{"id":"https://openalex.org/W7156628940","doi":"https://doi.org/10.48550/arxiv.2604.24647","title":"DepthKV: Layer-Dependent KV Cache Pruning for Long-Context LLM Inference","display_name":"DepthKV: Layer-Dependent KV Cache Pruning for Long-Context LLM Inference","publication_year":2026,"publication_date":"2026-04-27","ids":{"openalex":"https://openalex.org/W7156628940","doi":"https://doi.org/10.48550/arxiv.2604.24647"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2604.24647","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.24647","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2604.24647","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5120002142","display_name":"Zahra Dehghanighobadi","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Dehghanighobadi, Zahra","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5026151059","display_name":"Asja Fischer","orcid":"https://orcid.org/0000-0002-1916-7033"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Fischer, Asja","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":0.2540999948978424,"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":0.2540999948978424,"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":0.15690000355243683,"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":0.09080000221729279,"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/pruning","display_name":"Pruning","score":0.8389999866485596},{"id":"https://openalex.org/keywords/cache","display_name":"Cache","score":0.7177000045776367},{"id":"https://openalex.org/keywords/memory-footprint","display_name":"Memory footprint","score":0.6481999754905701},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.6075000166893005},{"id":"https://openalex.org/keywords/code","display_name":"Code (set theory)","score":0.4657000005245209},{"id":"https://openalex.org/keywords/footprint","display_name":"Footprint","score":0.4593000113964081},{"id":"https://openalex.org/keywords/sequence","display_name":"Sequence (biology)","score":0.4203999936580658}],"concepts":[{"id":"https://openalex.org/C108010975","wikidata":"https://www.wikidata.org/wiki/Q500094","display_name":"Pruning","level":2,"score":0.8389999866485596},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7211999893188477},{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.7177000045776367},{"id":"https://openalex.org/C74912251","wikidata":"https://www.wikidata.org/wiki/Q6815727","display_name":"Memory footprint","level":2,"score":0.6481999754905701},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.6075000166893005},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.4657000005245209},{"id":"https://openalex.org/C132943942","wikidata":"https://www.wikidata.org/wiki/Q2562511","display_name":"Footprint","level":2,"score":0.4593000113964081},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.42660000920295715},{"id":"https://openalex.org/C2778112365","wikidata":"https://www.wikidata.org/wiki/Q3511065","display_name":"Sequence (biology)","level":2,"score":0.4203999936580658},{"id":"https://openalex.org/C159877910","wikidata":"https://www.wikidata.org/wiki/Q2202883","display_name":"Autoregressive model","level":2,"score":0.4187000095844269},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.4068000018596649},{"id":"https://openalex.org/C189783530","wikidata":"https://www.wikidata.org/wiki/Q352090","display_name":"CPU cache","level":3,"score":0.3939000070095062},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3547999858856201},{"id":"https://openalex.org/C38556500","wikidata":"https://www.wikidata.org/wiki/Q13404475","display_name":"Cache algorithms","level":4,"score":0.34529998898506165},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.34450000524520874},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.32670000195503235},{"id":"https://openalex.org/C43126263","wikidata":"https://www.wikidata.org/wiki/Q128751","display_name":"Source code","level":2,"score":0.2881999909877777},{"id":"https://openalex.org/C169590947","wikidata":"https://www.wikidata.org/wiki/Q47506","display_name":"Compiler","level":2,"score":0.2687999904155731}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2604.24647","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.24647","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2604.24647","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.24647","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Long-context":[0],"reasoning":[1],"is":[2,88],"a":[3,41,66,102,108,122],"critical":[4],"capability":[5],"of":[6,145],"large":[7],"language":[8],"models":[9,127],"(LLMs),":[10],"enabling":[11],"applications":[12],"such":[13],"as":[14,90],"long-document":[15],"understanding,":[16],"summarization,":[17],"and":[18,128],"code":[19],"generation.":[20],"However,":[21],"efficient":[22],"autoregressive":[23],"inference":[24],"relies":[25],"on":[26,116],"the":[27,136,146],"key-value":[28],"(KV)":[29],"cache,":[30],"whose":[31],"memory":[32,43],"footprint":[33],"grows":[34],"linearly":[35],"with":[36,56],"sequence":[37],"length,":[38],"leading":[39],"to":[40,79,97],"major":[42],"bottleneck.":[44],"To":[45],"mitigate":[46],"this":[47,86],"overhead,":[48],"KV":[49,111,147],"cache":[50,148],"pruning":[51,68,104,134,139],"methods":[52,64],"discard":[53],"cached":[54],"tokens":[55],"low":[57],"attention":[58],"scores":[59],"during":[60],"inference.":[61],"Most":[62],"existing":[63],"apply":[65],"uniform":[67,123,133],"ratio":[69],"across":[70,113],"layers,":[71],"implicitly":[72],"assuming":[73],"that":[74,85,106],"all":[75],"layers":[76,91,114],"contribute":[77],"equally":[78],"overall":[80],"model":[81],"performance.":[82],"We":[83,99],"show":[84],"assumption":[87],"suboptimal,":[89],"differ":[92],"significantly":[93],"in":[94],"their":[95,117],"sensitivity":[96],"pruning.":[98],"propose":[100],"DepthKV,":[101],"layer-dependent":[103,151],"framework":[105],"allocates":[107],"fixed":[109],"global":[110,138],"budget":[112,149],"based":[115],"sensitivity,":[118],"rather":[119],"than":[120],"using":[121],"allocation.":[124,152],"Across":[125],"multiple":[126],"tasks,":[129],"DepthKV":[130],"consistently":[131],"outperforms":[132],"at":[135],"same":[137],"ratio,":[140],"demonstrating":[141],"more":[142],"effective":[143],"utilization":[144],"through":[150]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-04-29T00:00:00"}