{"id":"https://openalex.org/W7160632811","doi":"https://doi.org/10.48550/arxiv.2605.06221","title":"UniPrefill: Universal Long-Context Prefill Acceleration via Block-wise Dynamic Sparsification","display_name":"UniPrefill: Universal Long-Context Prefill Acceleration via Block-wise Dynamic Sparsification","publication_year":2026,"publication_date":"2026-05-07","ids":{"openalex":"https://openalex.org/W7160632811","doi":"https://doi.org/10.48550/arxiv.2605.06221"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.06221","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.06221","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":null,"license_id":null,"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.2605.06221","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5135690317","display_name":"Qihang Fan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Fan, Qihang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135719133","display_name":"Huaibo Huang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Huang, Huaibo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135663027","display_name":"Zhiying Wu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wu, Zhiying","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135699132","display_name":"Bingning Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Bingning","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5135699544","display_name":"Ran He","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"He, Ran","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/T10028","display_name":"Topic Modeling","score":0.1363999992609024,"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":0.1363999992609024,"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/T13702","display_name":"Machine Learning in Healthcare","score":0.13269999623298645,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.11559999734163284,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/speedup","display_name":"Speedup","score":0.8932999968528748},{"id":"https://openalex.org/keywords/acceleration","display_name":"Acceleration","score":0.7687000036239624},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.6193000078201294},{"id":"https://openalex.org/keywords/computation","display_name":"Computation","score":0.5412999987602234},{"id":"https://openalex.org/keywords/context","display_name":"Context (archaeology)","score":0.5142999887466431},{"id":"https://openalex.org/keywords/scalability","display_name":"Scalability","score":0.4812999963760376},{"id":"https://openalex.org/keywords/security-token","display_name":"Security token","score":0.4625999927520752}],"concepts":[{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.8932999968528748},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7803000211715698},{"id":"https://openalex.org/C117896860","wikidata":"https://www.wikidata.org/wiki/Q11376","display_name":"Acceleration","level":2,"score":0.7687000036239624},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.6193000078201294},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.5412999987602234},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.5142999887466431},{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.4812999963760376},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.4625999927520752},{"id":"https://openalex.org/C206729178","wikidata":"https://www.wikidata.org/wiki/Q2271896","display_name":"Scheduling (production processes)","level":2,"score":0.44909998774528503},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.3977999985218048},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.34389999508857727},{"id":"https://openalex.org/C17020691","wikidata":"https://www.wikidata.org/wiki/Q139677","display_name":"Operator (biology)","level":5,"score":0.3240000009536743},{"id":"https://openalex.org/C46743427","wikidata":"https://www.wikidata.org/wiki/Q1341685","display_name":"Inference engine","level":3,"score":0.322299987077713},{"id":"https://openalex.org/C13164978","wikidata":"https://www.wikidata.org/wiki/Q600158","display_name":"Hardware acceleration","level":3,"score":0.2976999878883362},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.28850001096725464},{"id":"https://openalex.org/C98045186","wikidata":"https://www.wikidata.org/wiki/Q205663","display_name":"Process (computing)","level":2,"score":0.28040000796318054},{"id":"https://openalex.org/C55416958","wikidata":"https://www.wikidata.org/wiki/Q6206757","display_name":"Job shop scheduling","level":3,"score":0.2750000059604645},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.2721000015735626},{"id":"https://openalex.org/C2780801425","wikidata":"https://www.wikidata.org/wiki/Q5164392","display_name":"Construct (python library)","level":2,"score":0.2597000002861023},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.2590000033378601},{"id":"https://openalex.org/C147297375","wikidata":"https://www.wikidata.org/wiki/Q6674930","display_name":"Look-ahead","level":2,"score":0.2567000091075897}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.06221","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.06221","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2605.06221","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.06221","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"As":[0],"large":[1],"language":[2],"models":[3],"(LLMs)":[4],"continue":[5],"to":[6,70,101,121,150,169],"advance":[7],"rapidly,":[8],"they":[9],"are":[10,92],"becoming":[11,178],"increasingly":[12,179],"capable":[13],"while":[14],"simultaneously":[15],"demanding":[16],"ever-longer":[17],"context":[18],"lengths.":[19],"To":[20,110],"improve":[21],"the":[22,39,129,133,176,182],"inference":[23,105],"efficiency":[24],"of":[25,42,184],"long-context":[26,43,49],"processing,":[27],"several":[28],"novel":[29],"low-complexity":[30],"hybrid":[31],"architectures":[32],"have":[33],"recently":[34],"been":[35],"proposed,":[36],"effectively":[37],"alleviating":[38],"computational":[40],"burden":[41],"inference.":[44],"However,":[45],"existing":[46],"research":[47],"on":[48,55,65],"prefill":[50,82,117],"acceleration":[51,83,118,177],"remains":[52],"predominantly":[53],"focused":[54],"sparse":[56],"attention":[57,75,80],"mechanisms,":[58],"which":[59,126],"achieve":[60],"their":[61],"maximum":[62],"speedup":[63,171],"only":[64],"full-attention":[66],"models.":[67],"When":[68],"transferred":[69],"emerging":[71],"architectures--such":[72],"as":[73,108,140,181],"linear/full":[74],"hybrids":[76],"or":[77],"sliding":[78],"window/full":[79],"hybrids--these":[81],"approaches":[84],"suffer":[85],"significant":[86],"performance":[87],"degradation.":[88],"Furthermore,":[89],"such":[90,107],"methods":[91],"generally":[93],"incompatible":[94],"with":[95,175],"continuous":[96,142],"batching,":[97],"making":[98],"them":[99],"difficult":[100],"integrate":[102],"into":[103,164],"modern":[104],"engines":[106],"vLLM.":[109,165],"this":[111],"end,":[112],"we":[113],"propose":[114],"UniPrefill,":[115,159],"a":[116,141],"framework":[119],"applicable":[120],"virtually":[122],"any":[123],"model":[124],"architecture,":[125],"directly":[127],"accelerates":[128],"model's":[130],"computation":[131],"at":[132],"token":[134],"level.":[135],"We":[136],"further":[137],"implement":[138],"UniPrefill":[139,166],"batching":[143],"operator":[144],"and":[145,155],"extend":[146],"vLLM's":[147],"scheduling":[148],"strategy":[149],"natively":[151],"support":[152],"prefill-decode":[153],"co-processing":[154],"tensor":[156],"parallel":[157],"for":[158],"enabling":[160],"its":[161],"seamless":[162],"integration":[163],"achieves":[167],"up":[168],"2.1x":[170],"in":[172],"Time-To-First-Token":[173],"(TTFT),":[174],"pronounced":[180],"number":[183],"concurrent":[185],"requests":[186],"grows.":[187]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-05-09T00:00:00"}