{"id":"https://openalex.org/W4404609179","doi":"https://doi.org/10.1162/tacl_a_00716","title":"FINCH: Prompt-guided Key-Value Cache Compression for Large Language Models","display_name":"FINCH: Prompt-guided Key-Value Cache Compression for Large Language Models","publication_year":2024,"publication_date":"2024-01-01","ids":{"openalex":"https://openalex.org/W4404609179","doi":"https://doi.org/10.1162/tacl_a_00716"},"language":"en","primary_location":{"id":"doi:10.1162/tacl_a_00716","is_oa":true,"landing_page_url":"https://doi.org/10.1162/tacl_a_00716","pdf_url":null,"source":{"id":"https://openalex.org/S2729999759","display_name":"Transactions of the Association for Computational Linguistics","issn_l":"2307-387X","issn":["2307-387X"],"is_oa":true,"is_in_doaj":true,"is_core":true,"host_organization":"https://openalex.org/P4310320244","host_organization_name":"Association for Computational Linguistics","host_organization_lineage":["https://openalex.org/P4310320244"],"host_organization_lineage_names":["Association for Computational Linguistics"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Transactions of the Association for Computational Linguistics","raw_type":"journal-article"},"type":"article","indexed_in":["crossref","doaj"],"open_access":{"is_oa":true,"oa_status":"diamond","oa_url":"https://doi.org/10.1162/tacl_a_00716","any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5106276521","display_name":"Giulio Corallo","orcid":null},"institutions":[{"id":"https://openalex.org/I1902872","display_name":"EURECOM","ror":"https://ror.org/00sse7z02","country_code":"FR","type":"education","lineage":["https://openalex.org/I1902872","https://openalex.org/I205703379"]}],"countries":["FR"],"is_corresponding":false,"raw_author_name":"Giulio Corallo","raw_affiliation_strings":["SAP Labs, France, EURECOM, France. giulio.corallo@sap.com"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"SAP Labs, France, EURECOM, France. giulio.corallo@sap.com","institution_ids":["https://openalex.org/I1902872"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5011336242","display_name":"Paolo Papotti","orcid":"https://orcid.org/0000-0003-0651-4128"},"institutions":[{"id":"https://openalex.org/I1902872","display_name":"EURECOM","ror":"https://ror.org/00sse7z02","country_code":"FR","type":"education","lineage":["https://openalex.org/I1902872","https://openalex.org/I205703379"]}],"countries":["FR"],"is_corresponding":false,"raw_author_name":"Paolo Papotti","raw_affiliation_strings":["EURECOM, France. papotti@eurecom.fr"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"EURECOM, France. papotti@eurecom.fr","institution_ids":["https://openalex.org/I1902872"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I1902872"],"apc_list":null,"apc_paid":null,"fwci":0.9737,"has_fulltext":false,"cited_by_count":4,"citation_normalized_percentile":{"value":0.80179273,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":94,"max":97},"biblio":{"volume":"12","issue":null,"first_page":"1517","last_page":"1532"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.9994999766349792,"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":0.9994999766349792,"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":0.9976999759674072,"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/T11269","display_name":"Algorithms and Data Compression","score":0.9945999979972839,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.8598847389221191},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.7161858677864075},{"id":"https://openalex.org/keywords/cache","display_name":"Cache","score":0.6415673494338989},{"id":"https://openalex.org/keywords/value","display_name":"Value (mathematics)","score":0.49733975529670715},{"id":"https://openalex.org/keywords/finch","display_name":"Finch","score":0.4873422086238861},{"id":"https://openalex.org/keywords/compression","display_name":"Compression (physics)","score":0.42284300923347473},{"id":"https://openalex.org/keywords/computer-security","display_name":"Computer security","score":0.1971752643585205},{"id":"https://openalex.org/keywords/parallel-computing","display_name":"Parallel computing","score":0.1773618459701538},{"id":"https://openalex.org/keywords/machine-learning","display_name":"Machine learning","score":0.11148366332054138},{"id":"https://openalex.org/keywords/geology","display_name":"Geology","score":0.08635491132736206},{"id":"https://openalex.org/keywords/paleontology","display_name":"Paleontology","score":0.07325991988182068}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8598847389221191},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.7161858677864075},{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.6415673494338989},{"id":"https://openalex.org/C2776291640","wikidata":"https://www.wikidata.org/wiki/Q2912517","display_name":"Value (mathematics)","level":2,"score":0.49733975529670715},{"id":"https://openalex.org/C2781352930","wikidata":"https://www.wikidata.org/wiki/Q160835","display_name":"Finch","level":2,"score":0.4873422086238861},{"id":"https://openalex.org/C180016635","wikidata":"https://www.wikidata.org/wiki/Q2712821","display_name":"Compression (physics)","level":2,"score":0.42284300923347473},{"id":"https://openalex.org/C38652104","wikidata":"https://www.wikidata.org/wiki/Q3510521","display_name":"Computer security","level":1,"score":0.1971752643585205},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.1773618459701538},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.11148366332054138},{"id":"https://openalex.org/C127313418","wikidata":"https://www.wikidata.org/wiki/Q1069","display_name":"Geology","level":0,"score":0.08635491132736206},{"id":"https://openalex.org/C151730666","wikidata":"https://www.wikidata.org/wiki/Q7205","display_name":"Paleontology","level":1,"score":0.07325991988182068},{"id":"https://openalex.org/C192562407","wikidata":"https://www.wikidata.org/wiki/Q228736","display_name":"Materials science","level":0,"score":0.0},{"id":"https://openalex.org/C159985019","wikidata":"https://www.wikidata.org/wiki/Q181790","display_name":"Composite material","level":1,"score":0.0}],"mesh":[],"locations_count":2,"locations":[{"id":"doi:10.1162/tacl_a_00716","is_oa":true,"landing_page_url":"https://doi.org/10.1162/tacl_a_00716","pdf_url":null,"source":{"id":"https://openalex.org/S2729999759","display_name":"Transactions of the Association for Computational Linguistics","issn_l":"2307-387X","issn":["2307-387X"],"is_oa":true,"is_in_doaj":true,"is_core":true,"host_organization":"https://openalex.org/P4310320244","host_organization_name":"Association for Computational Linguistics","host_organization_lineage":["https://openalex.org/P4310320244"],"host_organization_lineage_names":["Association for Computational Linguistics"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Transactions of the Association for Computational Linguistics","raw_type":"journal-article"},{"id":"pmh:oai:fr.eurecom:7983","is_oa":false,"landing_page_url":null,"pdf_url":null,"source":{"id":"https://openalex.org/S4377196942","display_name":"Graduate School and Research Center in Digital Science (EURECOM)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I1902872","host_organization_name":"EURECOM","host_organization_lineage":["https://openalex.org/I1902872"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"EMNLP 2024, Conference on Empirical Methods in Natural Language Processing, 12-16 November 2024, Miami, Florida, USA / Also published in Transactions of the Association for Computational Linguistics, Vol.12, 2024
","raw_type":"Conference"}],"best_oa_location":{"id":"doi:10.1162/tacl_a_00716","is_oa":true,"landing_page_url":"https://doi.org/10.1162/tacl_a_00716","pdf_url":null,"source":{"id":"https://openalex.org/S2729999759","display_name":"Transactions of the Association for Computational Linguistics","issn_l":"2307-387X","issn":["2307-387X"],"is_oa":true,"is_in_doaj":true,"is_core":true,"host_organization":"https://openalex.org/P4310320244","host_organization_name":"Association for Computational Linguistics","host_organization_lineage":["https://openalex.org/P4310320244"],"host_organization_lineage_names":["Association for Computational Linguistics"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Transactions of the Association for Computational Linguistics","raw_type":"journal-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":59,"referenced_works":["https://openalex.org/W2612690371","https://openalex.org/W2734443755","https://openalex.org/W2913668833","https://openalex.org/W2914304175","https://openalex.org/W2949555952","https://openalex.org/W2963323070","https://openalex.org/W2970641574","https://openalex.org/W2972324944","https://openalex.org/W3180037928","https://openalex.org/W4212774754","https://openalex.org/W4221159599","https://openalex.org/W4280546523","https://openalex.org/W4285199320","https://openalex.org/W4310390625","https://openalex.org/W4313483544","https://openalex.org/W4322718191","https://openalex.org/W4366330736","https://openalex.org/W4367000321","https://openalex.org/W4384918448","https://openalex.org/W4385567182","https://openalex.org/W4386655575","https://openalex.org/W4388979610","https://openalex.org/W4389519226","https://openalex.org/W4390833751","https://openalex.org/W4391766637","https://openalex.org/W4391876619","https://openalex.org/W4392222684","https://openalex.org/W4401042914","https://openalex.org/W4401307018","https://openalex.org/W4402671766","https://openalex.org/W4402671835","https://openalex.org/W6728274705","https://openalex.org/W6732703391","https://openalex.org/W6739901393","https://openalex.org/W6777615688","https://openalex.org/W6779089016","https://openalex.org/W6779313456","https://openalex.org/W6779436764","https://openalex.org/W6809646742","https://openalex.org/W6810738896","https://openalex.org/W6848451824","https://openalex.org/W6848909144","https://openalex.org/W6850625674","https://openalex.org/W6851682161","https://openalex.org/W6852408377","https://openalex.org/W6853192989","https://openalex.org/W6853251322","https://openalex.org/W6854866820","https://openalex.org/W6855125150","https://openalex.org/W6856529530","https://openalex.org/W6856619137","https://openalex.org/W6856712200","https://openalex.org/W6857061474","https://openalex.org/W6857690716","https://openalex.org/W6858023062","https://openalex.org/W6860201676","https://openalex.org/W6861707877","https://openalex.org/W6861744286","https://openalex.org/W6870342304"],"related_works":["https://openalex.org/W2808547747","https://openalex.org/W4322393049","https://openalex.org/W638130803","https://openalex.org/W3035050670","https://openalex.org/W1991595402","https://openalex.org/W4252712126","https://openalex.org/W4253352296","https://openalex.org/W2738685648","https://openalex.org/W4226426977","https://openalex.org/W1490540149"],"abstract_inverted_index":{"Abstract":[0],"Recent":[1],"large":[2,131],"language":[3],"model":[4,65],"applications,":[5],"such":[6,99],"as":[7],"Retrieval-Augmented":[8],"Generation":[9],"and":[10,73,85],"chatbots,":[11],"have":[12],"led":[13],"to":[14,18,56,129,138],"an":[15],"increased":[16],"need":[17,146],"process":[19],"longer":[20],"input":[21,59],"contexts.":[22],"However,":[23],"this":[24],"requirement":[25],"is":[26],"hampered":[27],"by":[28,35,61,112],"inherent":[29],"limitations.":[30],"Architecturally,":[31],"models":[32,128],"are":[33,101],"constrained":[34,111],"a":[36,52,71,74,118],"context":[37,60,114],"window":[38],"defined":[39],"during":[40],"training.":[41],"Additionally,":[42],"processing":[43],"extensive":[44],"texts":[45],"requires":[46],"substantial":[47],"GPU":[48],"memory.":[49],"We":[50],"propose":[51],"novel":[53],"approach,":[54],"Finch,":[55],"compress":[57],"the":[58,63,68,80,92,96,104,109,113,122,145],"leveraging":[62],"pre-trained":[64],"weights":[66],"of":[67,91,121],"self-attention.":[69],"Given":[70],"prompt":[72],"long":[75,123],"text,":[76],"Finch":[77],"iteratively":[78],"identifies":[79],"most":[81],"relevant":[82],"Key":[83],"(K)":[84],"Value":[86],"(V)":[87],"pairs":[88,100],"over":[89],"chunks":[90],"text":[93],"conditioned":[94],"on":[95],"prompt.":[97],"Only":[98],"stored":[102],"in":[103],"KV":[105],"cache,":[106],"which,":[107],"within":[108],"space":[110],"window,":[115],"ultimately":[116],"contains":[117],"compressed":[119],"version":[120],"text.":[124],"Our":[125],"proposal":[126],"enables":[127],"consume":[130],"inputs":[132],"even":[133],"with":[134],"high":[135],"compression":[136],"(up":[137],"93x)":[139],"while":[140],"preserving":[141],"semantic":[142],"integrity":[143],"without":[144],"for":[147],"fine-tuning.":[148]},"counts_by_year":[{"year":2026,"cited_by_count":1},{"year":2025,"cited_by_count":3}],"updated_date":"2026-07-22T07:51:19.307946","created_date":"2025-10-10T00:00:00"}