155613","display_name":"Po-An Tsai","orcid":"https://orcid.org/0000-0003-4561-6450"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Po-An Tsai","raw_affiliation_strings":["NVIDIA"],"affiliations":[{"raw_affiliation_string":"NVIDIA","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101944084","display_name":"Yu Zheng","orcid":"https://orcid.org/0000-0002-5885-2674"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhididng Yu","raw_affiliation_strings":["NVIDIA"],"affiliations":[{"raw_affiliation_string":"NVIDIA","institution_ids":[]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5048451114","display_name":"Alexey Tumanov","orcid":"https://orcid.org/0009-0005-7862-1477"},"institutions":[{"id":"https://openalex.org/I130701444","display_name":"Georgia Institute of Technology","ror":"https://ror.org/01zkghx44","country_code":"US","type":"education","lineage":["https://openalex.org/I130701444"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Alexey Tumanov","raw_affiliation_strings":["Georgia Tech"],"affiliations":[{"raw_affiliation_string":"Georgia Tech","institution_ids":["https://openalex.org/I130701444"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":6,"corresponding_author_ids":["https://openalex.org/A5032487759"],"corresponding_institution_ids":["https://openalex.org/I1304085615"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.11338471,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":"59","issue":"1","first_page":"46","last_page":"54"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.998199999332428,"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/T10028","display_name":"Topic Modeling","score":0.998199999332428,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.9965999722480774,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.9911999702453613,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.8407274484634399},{"id":"https://openalex.org/keywords/cache","display_name":"Cache","score":0.7913157939910889},{"id":"https://openalex.org/keywords/cache-algorithms","display_name":"Cache algorithms","score":0.586452305316925},{"id":"https://openalex.org/keywords/parallel-computing","display_name":"Parallel computing","score":0.5532578229904175},{"id":"https://openalex.org/keywords/cache-pollution","display_name":"Cache pollution","score":0.5102963447570801},{"id":"https://openalex.org/keywords/cpu-cache","display_name":"CPU cache","score":0.46600040793418884},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.4641934633255005},{"id":"https://openalex.org/keywords/security-token","display_name":"Security token","score":0.4486546218395233},{"id":"https://openalex.org/keywords/cache-coloring","display_name":"Cache coloring","score":0.4421252906322479},{"id":"https://openalex.org/keywords/computer-engineering","display_name":"Computer engineering","score":0.37271279096603394},{"id":"https://openalex.org/keywords/distributed-computing","display_name":"Distributed computing","score":0.331570029258728},{"id":"https://openalex.org/keywords/computer-network","display_name":"Computer network","score":0.16428232192993164},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.11698576807975769}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8407274484634399},{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.7913157939910889},{"id":"https://openalex.org/C38556500","wikidata":"https://www.wikidata.org/wiki/Q13404475","display_name":"Cache algorithms","level":4,"score":0.586452305316925},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.5532578229904175},{"id":"https://openalex.org/C113166858","wikidata":"https://www.wikidata.org/wiki/Q5015981","display_name":"Cache pollution","level":5,"score":0.5102963447570801},{"id":"https://openalex.org/C189783530","wikidata":"https://www.wikidata.org/wiki/Q352090","display_name":"CPU cache","level":3,"score":0.46600040793418884},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.4641934633255005},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.4486546218395233},{"id":"https://openalex.org/C201148951","wikidata":"https://www.wikidata.org/wiki/Q5015976","display_name":"Cache coloring","level":4,"score":0.4421252906322479},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.37271279096603394},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.331570029258728},{"id":"https://openalex.org/C31258907","wikidata":"https://www.wikidata.org/wiki/Q1301371","display_name":"Computer network","level":1,"score":0.16428232192993164},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.11698576807975769}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1145/3759441.3759448","is_oa":false,"landing_page_url":"https://doi.org/10.1145/3759441.3759448","pdf_url":null,"source":{"id":"https://openalex.org/S50071195","display_name":"ACM SIGOPS Operating Systems Review","issn_l":"0163-5980","issn":["0163-5980","1943-586X"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":"https://openalex.org/P4310319798","host_organization_name":"Association for Computing Machinery","host_organization_lineage":["https://openalex.org/P4310319798"],"host_organization_lineage_names":["Association for Computing Machinery"],"type":"journal"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"ACM SIGOPS Operating Systems Review","raw_type":"journal-article"}],"best_oa_location":null,"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":13,"referenced_works":["https://openalex.org/W3184606595","https://openalex.org/W4206289265","https://openalex.org/W4387321091","https://openalex.org/W4388979610","https://openalex.org/W6600384961","https://openalex.org/W6600446476","https://openalex.org/W6600459194","https://openalex.org/W6601144205","https://openalex.org/W6605768063","https://openalex.org/W6629340653","https://openalex.org/W6681176463","https://openalex.org/W6815446887","https://openalex.org/W6838957149"],"related_works":["https://openalex.org/W2734782074","https://openalex.org/W2114386333","https://openalex.org/W2133489088","https://openalex.org/W2012518269","https://openalex.org/W2086718556","https://openalex.org/W2167303720","https://openalex.org/W2066160958","https://openalex.org/W1496086148","https://openalex.org/W2146079099","https://openalex.org/W1860107648"],"abstract_inverted_index":{"Transformer-based":[0],"Large":[1],"Language":[2],"Models":[3],"(LLMs)":[4],"heavily":[5],"depend":[6],"on":[7,71],"the":[8,19,22,28,52,150],"KV":[9,23,53,89,124,132,155,184],"cache":[10,24,54,90,125,133,156,185],"for":[11,118,192],"efficient":[12,183],"handling":[13],"of":[14,21],"long":[15,193],"context":[16,194],"sequences.":[17],"However,":[18],"size":[20],"grows":[25],"linearly":[26],"with":[27],"input":[29],"sequence":[30],"length,":[31],"increasingly":[32],"straining":[33],"system":[34],"memory,":[35],"computational":[36,59,163],"resources,":[37],"bandwidth,":[38],"and":[39,81,121,160,169],"latency":[40],"during":[41,158],"decoding.":[42],"Although":[43],"recent":[44,101],"research":[45,178],"has":[46],"proposed":[47],"various":[48],"techniques":[49],"to":[50,154],"compress":[51],"-targeting":[55],"either":[56],"storage":[57,122,165],"or":[58],"efficiency-few":[60],"methods":[61,68],"effectively":[62],"achieve":[63],"both":[64],"simultaneously.":[65],"Additionally,":[66],"existing":[67],"primarily":[69],"rely":[70],"heuristic-driven":[72],"approaches,":[73],"lacking":[74],"comprehensive":[75],"insights":[76,139],"into":[77,140],"token":[78,91,142],"selection":[79],"criteria,":[80],"often":[82],"significantly":[83,188],"compromise":[84],"model":[85],"accuracy":[86,145],"under":[87],"strict":[88],"budget":[92],"constraints":[93],"(e.g.,":[94],"keeping":[95],"512":[96],"tokens).":[97],"Building":[98],"upon":[99],"our":[100],"work,":[102],"RocketKV,":[103],"this":[104],"paper":[105],"introduces":[106],"EMPIRIC":[107,136,174],"as":[108],"an":[109],"oracle-based":[110],"vision":[111],"study,":[112],"which":[113],"explicitly":[114],"defines":[115],"theoretical":[116],"bounds":[117],"accuracy,":[119],"computation,":[120],"in":[123,131],"compression.":[126],"By":[127],"analyzing":[128],"intrinsic":[129],"patterns":[130],"attention":[134],"heads,":[135],"provides":[137],"novel":[138],"effective":[141],"pruning":[143],"without":[144],"degradation.":[146],"This":[147],"work":[148],"clarifies":[149],"overlooked":[151],"elements":[152],"critical":[153],"compression":[157,186],"decoding":[159],"optimally":[161],"balances":[162],"efficiency,":[164],"optimization,":[166],"inference":[167,190],"latency,":[168],"accuracy.":[170],"We":[171],"envision":[172],"that":[173],"will":[175],"guide":[176],"future":[177],"efforts":[179],"toward":[180],"creating":[181],"scalable,":[182],"techniques,":[187],"improving":[189],"performance":[191],"LLM":[195],"inference.":[196]},"counts_by_year":[],"updated_date":"2025-11-06T03:46:38.306776","created_date":"2025-10-10T00:00:00"}