{"id":"https://openalex.org/W7137956095","doi":"https://doi.org/10.1609/aaai.v40i27.39403","title":"SMIDT: High-Performance Inference Framework for MoE Models with Dynamic Top-K Routing","display_name":"SMIDT: High-Performance Inference Framework for MoE Models with Dynamic Top-K Routing","publication_year":2026,"publication_date":"2026-03-14","ids":{"openalex":"https://openalex.org/W7137956095","doi":"https://doi.org/10.1609/aaai.v40i27.39403"},"language":"en","primary_location":{"id":"doi:10.1609/aaai.v40i27.39403","is_oa":true,"landing_page_url":"https://doi.org/10.1609/aaai.v40i27.39403","pdf_url":"https://ojs.aaai.org/index.php/AAAI/article/download/39403/43364","source":{"id":"https://openalex.org/S4210191458","display_name":"Proceedings of the AAAI Conference on Artificial Intelligence","issn_l":"2159-5399","issn":["2159-5399","2374-3468"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/P4310320058","host_organization_name":"Association for the Advancement of Artificial Intelligence","host_organization_lineage":["https://openalex.org/P4310320058"],"host_organization_lineage_names":["Association for the Advancement of Artificial Intelligence"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the AAAI Conference on Artificial Intelligence","raw_type":"journal-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"diamond","oa_url":"https://ojs.aaai.org/index.php/AAAI/article/download/39403/43364","any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5102982554","display_name":"Zewen Jin","orcid":"https://orcid.org/0000-0002-4762-6559"},"institutions":[{"id":"https://openalex.org/I126520041","display_name":"University of Science and Technology of China","ror":"https://ror.org/04c4dkn09","country_code":"CN","type":"education","lineage":["https://openalex.org/I126520041","https://openalex.org/I19820366"]}],"countries":["CN"],"is_corresponding":true,"raw_author_name":"Zewen Jin","raw_affiliation_strings":["University of Science and Technology of China\nInstitute of Artificial Intelligence, Hefei Comprehensive National Science Center"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Science and Technology of China\nInstitute of Artificial Intelligence, Hefei Comprehensive National Science Center","institution_ids":["https://openalex.org/I126520041"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5034333635","display_name":"Shen Yu Fu","orcid":null},"institutions":[{"id":"https://openalex.org/I126520041","display_name":"University of Science and Technology of China","ror":"https://ror.org/04c4dkn09","country_code":"CN","type":"education","lineage":["https://openalex.org/I126520041","https://openalex.org/I19820366"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Shen Fu","raw_affiliation_strings":["University of Science and Technology of China\nInstitute of Artificial Intelligence, Hefei Comprehensive National Science Center"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Science and Technology of China\nInstitute of Artificial Intelligence, Hefei Comprehensive National Science Center","institution_ids":["https://openalex.org/I126520041"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Chengjie Tang","orcid":null},"institutions":[{"id":"https://openalex.org/I4210137491","display_name":"National Science Centre","ror":"https://ror.org/03ha2q922","country_code":"PL","type":"government","lineage":["https://openalex.org/I4210137491"]}],"countries":["PL"],"is_corresponding":false,"raw_author_name":"Chengjie Tang","raw_affiliation_strings":["Shanxi University\nInstitute of Artificial Intelligence, Hefei Comprehensive National Science Center"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Shanxi University\nInstitute of Artificial Intelligence, Hefei Comprehensive National Science Center","institution_ids":["https://openalex.org/I4210137491"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5037054290","display_name":"Youhui Bai","orcid":null},"institutions":[{"id":"https://openalex.org/I126520041","display_name":"University of Science and Technology of China","ror":"https://ror.org/04c4dkn09","country_code":"CN","type":"education","lineage":["https://openalex.org/I126520041","https://openalex.org/I19820366"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Youhui Bai","raw_affiliation_strings":["University of Science and Technology of China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Science and Technology of China","institution_ids":["https://openalex.org/I126520041"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100343221","display_name":"Shengnan Wang","orcid":"https://orcid.org/0000-0002-2762-1339"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shengnan Wang","raw_affiliation_strings":["Independent Researcher"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Independent Researcher","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100370143","display_name":"Jiaan Zhu","orcid":"https://orcid.org/0000-0001-8700-639X"},"institutions":[{"id":"https://openalex.org/I126520041","display_name":"University of Science and Technology of China","ror":"https://ror.org/04c4dkn09","country_code":"CN","type":"education","lineage":["https://openalex.org/I126520041","https://openalex.org/I19820366"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Jiaan Zhu","raw_affiliation_strings":["University of Science and Technology of China\nInstitute of Artificial Intelligence, Hefei Comprehensive National Science Center"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Science and Technology of China\nInstitute of Artificial Intelligence, Hefei Comprehensive National Science Center","institution_ids":["https://openalex.org/I126520041"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129678079","display_name":"Chizheng Fang","orcid":null},"institutions":[{"id":"https://openalex.org/I126520041","display_name":"University of Science and Technology of China","ror":"https://ror.org/04c4dkn09","country_code":"CN","type":"education","lineage":["https://openalex.org/I126520041","https://openalex.org/I19820366"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Chizheng Fang","raw_affiliation_strings":["University of Science and Technology of China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Science and Technology of China","institution_ids":["https://openalex.org/I126520041"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129705643","display_name":"Ping Gong","orcid":null},"institutions":[{"id":"https://openalex.org/I126520041","display_name":"University of Science and Technology of China","ror":"https://ror.org/04c4dkn09","country_code":"CN","type":"education","lineage":["https://openalex.org/I126520041","https://openalex.org/I19820366"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Ping Gong","raw_affiliation_strings":["University of Science and Technology of China\nInstitute of Artificial Intelligence, Hefei Comprehensive National Science Center"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Science and Technology of China\nInstitute of Artificial Intelligence, Hefei Comprehensive National Science Center","institution_ids":["https://openalex.org/I126520041"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5129705722","display_name":"Cheng Li","orcid":null},"institutions":[{"id":"https://openalex.org/I126520041","display_name":"University of Science and Technology of China","ror":"https://ror.org/04c4dkn09","country_code":"CN","type":"education","lineage":["https://openalex.org/I126520041","https://openalex.org/I19820366"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Cheng Li","raw_affiliation_strings":["University of Science and Technology of China\nInstitute of Artificial Intelligence, Hefei Comprehensive National Science Center"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Science and Technology of China\nInstitute of Artificial Intelligence, Hefei Comprehensive National Science Center","institution_ids":["https://openalex.org/I126520041"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":9,"corresponding_author_ids":["https://openalex.org/A5102982554"],"corresponding_institution_ids":["https://openalex.org/I126520041"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.0,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":"40","issue":"27","first_page":"22444","last_page":"22453"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.28189998865127563,"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"}},"topics":[{"id":"https://openalex.org/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.28189998865127563,"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"}},{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.12780000269412994,"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":0.0843999981880188,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/pipeline","display_name":"Pipeline (software)","score":0.6092000007629395},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.5239999890327454},{"id":"https://openalex.org/keywords/routing","display_name":"Routing (electronic design automation)","score":0.492900013923645},{"id":"https://openalex.org/keywords/computation","display_name":"Computation","score":0.4724000096321106},{"id":"https://openalex.org/keywords/throughput","display_name":"Throughput","score":0.4553000032901764},{"id":"https://openalex.org/keywords/decoding-methods","display_name":"Decoding methods","score":0.45410001277923584},{"id":"https://openalex.org/keywords/scheme","display_name":"Scheme (mathematics)","score":0.4171000123023987},{"id":"https://openalex.org/keywords/parallelism","display_name":"Parallelism (grammar)","score":0.4104999899864197}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.788100004196167},{"id":"https://openalex.org/C43521106","wikidata":"https://www.wikidata.org/wiki/Q2165493","display_name":"Pipeline (software)","level":2,"score":0.6092000007629395},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5239999890327454},{"id":"https://openalex.org/C74172769","wikidata":"https://www.wikidata.org/wiki/Q1446839","display_name":"Routing (electronic design automation)","level":2,"score":0.492900013923645},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.4724000096321106},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.45570001006126404},{"id":"https://openalex.org/C157764524","wikidata":"https://www.wikidata.org/wiki/Q1383412","display_name":"Throughput","level":3,"score":0.4553000032901764},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.45410001277923584},{"id":"https://openalex.org/C77618280","wikidata":"https://www.wikidata.org/wiki/Q1155772","display_name":"Scheme (mathematics)","level":2,"score":0.4171000123023987},{"id":"https://openalex.org/C2781172179","wikidata":"https://www.wikidata.org/wiki/Q853109","display_name":"Parallelism (grammar)","level":2,"score":0.4104999899864197},{"id":"https://openalex.org/C37404715","wikidata":"https://www.wikidata.org/wiki/Q380679","display_name":"Dynamic programming","level":2,"score":0.3653999865055084},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.3634999990463257},{"id":"https://openalex.org/C24856439","wikidata":"https://www.wikidata.org/wiki/Q352483","display_name":"Adaptive routing","level":5,"score":0.35190001130104065},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.34850001335144043},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.3422999978065491},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.310699999332428},{"id":"https://openalex.org/C2775937380","wikidata":"https://www.wikidata.org/wiki/Q1232589","display_name":"Replica","level":2,"score":0.3068000078201294},{"id":"https://openalex.org/C61797465","wikidata":"https://www.wikidata.org/wiki/Q1188986","display_name":"Term (time)","level":2,"score":0.29649999737739563},{"id":"https://openalex.org/C12590798","wikidata":"https://www.wikidata.org/wiki/Q3933199","display_name":"Replication (statistics)","level":2,"score":0.2858000099658966},{"id":"https://openalex.org/C82876162","wikidata":"https://www.wikidata.org/wiki/Q17096504","display_name":"Latency (audio)","level":2,"score":0.2858000099658966},{"id":"https://openalex.org/C61483411","wikidata":"https://www.wikidata.org/wiki/Q3124522","display_name":"Data parallelism","level":3,"score":0.27649998664855957},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.2718999981880188},{"id":"https://openalex.org/C34165917","wikidata":"https://www.wikidata.org/wiki/Q188267","display_name":"Programming paradigm","level":2,"score":0.26660001277923584},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.2500999867916107}],"mesh":[],"locations_count":2,"locations":[{"id":"doi:10.1609/aaai.v40i27.39403","is_oa":true,"landing_page_url":"https://doi.org/10.1609/aaai.v40i27.39403","pdf_url":"https://ojs.aaai.org/index.php/AAAI/article/download/39403/43364","source":{"id":"https://openalex.org/S4210191458","display_name":"Proceedings of the AAAI Conference on Artificial Intelligence","issn_l":"2159-5399","issn":["2159-5399","2374-3468"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/P4310320058","host_organization_name":"Association for the Advancement of Artificial Intelligence","host_organization_lineage":["https://openalex.org/P4310320058"],"host_organization_lineage_names":["Association for the Advancement of Artificial Intelligence"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the AAAI Conference on Artificial Intelligence","raw_type":"journal-article"},{"id":"pmh:oai:ojs.aaai.org:article/39403","is_oa":false,"landing_page_url":"https://ojs.aaai.org/index.php/AAAI/article/view/39403","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"2159-5399","raw_type":"info:eu-repo/semantics/publishedVersion"}],"best_oa_location":{"id":"doi:10.1609/aaai.v40i27.39403","is_oa":true,"landing_page_url":"https://doi.org/10.1609/aaai.v40i27.39403","pdf_url":"https://ojs.aaai.org/index.php/AAAI/article/download/39403/43364","source":{"id":"https://openalex.org/S4210191458","display_name":"Proceedings of the AAAI Conference on Artificial Intelligence","issn_l":"2159-5399","issn":["2159-5399","2374-3468"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/P4310320058","host_organization_name":"Association for the Advancement of Artificial Intelligence","host_organization_lineage":["https://openalex.org/P4310320058"],"host_organization_lineage_names":["Association for the Advancement of Artificial Intelligence"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the AAAI Conference on Artificial Intelligence","raw_type":"journal-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7137956095.pdf","grobid_xml":"https://content.openalex.org/works/W7137956095.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"To":[0,103],"accelerate":[1],"Mixture-of-Experts":[2],"(MoE)":[3],"inference,":[4,64],"the":[5,18,34,55],"hybrid":[6],"parallelism":[7,13,161,188],"paradigm":[8],"is":[9],"first":[10],"applying":[11],"pipeline":[12],"(PP)":[14],"to":[15,61,66,96,129,158],"vertically":[16],"divide":[17],"model":[19],"into":[20],"stages,":[21,134],"with":[22,86,185],"each":[23],"stage":[24,90],"further":[25],"divided":[26],"horizontally":[27],"using":[28],"tensor":[29],"or":[30],"expert":[31,138],"parallelism.":[32],"On":[33],"algorithm":[35,151],"side,":[36],"dynamic":[37,58,117,156],"Top-K":[38,59,79,118],"routing":[39,60],"reduces":[40,143],"computation":[41,131],"by":[42,226],"activating":[43],"fewer":[44],"experts":[45],"per":[46],"token":[47],"on":[48,166,175],"average.":[49],"In":[50],"this":[51],"paper,":[52],"we":[53,107],"explore":[54],"application":[56],"of":[57],"PP-enabled":[62],"MoE":[63,112],"aiming":[65],"fully":[67],"unleash":[68],"their":[69,101],"combined":[70],"potential.":[71],"We":[72,163],"identify":[73],"key":[74],"performance":[75],"bottlenecks":[76],"arising":[77],"from":[78],"value":[80],"variation":[81],"across":[82,132],"layers,":[83],"which":[84],"conflicts":[85],"PP's":[87],"typically":[88],"uniform":[89],"partitioning,":[91],"as":[92,94],"well":[93],"opportunities":[95],"optimize":[97],"memory":[98],"usage":[99],"through":[100],"integration.":[102],"address":[104],"these":[105],"challenges,":[106],"present":[108],"SMIDT,":[109],"an":[110,123],"efficient":[111,160],"inference":[113,171],"framework":[114],"tailored":[115],"for":[116,205,210],"routing.":[119],"SMIDT":[120,165,200,219],"features:":[121],"(1)":[122],"adaptive,":[124],"module-level":[125],"uneven":[126],"partitioning":[127],"strategy":[128],"balance":[130],"PP":[133],"(2)":[135],"a":[136,148,168],"memory-aware":[137],"replication":[139],"scheme":[140],"(DPMoE)":[141],"that":[142],"communication":[144],"overhead,":[145],"and":[146,155,179,183,197,208,216,222,228],"(3)":[147],"lightweight":[149],"search":[150,154],"combining":[152],"binary":[153],"programming":[157],"generate":[159],"plans.":[162],"implement":[164],"SGLang,":[167],"state-of-the-art":[169],"LLM":[170],"framework,":[172],"evaluate":[173],"it":[174],"32":[176],"A40":[177],"GPUs":[178],"16":[180],"A100":[181],"GPUs,":[182],"compare":[184],"manually":[186],"tuned":[187],"strategies.":[189],"Experimental":[190],"results":[191],"show":[192],"that,":[193],"when":[194],"co-locating":[195],"prefill":[196,215],"decoding":[198,217],"phases,":[199],"achieves":[201],"1.20\u20133.13x":[202],"throughput":[203],"improvements":[204],"prefill-only":[206],"tasks":[207],"1.05\u20131.89x":[209],"prefill-decoding":[211],"tasks.":[212],"When":[213],"disaggregating":[214],"tasks,":[218],"improves":[220],"average":[221],"P99":[223],"time-to-first-token":[224],"(TTFT)":[225],"1.10\u20131.17x":[227],"1.21\u20131.26x,":[229],"respectively.":[230]},"counts_by_year":[],"updated_date":"2026-06-08T08:47:23.578185","created_date":"2026-03-18T00:00:00"}