{"id":"https://openalex.org/W7160847554","doi":"https://doi.org/10.48550/arxiv.2605.07363","title":"MISA: Mixture of Indexer Sparse Attention for Long-Context LLM Inference","display_name":"MISA: Mixture of Indexer Sparse Attention for Long-Context LLM Inference","publication_year":2026,"publication_date":"2026-05-08","ids":{"openalex":"https://openalex.org/W7160847554","doi":"https://doi.org/10.48550/arxiv.2605.07363"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.07363","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.07363","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.2605.07363","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5112689668","display_name":"R Zhou","orcid":"https://orcid.org/0009-0003-7391-4794"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhou, Ruijie","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135840460","display_name":"Fanxu Meng","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Meng, Fanxu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135851128","display_name":"Yufei Xu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xu, Yufei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135906715","display_name":"Tongxuan Liu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Tongxuan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135875264","display_name":"Guangming Lu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lu, Guangming","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135871189","display_name":"Muhan Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Muhan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5078487642","display_name":"Wenjie Pei","orcid":"https://orcid.org/0000-0001-8117-2696"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Pei, Wenjie","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":0.3815000057220459,"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":0.3815000057220459,"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/T10601","display_name":"Handwritten Text Recognition Techniques","score":0.05420000106096268,"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/T10028","display_name":"Topic Modeling","score":0.05220000073313713,"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/security-token","display_name":"Security token","score":0.704200029373169},{"id":"https://openalex.org/keywords/set","display_name":"Set (abstract data type)","score":0.65420001745224},{"id":"https://openalex.org/keywords/context","display_name":"Context (archaeology)","score":0.6434999704360962},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.5701000094413757},{"id":"https://openalex.org/keywords/prefix","display_name":"Prefix","score":0.521399974822998},{"id":"https://openalex.org/keywords/router","display_name":"Router","score":0.44769999384880066}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7218999862670898},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.704200029373169},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.65420001745224},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.6434999704360962},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5701000094413757},{"id":"https://openalex.org/C141603448","wikidata":"https://www.wikidata.org/wiki/Q134830","display_name":"Prefix","level":2,"score":0.521399974822998},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4544999897480011},{"id":"https://openalex.org/C2775896111","wikidata":"https://www.wikidata.org/wiki/Q642560","display_name":"Router","level":2,"score":0.44769999384880066},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.3325999975204468},{"id":"https://openalex.org/C23123220","wikidata":"https://www.wikidata.org/wiki/Q816826","display_name":"Information retrieval","level":1,"score":0.3167000114917755},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.29170000553131104},{"id":"https://openalex.org/C151810110","wikidata":"https://www.wikidata.org/wiki/Q7307074","display_name":"RefSeq","level":4,"score":0.2648000121116638},{"id":"https://openalex.org/C61797465","wikidata":"https://www.wikidata.org/wiki/Q1188986","display_name":"Term (time)","level":2,"score":0.25699999928474426}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.07363","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.07363","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.2605.07363","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.07363","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":{"DeepSeek":[0],"Sparse":[1,78],"Attention":[2],"(DSA)":[3],"sets":[4],"the":[5,8,28,33,39,52,64,66,84,120,126,129,135,179,193,199,216,260,264],"state":[6],"of":[7,76,95,109,128,153,166,175,259],"art":[9],"for":[10,32,83],"fine-grained":[11],"inference-time":[12],"sparse":[13],"attention":[14],"by":[15,263],"introducing":[16],"a":[17,80,93,106,111,151,157,163,172,251,274,284],"learned":[18],"token-wise":[19],"indexer":[20,40,65,86,90,131,196,219,234,266,281],"that":[21,50,87,177],"scores":[22],"every":[23,140,144],"prefix":[24,141],"token":[25,55,142],"and":[26,115,188,210,224,230,237,254],"selects":[27],"most":[29],"relevant":[30],"ones":[31],"main":[34],"attention.":[35],"To":[36],"remain":[37],"expressive,":[38],"uses":[41,100,178],"many":[42],"query":[43],"heads":[44,91,118,209,235],"(for":[45],"example,":[46],"64":[47],"on":[48,69,162,220,240,283],"DeepSeek-V3.2)":[49],"share":[51],"same":[53],"selected":[54,201,262],"set;":[56],"this":[57],"multi-head":[58],"design":[59],"is":[60],"precisely":[61],"what":[62],"makes":[63],"dominant":[67],"cost":[68,137],"long":[70],"contexts.":[71],"We":[72,169],"propose":[73],"MISA":[74,176,214],"(Mixture":[75],"Indexer":[77],"Attention),":[79],"drop-in":[81],"replacement":[82],"DSA":[85,195,218,265],"treats":[88],"its":[89],"as":[92],"pool":[94,132],"mixture-of-experts.":[96],"A":[97],"lightweight":[98],"router":[99,159],"cheap":[101],"block-level":[102],"statistics":[103],"to":[104,146,182,197,250],"pick":[105],"query-dependent":[107],"subset":[108],"only":[110,116,150,206],"few":[112],"active":[113,208],"heads,":[114,155],"those":[117],"run":[119],"heavy":[121],"token-level":[122],"scoring.":[123],"This":[124],"preserves":[125,244],"diversity":[127],"original":[130,194,280],"while":[133,226],"reducing":[134],"per-query":[136],"from":[138],"scoring":[139,147],"with":[143,149,192,228],"head":[145],"it":[148,191],"handful":[152],"routed":[154,180],"plus":[156],"negligible":[158],"term":[160],"computed":[161],"small":[164],"set":[165,187],"pooled":[167],"keys.":[168],"further":[170],"introduce":[171],"hierarchical":[173],"variant":[174],"pass":[181],"keep":[183],"an":[184],"enlarged":[185],"candidate":[186],"then":[189],"re-ranks":[190],"recover":[198],"final":[200],"tokens":[202,261],"almost":[203],"exactly.":[204],"With":[205],"eight":[207,229],"no":[211],"additional":[212],"training,":[213],"matches":[215],"dense":[217],"LongBench":[221],"across":[222],"DeepSeek-V3.2":[223],"GLM-5":[225],"running":[227],"four":[231],"times":[232,276],"fewer":[233],"respectively,":[236],"outperforms":[238],"HISA":[239],"average.":[241],"It":[242],"also":[243],"fully":[245],"green":[246],"Needle-in-a-Haystack":[247],"heatmaps":[248],"up":[249],"128K-token":[252],"context":[253],"recovers":[255],"more":[256],"than":[257],"92%":[258],"per":[267],"layer.":[268],"Our":[269],"TileLang":[270],"kernel":[271,282],"delivers":[272],"roughly":[273],"3.82":[275],"speedup":[277],"over":[278],"DSA's":[279],"single":[285],"NVIDIA":[286],"H200":[287],"GPU.":[288]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-05-12T00:00:00"}