{"id":"https://openalex.org/W6929482003","doi":"https://doi.org/10.48550/arxiv.2503.04973","title":"Beyond RAG: Task-Aware KV Cache Compression for Comprehensive Knowledge Reasoning","display_name":"Beyond RAG: Task-Aware KV Cache Compression for Comprehensive Knowledge Reasoning","publication_year":2025,"publication_date":"2025-03-06","ids":{"openalex":"https://openalex.org/W6929482003","doi":"https://doi.org/10.48550/arxiv.2503.04973"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2503.04973","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2503.04973","pdf_url":"https://arxiv.org/pdf/2503.04973","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/2503.04973","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5106276521","display_name":"Giulio Corallo","orcid":"https://orcid.org/0009-0009-5561-5479","observed_orcids":["https://orcid.org/0009-0009-5561-5479"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Corallo, Giulio","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5056096860","display_name":"Orion Weller","orcid":"https://orcid.org/0000-0003-4148-5430","observed_orcids":["https://orcid.org/0000-0003-4148-5430"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Weller, Orion","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5093091829","display_name":"Fabio Petroni","orcid":"https://orcid.org/0009-0005-5996-9830","observed_orcids":["https://orcid.org/0009-0005-5996-9830"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Petroni, Fabio","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5011336242","display_name":"Paolo Papotti","orcid":"https://orcid.org/0000-0003-0651-4128","observed_orcids":["https://orcid.org/0000-0003-0651-4128"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Papotti, Paolo","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":true,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.9634000062942505,"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":0.9634000062942505,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.015200000256299973,"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/T10286","display_name":"Information Retrieval and Search Behavior","score":0.011500000022351742,"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/key-value-cache-compression","display_name":"key-value cache compression","score":0.8999999761581421},{"id":"https://openalex.org/keywords/knowledge-based-reasoning","display_name":"knowledge reasoning","score":0.5099999904632568},{"id":"https://openalex.org/keywords/retrieval-augmented-generation","display_name":"retrieval-augmented generation","score":0.9049999713897705},{"id":"https://openalex.org/keywords/external-knowledge","display_name":"external knowledge","score":0.3959999978542328},{"id":"https://openalex.org/keywords/long-context-models","display_name":"long-context models","score":0.6539999842643738},{"id":"https://openalex.org/keywords/inference-latency","display_name":"inference latency","score":0.5559999942779541}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8285999894142151},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.7106999754905701},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.5580000281333923},{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.5519000291824341},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.5428000092506409},{"id":"https://openalex.org/C180016635","wikidata":"https://www.wikidata.org/wiki/Q2712821","display_name":"Compression (physics)","level":2,"score":0.5045999884605408},{"id":"https://openalex.org/C161301231","wikidata":"https://www.wikidata.org/wiki/Q3478658","display_name":"Knowledge representation and reasoning","level":2,"score":0.4747999906539917},{"id":"https://openalex.org/C82876162","wikidata":"https://www.wikidata.org/wiki/Q17096504","display_name":"Latency (audio)","level":2,"score":0.4722999930381775},{"id":"https://openalex.org/C98045186","wikidata":"https://www.wikidata.org/wiki/Q205663","display_name":"Process (computing)","level":2,"score":0.4683000147342682},{"id":"https://openalex.org/C78548338","wikidata":"https://www.wikidata.org/wiki/Q2493","display_name":"Data compression","level":2,"score":0.4652999937534332},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.43860000371932983},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4138999879360199},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4004000127315521},{"id":"https://openalex.org/C2776359362","wikidata":"https://www.wikidata.org/wiki/Q2145286","display_name":"Representation (politics)","level":3,"score":0.396699994802475},{"id":"https://openalex.org/C103278499","wikidata":"https://www.wikidata.org/wiki/Q254465","display_name":"Similarity (geometry)","level":3,"score":0.38690000772476196},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.3610999882221222},{"id":"https://openalex.org/C115925183","wikidata":"https://www.wikidata.org/wiki/Q1412694","display_name":"Knowledge-based systems","level":2,"score":0.323199987411499},{"id":"https://openalex.org/C23123220","wikidata":"https://www.wikidata.org/wiki/Q816826","display_name":"Information retrieval","level":1,"score":0.2962999939918518},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.29260000586509705},{"id":"https://openalex.org/C2778751112","wikidata":"https://www.wikidata.org/wiki/Q835016","display_name":"Window (computing)","level":2,"score":0.2802000045776367},{"id":"https://openalex.org/C189783530","wikidata":"https://www.wikidata.org/wiki/Q352090","display_name":"CPU cache","level":3,"score":0.27799999713897705},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.27090001106262207},{"id":"https://openalex.org/C120567893","wikidata":"https://www.wikidata.org/wiki/Q1582085","display_name":"Knowledge extraction","level":2,"score":0.2703000009059906},{"id":"https://openalex.org/C195344581","wikidata":"https://www.wikidata.org/wiki/Q2555318","display_name":"Automated reasoning","level":2,"score":0.2583000063896179},{"id":"https://openalex.org/C197914299","wikidata":"https://www.wikidata.org/wiki/Q18650","display_name":"Semantic memory","level":3,"score":0.2563000023365021},{"id":"https://openalex.org/C207685749","wikidata":"https://www.wikidata.org/wiki/Q2088941","display_name":"Domain knowledge","level":2,"score":0.2554999887943268}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2503.04973","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2503.04973","pdf_url":"https://arxiv.org/pdf/2503.04973","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.2503.04973","doi":"https://doi.org/10.48550/arxiv.2503.04973","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2503.04973","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2503.04973","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2503.04973","pdf_url":"https://arxiv.org/pdf/2503.04973","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":[],"sustainable_development_goals_aurora":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Incorporating":[0],"external":[1,70],"knowledge":[2,71,149],"in":[3,72],"large":[4],"language":[5],"models":[6,37],"(LLMs)":[7],"enhances":[8],"their":[9],"utility":[10],"across":[11],"diverse":[12],"applications,":[13],"but":[14,27,42],"existing":[15],"methods":[16],"have":[17],"trade-offs.":[18],"Retrieval-Augmented":[19],"Generation":[20],"(RAG)":[21],"fetches":[22],"evidence":[23,140],"via":[24],"similarity":[25],"search,":[26],"key":[28],"information":[29],"may":[30],"fall":[31],"outside":[32],"top":[33],"ranked":[34],"results.":[35],"Long-context":[36],"can":[38],"process":[39],"multiple":[40],"documents":[41],"are":[43],"computationally":[44],"expensive":[45],"and":[46,99],"limited":[47],"by":[48,53,109],"context":[49],"window":[50],"size.":[51],"Inspired":[52],"students":[54],"condensing":[55],"study":[56],"material":[57],"for":[58,147],"open-book":[59],"exams,":[60],"we":[61],"propose":[62],"task-aware":[63,143],"key-value":[64],"(KV)":[65],"cache":[66],"compression,":[67],"which":[68],"compresses":[69],"a":[73,85,118],"zero-":[74],"or":[75],"few-shot":[76],"setup.":[77],"This":[78],"enables":[79],"LLMs":[80],"to":[81,111,128],"reason":[82],"efficiently":[83],"over":[84,115],"compacted":[86],"representation":[87],"of":[88],"all":[89],"relevant":[90],"information.":[91],"Experiments":[92],"show":[93],"our":[94],"approach":[95],"outperforms":[96],"both":[97],"RAG":[98,116,135],"task-agnostic":[100],"compression":[101,120,144],"methods.":[102],"On":[103],"LongBench":[104],"v2,":[105],"it":[106],"improves":[107],"accuracy":[108],"up":[110],"7":[112],"absolute":[113],"points":[114],"with":[117],"30x":[119],"rate,":[121],"while":[122],"reducing":[123],"inference":[124],"latency":[125],"from":[126],"0.43s":[127],"0.16s.":[129],"A":[130],"synthetic":[131],"dataset":[132],"highlights":[133],"that":[134],"performs":[136],"well":[137],"when":[138],"sparse":[139],"suffices,":[141],"whereas":[142],"is":[145],"superior":[146],"broad":[148],"tasks.":[150]},"counts_by_year":[],"updated_date":"2026-10-12T09:40:04.860316","created_date":"2025-10-10T00:00:00"}