{"id":"https://openalex.org/W4407831517","doi":"https://doi.org/10.48550/arxiv.2502.14866","title":"LServe: Efficient Long-sequence LLM Serving with Unified Sparse Attention","display_name":"LServe: Efficient Long-sequence LLM Serving with Unified Sparse Attention","publication_year":2025,"publication_date":"2025-02-20","ids":{"openalex":"https://openalex.org/W4407831517","doi":"https://doi.org/10.48550/arxiv.2502.14866"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2502.14866","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2502.14866","pdf_url":"https://arxiv.org/pdf/2502.14866","source":{"id":"https://openalex.org/S4393918464","display_name":"ArXiv.org","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2502.14866","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5101467057","display_name":"Yang Shang","orcid":"https://orcid.org/0000-0002-5836-5016"},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Yang, Shang","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5107749807","display_name":"J.Z. Li D.B. Guo","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Guo, Junxian","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5054102866","display_name":"Haotian Tang","orcid":"https://orcid.org/0000-0002-6567-6170"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tang, Haotian","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5035921543","display_name":"Qinghao Hu","orcid":"https://orcid.org/0000-0003-1034-7502"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hu, Qinghao","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5009724487","display_name":"Guangxuan Xiao","orcid":"https://orcid.org/0000-0002-7182-9284"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xiao, Guangxuan","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Tang, Jiaming","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tang, Jiaming","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Lin, Yujun","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lin, Yujun","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100750262","display_name":"Zhijian Liu","orcid":"https://orcid.org/0000-0003-1841-2671"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Zhijian","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5102001223","display_name":"Yao Lu","orcid":"https://orcid.org/0000-0002-1587-0135"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lu, Yao","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5070926896","display_name":"Song Han","orcid":"https://orcid.org/0000-0002-4186-7618"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Han, Song","raw_affiliation_strings":[],"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":10,"corresponding_author_ids":["https://openalex.org/A5101467057"],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"cited_by_count":1,"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/T11447","display_name":"Blind Source Separation Techniques","score":0.9854999780654907,"subfield":{"id":"https://openalex.org/subfields/1711","display_name":"Signal Processing"},"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/T11447","display_name":"Blind Source Separation Techniques","score":0.9854999780654907,"subfield":{"id":"https://openalex.org/subfields/1711","display_name":"Signal Processing"},"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/T13717","display_name":"Advanced Algorithms and Applications","score":0.9735000133514404,"subfield":{"id":"https://openalex.org/subfields/2207","display_name":"Control and Systems Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10320","display_name":"Neural Networks and Applications","score":0.968500018119812,"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/sequence","display_name":"Sequence (biology)","score":0.5630218386650085},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.39929455518722534},{"id":"https://openalex.org/keywords/chemistry","display_name":"Chemistry","score":0.18903625011444092}],"concepts":[{"id":"https://openalex.org/C2778112365","wikidata":"https://www.wikidata.org/wiki/Q3511065","display_name":"Sequence (biology)","level":2,"score":0.5630218386650085},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.39929455518722534},{"id":"https://openalex.org/C185592680","wikidata":"https://www.wikidata.org/wiki/Q2329","display_name":"Chemistry","level":0,"score":0.18903625011444092},{"id":"https://openalex.org/C55493867","wikidata":"https://www.wikidata.org/wiki/Q7094","display_name":"Biochemistry","level":1,"score":0.0}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2502.14866","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2502.14866","pdf_url":"https://arxiv.org/pdf/2502.14866","source":{"id":"https://openalex.org/S4393918464","display_name":"ArXiv.org","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2502.14866","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2502.14866","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":"article"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2502.14866","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2502.14866","pdf_url":"https://arxiv.org/pdf/2502.14866","source":{"id":"https://openalex.org/S4393918464","display_name":"ArXiv.org","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2899084033","https://openalex.org/W2748952813","https://openalex.org/W2390279801","https://openalex.org/W4391913857","https://openalex.org/W2358668433","https://openalex.org/W4396701345","https://openalex.org/W2376932109","https://openalex.org/W2001405890","https://openalex.org/W4396696052"],"abstract_inverted_index":{"Large":[0],"language":[1],"models":[2,20],"(LLMs)":[3],"have":[4],"shown":[5],"remarkable":[6],"potential":[7],"in":[8,31,44,103,129],"processing":[9],"long":[10],"sequences":[11],"and":[12,35,78,100,133,152,187],"complex":[13],"reasoning":[14,153],"tasks,":[15],"yet":[16],"efficiently":[17],"serving":[18,62],"these":[19,50,114],"remains":[21],"challenging":[22],"due":[23],"to":[24,124,149,185],"the":[25,32,36,41,45,96,121,131],"quadratic":[26],"computational":[27],"complexity":[28],"of":[29,40,98,120,144,156],"attention":[30,80,122],"prefilling":[33,77,132,182],"stage":[34],"large":[37],"memory":[38],"footprint":[39],"KV":[42,145,164,171],"cache":[43],"decoding":[46,79,134,188],"stage.":[47],"To":[48],"address":[49],"issues,":[51],"we":[52,117,137],"introduce":[53],"LServe,":[54],"an":[55],"efficient":[56],"system":[57],"that":[58,139,168],"accelerates":[59,180],"long-sequence":[60],"LLM":[61,105,181],"via":[63],"hybrid":[64],"sparse":[65],"attention.":[66,106],"This":[67,107],"method":[68],"unifies":[69],"different":[70],"hardware-friendly,":[71],"structured":[72],"sparsity":[73,102],"patterns":[74],"for":[75],"both":[76,130],"into":[81],"a":[82,141,162],"single":[83],"framework,":[84],"where":[85],"computations":[86],"on":[87,174],"less":[88],"important":[89],"tokens":[90],"are":[91],"skipped":[92],"block-wise.":[93],"LServe":[94,179],"demonstrates":[95],"compatibility":[97],"static":[99],"dynamic":[101],"long-context":[104,151,194],"design":[108,161],"enables":[109],"multiplicative":[110],"speedups":[111],"by":[112,183,189],"combining":[113],"optimizations.":[115],"Specifically,":[116],"convert":[118],"half":[119],"heads":[123,128],"nearly":[125],"free":[126],"streaming":[127],"stages.":[135],"Additionally,":[136],"find":[138],"only":[140],"constant":[142],"number":[143],"pages":[146,172],"is":[147,197],"required":[148],"preserve":[150],"capabilities,":[154],"irrespective":[155],"context":[157],"length.":[158],"We":[159],"then":[160],"hierarchical":[163],"page":[165],"selection":[166],"policy":[167],"dynamically":[169],"prunes":[170],"based":[173],"query-centric":[175],"similarity.":[176],"On":[177],"average,":[178],"up":[184],"2.9x":[186],"1.3-2.1x":[190],"over":[191],"vLLM,":[192],"maintaining":[193],"accuracy.":[195],"Code":[196],"released":[198],"at":[199],"https://github.com/mit-han-lab/omniserve.":[200]},"counts_by_year":[{"year":2025,"cited_by_count":1}],"updated_date":"2026-04-21T08:09:41.155169","created_date":"2025-10-10T00:00:00"}
