{"id":"https://openalex.org/W7204900938","doi":"https://doi.org/10.48550/arxiv.2608.30295","title":"CateKV: On Sequential Consistency for Long-Context LLM Inference Acceleration","display_name":"CateKV: On Sequential Consistency for Long-Context LLM Inference Acceleration","publication_year":2026,"publication_date":"2026-08-31","ids":{"openalex":"https://openalex.org/W7204900938","doi":"https://doi.org/10.48550/arxiv.2608.30295"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2608.30295","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.30295","pdf_url":"https://arxiv.org/pdf/2608.30295","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2608.30295","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5012162299","display_name":"Haoyun Jiang","orcid":"https://orcid.org/0000-0002-4499-2043","observed_orcids":["https://orcid.org/0000-0002-4499-2043"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jiang, Haoyun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5148912573","display_name":"Haolin Li","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Haolin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5148927841","display_name":"Jianwei Zhang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Jianwei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138946909","display_name":"Fei Huang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Huang, Fei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5134055157","display_name":"Qiang Hu","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hu, Qiang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5132577067","display_name":"Minmin Sun","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sun, Minmin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5148916718","display_name":"Shuai Xiao","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xiao, Shuai","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5148858536","display_name":"Yong Li","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Yong","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5124806358","display_name":"Junyang Lin","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lin, Junyang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5148900137","display_name":"Jiangchao Yao","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yao, Jiangchao","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":0.3546999990940094,"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":0.3546999990940094,"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/T10028","display_name":"Topic Modeling","score":0.30880001187324524,"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":0.14659999310970306,"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/large-language-models","display_name":"large language models","score":0.777999997138977},{"id":"https://openalex.org/keywords/long-context-inference","display_name":"long-context inference","score":0.9679999947547913},{"id":"https://openalex.org/keywords/sequential-consistency","display_name":"sequential consistency","score":0.6579999923706055},{"id":"https://openalex.org/keywords/attention-heads","display_name":"attention heads","score":0.9459999799728394},{"id":"https://openalex.org/keywords/coefficient-of-variation","display_name":"coefficient of variation","score":0.9739999771118164},{"id":"https://openalex.org/keywords/memory-efficiency","display_name":"memory efficiency","score":0.7599999904632568},{"id":"https://openalex.org/keywords/decoding-acceleration","display_name":"decoding acceleration","score":0.7900000214576721}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7967000007629395},{"id":"https://openalex.org/C2776436953","wikidata":"https://www.wikidata.org/wiki/Q5163215","display_name":"Consistency (knowledge bases)","level":2,"score":0.6729000210762024},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.6557999849319458},{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.6104000210762024},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.5475999712944031},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.4973999857902527},{"id":"https://openalex.org/C117896860","wikidata":"https://www.wikidata.org/wiki/Q11376","display_name":"Acceleration","level":2,"score":0.477400004863739},{"id":"https://openalex.org/C157764524","wikidata":"https://www.wikidata.org/wiki/Q1383412","display_name":"Throughput","level":3,"score":0.4577000141143799},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.4302999973297119},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.4034000039100647},{"id":"https://openalex.org/C37279795","wikidata":"https://www.wikidata.org/wiki/Q2492305","display_name":"Consistency model","level":3,"score":0.39160001277923584},{"id":"https://openalex.org/C2778029271","wikidata":"https://www.wikidata.org/wiki/Q5421931","display_name":"Extension (predicate logic)","level":2,"score":0.3813999891281128},{"id":"https://openalex.org/C31388003","wikidata":"https://www.wikidata.org/wiki/Q7624548","display_name":"Strong consistency","level":3,"score":0.3433000147342682},{"id":"https://openalex.org/C189783530","wikidata":"https://www.wikidata.org/wiki/Q352090","display_name":"CPU cache","level":3,"score":0.3384999930858612},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.28610000014305115},{"id":"https://openalex.org/C141917322","wikidata":"https://www.wikidata.org/wiki/Q1025017","display_name":"Cache coherence","level":5,"score":0.28540000319480896},{"id":"https://openalex.org/C80478641","wikidata":"https://www.wikidata.org/wiki/Q195771","display_name":"Sequential analysis","level":2,"score":0.26980000734329224},{"id":"https://openalex.org/C86426650","wikidata":"https://www.wikidata.org/wiki/Q7452504","display_name":"Sequential estimation","level":2,"score":0.26589998602867126},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.26100000739097595},{"id":"https://openalex.org/C2777472644","wikidata":"https://www.wikidata.org/wiki/Q16968992","display_name":"Approximate inference","level":3,"score":0.2549999952316284},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.2547000050544739}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2608.30295","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.30295","pdf_url":"https://arxiv.org/pdf/2608.30295","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2608.30295","doi":"https://doi.org/10.48550/arxiv.2608.30295","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2608.30295","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2608.30295","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.30295","pdf_url":"https://arxiv.org/pdf/2608.30295","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4403571350","https://openalex.org/W4416866664","https://openalex.org/W4415101136","https://openalex.org/W7208771792","https://openalex.org/W7127907339","https://openalex.org/W4416035984","https://openalex.org/W7128525388","https://openalex.org/W4415068608","https://openalex.org/W7162449816","https://openalex.org/W7128489630"],"abstract_inverted_index":{"Large":[0],"language":[1],"models":[2],"(LLMs)":[3],"have":[4],"demonstrated":[5],"strong":[6],"capabilities":[7],"in":[8,40,89,138,146],"handling":[9],"long-context":[10,114],"tasks,":[11],"but":[12],"processing":[13],"such":[14],"long":[15],"contexts":[16],"remains":[17],"challenging":[18],"due":[19],"to":[20,92,122,131],"the":[21,84,98],"substantial":[22],"memory":[23,127],"requirements":[24],"and":[25,79,104,133,141],"inference":[26],"latency.":[27],"In":[28],"this":[29,55],"work,":[30],"we":[31,57],"discover":[32],"that":[33,65],"certain":[34],"attention":[35,42],"heads":[36,91],"exhibit":[37],"sequential":[38],"consistency":[39],"their":[41],"patterns,":[43],"which":[44],"can":[45],"be":[46],"persistently":[47],"identified":[48],"using":[49],"a":[50,60],"coefficient-of-variation-based":[51],"algorithm.":[52],"Inspired":[53],"by":[54,129,136,144],"observation,":[56],"propose":[58],"CateKV,":[59],"hybrid":[61],"KV":[62,76,87],"cache":[63,77],"method":[64],"retains":[66],"only":[67],"critical":[68],"token":[69],"information":[70],"for":[71],"consistent":[72],"heads,":[73],"thereby":[74],"reducing":[75],"size":[78],"computational":[80],"overhead,":[81],"while":[82,118],"preserving":[83],"majority":[85],"of":[86,101],"pairs":[88],"adaptive":[90],"ensure":[93],"high":[94],"accuracy.":[95],"We":[96],"show":[97,116],"unique":[99],"characteristics":[100],"our":[102],"algorithm":[103],"its":[105],"extension":[106],"with":[107],"existing":[108],"acceleration":[109],"methods.":[110],"Comprehensive":[111],"evaluations":[112],"on":[113],"benchmarks":[115],"that,":[117],"maintaining":[119],"accuracy":[120],"comparable":[121],"full":[123],"attention,":[124],"CateKV":[125],"reduces":[126],"usage":[128],"up":[130],"$2.72\\times$":[132],"accelerates":[134],"decoding":[135],"$2.18\\times$":[137],"single-sample":[139],"inputs,":[140],"boosts":[142],"throughput":[143],"$3.96\\times$":[145],"batch":[147],"scenarios.":[148]},"counts_by_year":[],"updated_date":"2026-10-28T10:33:24.094632","created_date":"2026-09-02T00:00:00"}