{"id":"https://openalex.org/W7214786013","doi":"https://doi.org/10.48550/arxiv.2609.35139","title":"CacheRepair: Learning to Repair Cross-Chunk Context in RAG for KV Cache Fusion","display_name":"CacheRepair: Learning to Repair Cross-Chunk Context in RAG for KV Cache Fusion","publication_year":2026,"publication_date":"2026-09-28","ids":{"openalex":"https://openalex.org/W7214786013","doi":"https://doi.org/10.48550/arxiv.2609.35139"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2609.35139","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.35139","pdf_url":"https://arxiv.org/pdf/2609.35139","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/2609.35139","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5027086602","display_name":"Genglin Wang","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I177725633","display_name":"Chinese University of Hong Kong","ror":"https://ror.org/00t33hh48","country_code":"HK","type":"education","lineage":["https://openalex.org/I177725633"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Wang, Genglin","raw_affiliation_strings":["The Chinese University of Hong Kong"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"The Chinese University of Hong Kong","institution_ids":["https://openalex.org/I177725633"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101293258","display_name":"Wangsong Yin","orcid":"https://orcid.org/0009-0000-6242-4368","observed_orcids":["https://orcid.org/0009-0000-6242-4368"]},"institutions":[{"id":"https://openalex.org/I20231570","display_name":"Peking University","ror":"https://ror.org/02v51f717","country_code":"CN","type":"education","lineage":["https://openalex.org/I20231570"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yin, Wangsong","raw_affiliation_strings":["Peking University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Peking University","institution_ids":["https://openalex.org/I20231570"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5119935458","display_name":"Yeerzhati Abudunuer","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I177725633","display_name":"Chinese University of Hong Kong","ror":"https://ror.org/00t33hh48","country_code":"HK","type":"education","lineage":["https://openalex.org/I177725633"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Abudunuer, Yeerzhati","raw_affiliation_strings":["The Chinese University of Hong Kong"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"The Chinese University of Hong Kong","institution_ids":["https://openalex.org/I177725633"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5153528216","display_name":"Haoxuan Xu","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I200769079","display_name":"Hong Kong University of Science and Technology","ror":"https://ror.org/00q4vv597","country_code":"HK","type":"education","lineage":["https://openalex.org/I200769079"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Xu, Haoxuan","raw_affiliation_strings":["The Hong Kong University of Science and Technology"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"The Hong Kong University of Science and Technology","institution_ids":["https://openalex.org/I200769079"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5072780737","display_name":"Guoliang Xing","orcid":"https://orcid.org/0000-0003-1772-7751","observed_orcids":["https://orcid.org/0000-0003-1772-7751"]},"institutions":[{"id":"https://openalex.org/I177725633","display_name":"Chinese University of Hong Kong","ror":"https://ror.org/00t33hh48","country_code":"HK","type":"education","lineage":["https://openalex.org/I177725633"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Xing, Guoliang","raw_affiliation_strings":["The Chinese University of Hong Kong"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"The Chinese University of Hong Kong","institution_ids":["https://openalex.org/I177725633"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5153417564","display_name":"Zhenyu Yan","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I177725633","display_name":"Chinese University of Hong Kong","ror":"https://ror.org/00t33hh48","country_code":"HK","type":"education","lineage":["https://openalex.org/I177725633"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Yan, Zhenyu","raw_affiliation_strings":["The Chinese University of Hong Kong"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"The Chinese University of Hong Kong","institution_ids":["https://openalex.org/I177725633"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":3,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"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/T10028","display_name":"Topic Modeling","score":0.8838000297546387,"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":0.8838000297546387,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.05979999899864197,"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/T10286","display_name":"Information Retrieval and Search Behavior","score":0.02889999933540821,"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/retrieval-augmented-generation","display_name":"retrieval-augmented generation","score":0.8920000195503235},{"id":"https://openalex.org/keywords/time-to-first-token","display_name":"time-to-first-token","score":0.7139999866485596},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.5400000214576721}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.862500011920929},{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.7336000204086304},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.6617000102996826},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.6557999849319458},{"id":"https://openalex.org/C98045186","wikidata":"https://www.wikidata.org/wiki/Q205663","display_name":"Process (computing)","level":2,"score":0.4876999855041504},{"id":"https://openalex.org/C2777210771","wikidata":"https://www.wikidata.org/wiki/Q4927124","display_name":"Block (permutation group theory)","level":2,"score":0.4690000116825104},{"id":"https://openalex.org/C155512373","wikidata":"https://www.wikidata.org/wiki/Q287450","display_name":"Residual","level":2,"score":0.36160001158714294},{"id":"https://openalex.org/C2776175482","wikidata":"https://www.wikidata.org/wiki/Q1195816","display_name":"Transfer (computing)","level":2,"score":0.3379000127315521},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.33239999413490295},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.32989999651908875},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.32850000262260437},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.31610000133514404},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.31040000915527344},{"id":"https://openalex.org/C23123220","wikidata":"https://www.wikidata.org/wiki/Q816826","display_name":"Information retrieval","level":1,"score":0.2912999987602234},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.28700000047683716},{"id":"https://openalex.org/C189783530","wikidata":"https://www.wikidata.org/wiki/Q352090","display_name":"CPU cache","level":3,"score":0.27239999175071716},{"id":"https://openalex.org/C103278499","wikidata":"https://www.wikidata.org/wiki/Q254465","display_name":"Similarity (geometry)","level":3,"score":0.2712000012397766},{"id":"https://openalex.org/C2776207758","wikidata":"https://www.wikidata.org/wiki/Q5303302","display_name":"Downstream (manufacturing)","level":2,"score":0.2556000053882599},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.25519999861717224}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2609.35139","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.35139","pdf_url":"https://arxiv.org/pdf/2609.35139","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.2609.35139","doi":"https://doi.org/10.48550/arxiv.2609.35139","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2609.35139","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:2609.35139","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.35139","pdf_url":"https://arxiv.org/pdf/2609.35139","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":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7214786013.pdf","grobid_xml":"https://content.openalex.org/works/W7214786013.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4411403197","https://openalex.org/W7140127569","https://openalex.org/W7090856496","https://openalex.org/W7125280553","https://openalex.org/W7171386993","https://openalex.org/W7127607560","https://openalex.org/W7214770294","https://openalex.org/W7163638869","https://openalex.org/W4416037017","https://openalex.org/W7140110628"],"abstract_inverted_index":{"Multi-document":[0],"retrieval-augmented":[1],"generation":[2],"(RAG)":[3],"requires":[4],"a":[5,16,71,141,147],"language":[6],"model":[7],"to":[8,113,130,197],"process":[9],"multiple":[10],"retrieved":[11,32],"text":[12],"chunks":[13,30,89],"before":[14],"answering":[15],"question.":[17],"Precomputing":[18],"each":[19,107],"chunk's":[20],"KV":[21,81,95,162],"cache":[22,40,122,203,233],"independently":[23,79],"and":[24,83,100,109,124,151,168,177,205,223],"concatenating":[25],"the":[26,29,38,52,58,76,88,120,125,184,211],"caches":[27,82],"when":[28],"are":[31],"can":[33],"accelerate":[34],"this":[35],"step.":[36],"However,":[37],"assembled":[39],"lacks":[41],"cross-chunk":[42,54],"attention":[43,102],"information,":[44],"reducing":[45],"answer":[46],"quality.":[47],"Selective":[48],"recomputation":[49],"methods":[50],"recover":[51],"missing":[53],"context":[55],"by":[56,86,227],"rerunning":[57],"target":[59,144,175],"LLM":[60,145],"on":[61,146,183],"selected":[62],"tokens,":[63],"incurring":[64],"substantial":[65],"online":[66,202],"computation.":[67],"We":[68],"introduce":[69],"CacheRepair,":[70],"lightweight":[72],"network":[73,92,137],"that":[74,103,159],"learns":[75],"difference":[77],"between":[78],"computed":[80],"those":[84],"produced":[85],"processing":[87],"together.":[90],"The":[91],"combines":[93],"compressed":[94,121],"features":[96],"with":[97],"token":[98,199],"embeddings":[99],"uses":[101],"is":[104,128,138],"bidirectional":[105],"within":[106],"chunk":[108,166,170],"flows":[110],"from":[111],"earlier":[112],"later":[114],"chunks.":[115],"Each":[116,135],"repair":[117,136,160],"block":[118],"receives":[119],"features,":[123],"predicted":[126],"residual":[127],"added":[129],"every":[131],"document":[132],"token's":[133],"cache.":[134],"trained":[139],"for":[140],"specific":[142],"frozen":[143],"generic":[148],"retrieval":[149],"corpus":[150],"reused":[152],"across":[153,173],"downstream":[154,179],"datasets.":[155],"Our":[156],"analysis":[157],"shows":[158],"reduces":[161],"errors":[163],"both":[164],"near":[165],"boundaries":[167],"throughout":[169],"interiors.":[171],"Evaluation":[172],"three":[174],"LLMs":[176],"four":[178],"datasets":[180],"places":[181],"CacheRepair":[182],"measured":[185],"answer-quality-latency":[186],"Pareto":[187],"frontier":[188],"in":[189,217],"eleven":[190],"of":[191],"twelve":[192,209],"model-dataset":[193],"combinations.":[194],"Reported":[195],"time":[196],"first":[198],"(TTFT)":[200],"includes":[201],"transfer":[204],"repair.":[206],"Across":[207],"all":[208],"combinations,":[210],"largest":[212],"repairers":[213],"achieve":[214],"1.69-4.61$\\times$":[215],"speedups":[216],"median":[218],"TTFT":[219],"over":[220,231],"full":[221],"prefill":[222],"improve":[224],"mean":[225],"F1":[226],"2.1-26.1":[228],"percentage":[229],"points":[230],"direct":[232],"reuse.":[234]},"counts_by_year":[],"updated_date":"2026-10-28T10:33:24.094632","created_date":"2026-09-30T00:00:00"}