{"id":"https://openalex.org/W7161727518","doi":"https://doi.org/10.48550/arxiv.2605.16360","title":"ProxyKV: Cross-Model Proxy Pruning for Efficient Long-Context LLM Inference","display_name":"ProxyKV: Cross-Model Proxy Pruning for Efficient Long-Context LLM Inference","publication_year":2026,"publication_date":"2026-05-09","ids":{"openalex":"https://openalex.org/W7161727518","doi":"https://doi.org/10.48550/arxiv.2605.16360"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.16360","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.16360","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,"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.2605.16360","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5136478421","display_name":"Junjie Li","orcid":"https://orcid.org/0000-0002-9274-9902"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Junjie","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5062969662","display_name":"Jiong Lou","orcid":"https://orcid.org/0000-0001-9245-2626"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lou, Jiong","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5136500152","display_name":"Jie Li","orcid":"https://orcid.org/0009-0007-2929-4267"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Jie","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/T10036","display_name":"Advanced Neural Network Applications","score":0.10360000282526016,"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"}},"topics":[{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.10360000282526016,"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/T14347","display_name":"Big Data and Digital Economy","score":0.1006999984383583,"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"}},{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.09399999678134918,"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/speedup","display_name":"Speedup","score":0.8246999979019165},{"id":"https://openalex.org/keywords/pruning","display_name":"Pruning","score":0.7045999765396118},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.6438999772071838},{"id":"https://openalex.org/keywords/heuristics","display_name":"Heuristics","score":0.6419000029563904},{"id":"https://openalex.org/keywords/proxy","display_name":"Proxy (statistics)","score":0.569100022315979},{"id":"https://openalex.org/keywords/security-token","display_name":"Security token","score":0.5187000036239624},{"id":"https://openalex.org/keywords/regression","display_name":"Regression","score":0.39579999446868896},{"id":"https://openalex.org/keywords/cache","display_name":"Cache","score":0.3804999887943268}],"concepts":[{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.8246999979019165},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7968999743461609},{"id":"https://openalex.org/C108010975","wikidata":"https://www.wikidata.org/wiki/Q500094","display_name":"Pruning","level":2,"score":0.7045999765396118},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.6438999772071838},{"id":"https://openalex.org/C127705205","wikidata":"https://www.wikidata.org/wiki/Q5748245","display_name":"Heuristics","level":2,"score":0.6419000029563904},{"id":"https://openalex.org/C2780148112","wikidata":"https://www.wikidata.org/wiki/Q1432581","display_name":"Proxy (statistics)","level":2,"score":0.569100022315979},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.560699999332428},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.5187000036239624},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.48559999465942383},{"id":"https://openalex.org/C83546350","wikidata":"https://www.wikidata.org/wiki/Q1139051","display_name":"Regression","level":2,"score":0.39579999446868896},{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.3804999887943268},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.37130001187324524},{"id":"https://openalex.org/C189430467","wikidata":"https://www.wikidata.org/wiki/Q7293293","display_name":"Ranking (information retrieval)","level":2,"score":0.3702999949455261},{"id":"https://openalex.org/C100776233","wikidata":"https://www.wikidata.org/wiki/Q2532492","display_name":"Bridge (graph theory)","level":2,"score":0.3666999936103821},{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.35499998927116394},{"id":"https://openalex.org/C53811970","wikidata":"https://www.wikidata.org/wiki/Q5062194","display_name":"Centrality","level":2,"score":0.33320000767707825},{"id":"https://openalex.org/C4679612","wikidata":"https://www.wikidata.org/wiki/Q866298","display_name":"Aggregate (composite)","level":2,"score":0.33230000734329224},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.3172999918460846},{"id":"https://openalex.org/C52622490","wikidata":"https://www.wikidata.org/wiki/Q1026626","display_name":"Feature extraction","level":2,"score":0.29409998655319214},{"id":"https://openalex.org/C108583219","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep learning","level":2,"score":0.28600001335144043},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.2851000130176544},{"id":"https://openalex.org/C113174947","wikidata":"https://www.wikidata.org/wiki/Q2859736","display_name":"Tree (set theory)","level":2,"score":0.27790001034736633},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.2572999894618988},{"id":"https://openalex.org/C189783530","wikidata":"https://www.wikidata.org/wiki/Q352090","display_name":"CPU cache","level":3,"score":0.25099998712539673},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.2506999969482422}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.16360","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.16360","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,"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.2605.16360","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.16360","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,"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":{"Efficient":[0],"long-context":[1],"inference":[2],"in":[3],"Large":[4],"Language":[5],"Models":[6],"(LLMs)":[7],"is":[8],"severely":[9],"constrained":[10],"by":[11],"the":[12,66,71,79,97,108,154],"Key-Value":[13],"(KV)":[14],"cache":[15],"memory":[16],"wall,":[17],"yet":[18],"existing":[19],"pruning":[20,51],"methods":[21,34],"force":[22],"a":[23,48,58,91,142],"choice":[24],"between":[25,74],"low-latency":[26],"heuristics":[27],"that":[28,35,53,95],"sacrifice":[29],"precision":[30],"and":[31,111,125,152],"high-precision":[32],"reconstruction":[33],"incur":[36],"prohibitive":[37],"prefilling":[38,144],"overhead.":[39],"To":[40,69],"bridge":[41,70],"this":[42],"scoring-cost--accuracy":[43],"gap,":[44],"we":[45,77],"propose":[46],"ProxyKV,":[47],"cross-model":[49],"proxy":[50],"framework":[52],"offloads":[54],"importance":[55],"scoring":[56],"to":[57,65,103,119,141,159],"lightweight":[59],"intra-family":[60],"Small-Model":[61],"Proxy":[62],"executed":[63],"asynchronously":[64],"Large-Model":[67],"Target.":[68],"architectural":[72],"gap":[73],"heterogeneous":[75],"models,":[76],"design":[78],"HybridAxialMapper,":[80],"which":[81],"disentangles":[82],"temporal":[83],"feature":[84],"extraction":[85],"from":[86,100,116],"cross-head":[87],"alignment,":[88],"together":[89],"with":[90],"Multi-Granularity":[92],"Hybrid":[93],"Loss":[94],"shifts":[96],"learning":[98],"objective":[99],"rigid":[101],"regression":[102],"relative":[104],"ranking":[105],"consistency.":[106],"Across":[107],"Llama-3.1,":[109],"Qwen-2.5,":[110],"Qwen-3":[112],"families":[113],"spanning":[114],"targets":[115],"7B":[117],"up":[118,140,158],"32B":[120],"parameters":[121],"on":[122,130,146,162],"LongBench,":[123],"SCBench,":[124],"RULER,":[126],"ProxyKV":[127],"matches":[128],"KVZip":[129],"aggregate":[131],"(recovering":[132],"$\\sim$$98.7\\%$":[133],"of":[134],"its":[135],"mean":[136],"accuracy)":[137],"while":[138],"delivering":[139],"$3.21\\times$":[143],"speedup":[145,155],"Llama-3.1-8B":[147],"(dual-GPU;":[148],"$\\sim$$1.5\\times$":[149],"shared":[150],"single-GPU)":[151],"sustaining":[153],"at":[156],"contexts":[157],"170k":[160],"tokens":[161],"Qwen-2.5-7B.":[163]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-05-20T00:00:00"}