{"id":"https://openalex.org/W7225423382","doi":"https://doi.org/10.5281/zenodo.23997983","title":"Predictive KV Cache Preparation for Generative AI: Anticipating and Precomputing Reusable Transformer States Before Future Requests Arrive","display_name":"Predictive KV Cache Preparation for Generative AI: Anticipating and Precomputing Reusable Transformer States Before Future Requests Arrive","publication_year":2026,"publication_date":"2026-11-06","ids":{"openalex":"https://openalex.org/W7225423382","doi":"https://doi.org/10.5281/zenodo.23997983"},"language":"en","primary_location":{"id":"doi:10.5281/zenodo.23997983","doi":"https://doi.org/10.5281/zenodo.23997983","is_oa":true,"landing_page_url":"https://doi.org/10.5281/zenodo.23997983","pdf_url":null,"source":{"id":"https://openalex.org/S4306400562","display_name":"Zenodo (CERN European Organization for Nuclear Research)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I67311998","host_organization_name":"European Organization for Nuclear Research","host_organization_lineage":["https://openalex.org/I67311998"],"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"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.5281/zenodo.23997983","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5100638045","display_name":"Jincheng Zhang","orcid":"https://orcid.org/0000-0001-7332-6704","observed_orcids":["https://orcid.org/0000-0001-7332-6704","https://orcid.org/0009-0005-1860-0009"]},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Zhang, Jincheng","raw_affiliation_strings":[],"raw_orcid":"https://orcid.org/0009-0005-1860-0009","affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":["https://openalex.org/A5100638045"],"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":0.329800009727478,"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":0.329800009727478,"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":0.22910000383853912,"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":0.18160000443458557,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/kv-cache-management","display_name":"KV cache management","score":0.7580000162124634},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.675000011920929},{"id":"https://openalex.org/keywords/cache-prefetching","display_name":"cache prefetching","score":0.6869999766349792},{"id":"https://openalex.org/keywords/retrieval-augmented-generation","display_name":"retrieval-augmented generation","score":0.8820000290870667},{"id":"https://openalex.org/keywords/optimization-model","display_name":"optimization model","score":0.46399998664855957},{"id":"https://openalex.org/keywords/time-to-first-token","display_name":"time-to-first-token","score":0.6660000085830688},{"id":"https://openalex.org/keywords/memory-usage","display_name":"memory usage","score":0.5299999713897705},{"id":"https://openalex.org/keywords/energy-consumption","display_name":"energy consumption","score":0.7379999756813049},{"id":"https://openalex.org/keywords/autoregressive-generation","display_name":"autoregressive generation","score":0.574999988079071}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8456000089645386},{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.7781000137329102},{"id":"https://openalex.org/C38556500","wikidata":"https://www.wikidata.org/wiki/Q13404475","display_name":"Cache algorithms","level":4,"score":0.5011000037193298},{"id":"https://openalex.org/C206588197","wikidata":"https://www.wikidata.org/wiki/Q846574","display_name":"Reuse","level":2,"score":0.483599990606308},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.4641000032424927},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.4081999957561493},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.39410001039505005},{"id":"https://openalex.org/C141603448","wikidata":"https://www.wikidata.org/wiki/Q134830","display_name":"Prefix","level":2,"score":0.3327000141143799},{"id":"https://openalex.org/C25536678","wikidata":"https://www.wikidata.org/wiki/Q5015977","display_name":"Cache invalidation","level":5,"score":0.2930999994277954},{"id":"https://openalex.org/C59687516","wikidata":"https://www.wikidata.org/wiki/Q5015938","display_name":"Cache-oblivious algorithm","level":5,"score":0.2865999937057495},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.2833999991416931},{"id":"https://openalex.org/C167713795","wikidata":"https://www.wikidata.org/wiki/Q352090","display_name":"Smart Cache","level":5,"score":0.2791000008583069},{"id":"https://openalex.org/C189783530","wikidata":"https://www.wikidata.org/wiki/Q352090","display_name":"CPU cache","level":3,"score":0.27630001306533813},{"id":"https://openalex.org/C201148951","wikidata":"https://www.wikidata.org/wiki/Q5015976","display_name":"Cache coloring","level":4,"score":0.2596000134944916}],"mesh":[],"locations_count":2,"locations":[{"id":"doi:10.5281/zenodo.23997983","doi":"https://doi.org/10.5281/zenodo.23997983","is_oa":true,"landing_page_url":"https://doi.org/10.5281/zenodo.23997983","pdf_url":null,"source":{"id":"https://openalex.org/S4306400562","display_name":"Zenodo (CERN European Organization for Nuclear Research)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I67311998","host_organization_name":"European Organization for Nuclear Research","host_organization_lineage":["https://openalex.org/I67311998"],"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"},{"id":"doi:10.5281/zenodo.23997984","doi":"https://doi.org/10.5281/zenodo.23997984","is_oa":true,"landing_page_url":"https://doi.org/10.5281/zenodo.23997984","pdf_url":null,"source":{"id":"https://openalex.org/S4306400562","display_name":"Zenodo (CERN European Organization for Nuclear Research)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I67311998","host_organization_name":"European Organization for Nuclear Research","host_organization_lineage":["https://openalex.org/I67311998"],"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":"doi:10.5281/zenodo.23997983","doi":"https://doi.org/10.5281/zenodo.23997983","is_oa":true,"landing_page_url":"https://doi.org/10.5281/zenodo.23997983","pdf_url":null,"source":{"id":"https://openalex.org/S4306400562","display_name":"Zenodo (CERN European Organization for Nuclear Research)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I67311998","host_organization_name":"European Organization for Nuclear Research","host_organization_lineage":["https://openalex.org/I67311998"],"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"},"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/W7225384377","https://openalex.org/W7225392952","https://openalex.org/W7225275234","https://openalex.org/W7225391670","https://openalex.org/W7109118627","https://openalex.org/W7225375420","https://openalex.org/W4392867129","https://openalex.org/W7225267314","https://openalex.org/W7225332613","https://openalex.org/W4400222625"],"abstract_inverted_index":{"The":[0,170,193,262,393],"increasing":[1],"deployment":[2],"of":[3,17,234,411,418],"large":[4],"language":[5],"models":[6],"(LLMs)":[7],"has":[8],"made":[9],"Key-Value":[10],"(KV)":[11],"cache":[12,61,113,188,204,215,219,253,296,346,354,358,364,384],"management":[13,62],"a":[14,107,186,322],"critical":[15],"component":[16],"generative":[18],"AI":[19],"inference":[20,75,127,439],"infrastructure.":[21],"During":[22],"transformer":[23,122,446],"inference,":[24],"the":[25,230,409,415,424,444],"prompt":[26,139],"prefill":[27,176,245,371],"stage":[28],"computes":[29],"key":[30],"and":[31,54,115,119,155,182,217,259,275,312,356,389,423],"value":[32,233],"representations":[33],"for":[34,109,133,190,241],"input":[35],"tokens,":[36,307],"which":[37],"are":[38,165],"subsequently":[39],"retained":[40],"to":[41,137,160,167,228,342,378,426,441,449],"support":[42,56],"autoregressive":[43],"generation.":[44],"Although":[45],"KV":[46,80,103,112,180,208,237,295,345,400],"caching":[47],"reduces":[48],"redundant":[49],"computation":[50,246],"during":[51],"token":[52,162,380],"generation":[53,249,278],"can":[55,332],"reuse":[57,243,368],"across":[58],"requests,":[59],"conventional":[60,349],"generally":[63],"depends":[64,407],"on":[65,408],"computational":[66,123,232,335,391,416,429],"states":[67,81,124,238,447],"that":[68,164,287,328,398,451],"have":[69,96],"already":[70],"been":[71,97],"generated":[72],"by":[73],"previous":[74],"operations.":[76],"Consequently,":[77],"potentially":[78],"reusable":[79,121],"may":[82],"remain":[83],"unavailable":[84],"when":[85,90],"future":[86,111,157,203,242,412,438],"requests":[87,128,291],"arrive,":[88],"even":[89],"their":[91],"associated":[92],"contextual":[93,150],"requirements":[94,114],"could":[95],"anticipated.":[98],"This":[99,431],"paper":[100],"introduces":[101],"Predictive":[102],"Cache":[104],"Preparation":[105],"(PKVCP),":[106],"framework":[108,171,263,394],"predicting":[110],"proactively":[116,442],"generating,":[117],"storing,":[118],"managing":[120],"before":[125],"corresponding":[126],"materialize.":[129],"Rather":[130],"than":[131,326],"waiting":[132],"an":[134],"incoming":[135],"request":[136,144],"trigger":[138],"prefill,":[140],"PKVCP":[141,317],"analyzes":[142],"historical":[143],"patterns,":[145],"emerging":[146],"information":[147],"trends,":[148],"recurring":[149],"structures,":[151],"anticipated":[152,276],"application":[153],"workloads,":[154],"predicted":[156,367],"query":[158],"distributions":[159],"identify":[161],"prefixes":[163],"likely":[166],"be":[168],"reused.":[169],"then":[172],"selectively":[173],"executes":[174],"advance":[175,248,399],"operations,":[177],"generates":[178],"compatible":[179],"representations,":[181],"stores":[183],"them":[184],"in":[185,302],"managed":[187],"hierarchy":[189],"subsequent":[191],"retrieval.":[192],"proposed":[194,341],"architecture":[195],"integrates":[196],"six":[197],"principal":[198],"functions:":[199],"predictive":[200,344,434],"prefix":[201,306,351,363],"identification,":[202],"demand":[205,440],"estimation,":[206],"compatibility-aware":[207],"state":[209,273],"generation,":[210],"cost-sensitive":[211],"preparation":[212,347,401],"scheduling,":[213],"hierarchical":[214],"placement,":[216],"adaptive":[218],"lifecycle":[220],"management.":[221],"A":[222],"mathematical":[223],"optimization":[224],"model":[225,303],"is":[226,286,340,402],"developed":[227],"evaluate":[229],"expected":[231],"preparing":[235,443],"individual":[236],"while":[239],"accounting":[240],"probability,":[244],"avoided,":[247],"expenditure,":[250],"memory":[251,382,421],"occupancy,":[252],"transfer":[254,385],"latency,":[255],"storage":[256],"costs,":[257],"expiration,":[258],"compatibility":[260,319],"uncertainty.":[261],"further":[264],"distinguishes":[265],"between":[266,290],"complete-prefix":[267],"preparation,":[268,270,274,419],"shared-prefix":[269],"multi-turn":[271],"conversational":[272],"retrieval-augmented":[277],"(RAG)":[279],"context":[280],"preparation.":[281],"An":[282,337],"important":[283],"design":[284],"principle":[285],"semantic":[288],"similarity":[289],"does":[292,395],"not":[293,396],"establish":[294],"compatibility.":[297,430],"Reuse":[298],"requires":[299],"appropriate":[300],"consistency":[301],"parameters,":[304],"tokenization,":[305],"positional":[308],"configuration,":[309],"attention":[310],"implementation,":[311],"relevant":[313],"execution":[314,375],"settings.":[315],"Accordingly,":[316],"treats":[318],"verification":[320],"as":[321],"mandatory":[323],"condition":[324],"rather":[325],"assuming":[327],"semantically":[329],"related":[330],"questions":[331],"share":[333],"arbitrary":[334],"states.":[336],"experimental":[338],"methodology":[339],"compare":[343],"against":[348],"reactive":[350],"caching,":[352],"historical-frequency-based":[353],"retention,":[355],"workload-aware":[357],"prefetching.":[359],"Evaluation":[360],"metrics":[361],"include":[362],"hit":[365],"rate,":[366],"accuracy,":[369],"avoided":[370],"computation,":[372],"total":[373,390],"GPU":[374],"time,":[376],"time":[377],"first":[379],"(TTFT),":[381],"utilization,":[383],"overhead,":[386],"energy":[387],"consumption,":[388],"expenditure.":[392],"assume":[397],"universally":[403],"beneficial.":[404],"Its":[405],"effectiveness":[406],"predictability":[410],"token-prefix":[413],"reuse,":[414],"cost":[417],"available":[420],"capacity,":[422],"ability":[425],"maintain":[427],"exact":[428],"research":[432],"extends":[433],"computing":[435],"from":[436],"anticipating":[437],"intermediate":[445],"required":[448],"satisfy":[450],"demand.":[452]},"counts_by_year":[],"updated_date":"2026-11-08T12:12:28.789561","created_date":"2026-11-07T00:00:00"}