{"id":"https://openalex.org/W4364382874","doi":"https://doi.org/10.1145/3588964","title":"FlexMoE: Scaling Large-scale Sparse Pre-trained Model Training via Dynamic Device Placement","display_name":"FlexMoE: Scaling Large-scale Sparse Pre-trained Model Training via Dynamic Device Placement","publication_year":2023,"publication_date":"2023-05-26","ids":{"openalex":"https://openalex.org/W4364382874","doi":"https://doi.org/10.1145/3588964"},"language":"en","primary_location":{"id":"doi:10.1145/3588964","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3588964","pdf_url":"https://dl.acm.org/doi/pdf/10.1145/3588964","source":{"id":"https://openalex.org/S4387289859","display_name":"Proceedings of the ACM on Management of Data","issn_l":"2836-6573","issn":["2836-6573"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":"https://openalex.org/P4310319798","host_organization_name":"Association for Computing Machinery","host_organization_lineage":["https://openalex.org/P4310319798"],"host_organization_lineage_names":["Association for Computing Machinery"],"type":"journal"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the ACM on Management of Data","raw_type":"journal-article"},"type":"article","indexed_in":["arxiv","crossref"],"open_access":{"is_oa":true,"oa_status":"bronze","oa_url":"https://dl.acm.org/doi/pdf/10.1145/3588964","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5059601307","display_name":"Xiaonan Nie","orcid":"https://orcid.org/0000-0001-6766-757X"},"institutions":[{"id":"https://openalex.org/I20231570","display_name":"Peking University","ror":"https://ror.org/02v51f717","country_code":"CN","type":"education","lineage":["https://openalex.org/I20231570"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Xiaonan Nie","raw_affiliation_strings":["Peking University, Beijing, China"],"raw_orcid":"https://orcid.org/0000-0001-6766-757X","affiliations":[{"raw_affiliation_string":"Peking University, Beijing, China","institution_ids":["https://openalex.org/I20231570"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5015552951","display_name":"Xupeng Miao","orcid":"https://orcid.org/0000-0002-9371-8358"},"institutions":[{"id":"https://openalex.org/I74973139","display_name":"Carnegie Mellon University","ror":"https://ror.org/05x2bcf33","country_code":"US","type":"education","lineage":["https://openalex.org/I74973139"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Xupeng Miao","raw_affiliation_strings":["Carnegie Mellon University, Pittsburgh, PA, USA"],"raw_orcid":"https://orcid.org/0000-0002-9371-8358","affiliations":[{"raw_affiliation_string":"Carnegie Mellon University, Pittsburgh, PA, USA","institution_ids":["https://openalex.org/I74973139"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100384118","display_name":"Zilong Wang","orcid":"https://orcid.org/0009-0000-4026-0104"},"institutions":[{"id":"https://openalex.org/I4210113369","display_name":"Microsoft Research Asia (China)","ror":"https://ror.org/0300m5276","country_code":"CN","type":"company","lineage":["https://openalex.org/I1290206253","https://openalex.org/I4210113369"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Zilong Wang","raw_affiliation_strings":["Microsoft, Beijing, China"],"raw_orcid":"https://orcid.org/0009-0000-4026-0104","affiliations":[{"raw_affiliation_string":"Microsoft, Beijing, China","institution_ids":["https://openalex.org/I4210113369"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101461004","display_name":"Zichao Yang","orcid":"https://orcid.org/0009-0001-7734-5766"},"institutions":[{"id":"https://openalex.org/I74973139","display_name":"Carnegie Mellon University","ror":"https://ror.org/05x2bcf33","country_code":"US","type":"education","lineage":["https://openalex.org/I74973139"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Zichao Yang","raw_affiliation_strings":["Carnegie Mellon University, Pittsburgh, PA, China"],"raw_orcid":"https://orcid.org/0009-0001-7734-5766","affiliations":[{"raw_affiliation_string":"Carnegie Mellon University, Pittsburgh, PA, China","institution_ids":["https://openalex.org/I74973139"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101635748","display_name":"Jilong Xue","orcid":"https://orcid.org/0000-0002-4495-1997"},"institutions":[{"id":"https://openalex.org/I4210113369","display_name":"Microsoft Research Asia (China)","ror":"https://ror.org/0300m5276","country_code":"CN","type":"company","lineage":["https://openalex.org/I1290206253","https://openalex.org/I4210113369"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Jilong Xue","raw_affiliation_strings":["Microsoft Research, Beijing, China"],"raw_orcid":"https://orcid.org/0000-0002-4495-1997","affiliations":[{"raw_affiliation_string":"Microsoft Research, Beijing, China","institution_ids":["https://openalex.org/I4210113369"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5102874660","display_name":"Lingxiao Ma","orcid":"https://orcid.org/0009-0009-9524-5476"},"institutions":[{"id":"https://openalex.org/I4210113369","display_name":"Microsoft Research Asia (China)","ror":"https://ror.org/0300m5276","country_code":"CN","type":"company","lineage":["https://openalex.org/I1290206253","https://openalex.org/I4210113369"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Lingxiao Ma","raw_affiliation_strings":["Microsoft Research, Beijing, China"],"raw_orcid":"https://orcid.org/0009-0009-9524-5476","affiliations":[{"raw_affiliation_string":"Microsoft Research, Beijing, China","institution_ids":["https://openalex.org/I4210113369"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Gang Cao","orcid":"https://orcid.org/0009-0005-1142-0128"},"institutions":[{"id":"https://openalex.org/I4210100255","display_name":"Beijing Academy of Artificial Intelligence","ror":"https://ror.org/016a74861","country_code":"CN","type":"other","lineage":["https://openalex.org/I4210100255"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Gang Cao","raw_affiliation_strings":["Beijing Academy of Artificial Intelligence, Beijing, China"],"raw_orcid":"https://orcid.org/0009-0005-1142-0128","affiliations":[{"raw_affiliation_string":"Beijing Academy of Artificial Intelligence, Beijing, China","institution_ids":["https://openalex.org/I4210100255"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5062357883","display_name":"Bin Cui","orcid":"https://orcid.org/0000-0003-1681-4677"},"institutions":[{"id":"https://openalex.org/I20231570","display_name":"Peking University","ror":"https://ror.org/02v51f717","country_code":"CN","type":"education","lineage":["https://openalex.org/I20231570"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Bin Cui","raw_affiliation_strings":["Peking University, Beijing, China"],"raw_orcid":"https://orcid.org/0000-0003-1681-4677","affiliations":[{"raw_affiliation_string":"Peking University, Beijing, China","institution_ids":["https://openalex.org/I20231570"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":4,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":3.489,"has_fulltext":true,"cited_by_count":36,"citation_normalized_percentile":{"value":0.94198231,"is_in_top_1_percent":false,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":97,"max":100},"biblio":{"volume":"1","issue":"1","first_page":"1","last_page":"19"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.9976000189781189,"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":0.9976000189781189,"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/T12127","display_name":"Software System Performance and Reliability","score":0.9975000023841858,"subfield":{"id":"https://openalex.org/subfields/1705","display_name":"Computer Networks and Communications"},"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/T10273","display_name":"IoT and Edge/Fog Computing","score":0.9940999746322632,"subfield":{"id":"https://openalex.org/subfields/1705","display_name":"Computer Networks and Communications"},"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/computer-science","display_name":"Computer science","score":0.840087890625},{"id":"https://openalex.org/keywords/dataflow","display_name":"Dataflow","score":0.7764114141464233},{"id":"https://openalex.org/keywords/scalability","display_name":"Scalability","score":0.7199811935424805},{"id":"https://openalex.org/keywords/scheduling","display_name":"Scheduling (production processes)","score":0.5642809271812439},{"id":"https://openalex.org/keywords/computation","display_name":"Computation","score":0.4586292803287506},{"id":"https://openalex.org/keywords/heuristic","display_name":"Heuristic","score":0.45669347047805786},{"id":"https://openalex.org/keywords/machine-learning","display_name":"Machine learning","score":0.4560229480266571},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.4528749883174896},{"id":"https://openalex.org/keywords/distributed-computing","display_name":"Distributed computing","score":0.3763021230697632},{"id":"https://openalex.org/keywords/parallel-computing","display_name":"Parallel computing","score":0.2266712486743927},{"id":"https://openalex.org/keywords/database","display_name":"Database","score":0.12485554814338684},{"id":"https://openalex.org/keywords/algorithm","display_name":"Algorithm","score":0.12256738543510437}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.840087890625},{"id":"https://openalex.org/C96324660","wikidata":"https://www.wikidata.org/wiki/Q205446","display_name":"Dataflow","level":2,"score":0.7764114141464233},{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.7199811935424805},{"id":"https://openalex.org/C206729178","wikidata":"https://www.wikidata.org/wiki/Q2271896","display_name":"Scheduling (production processes)","level":2,"score":0.5642809271812439},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.4586292803287506},{"id":"https://openalex.org/C173801870","wikidata":"https://www.wikidata.org/wiki/Q201413","display_name":"Heuristic","level":2,"score":0.45669347047805786},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4560229480266571},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4528749883174896},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.3763021230697632},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.2266712486743927},{"id":"https://openalex.org/C77088390","wikidata":"https://www.wikidata.org/wiki/Q8513","display_name":"Database","level":1,"score":0.12485554814338684},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.12256738543510437},{"id":"https://openalex.org/C162324750","wikidata":"https://www.wikidata.org/wiki/Q8134","display_name":"Economics","level":0,"score":0.0},{"id":"https://openalex.org/C21547014","wikidata":"https://www.wikidata.org/wiki/Q1423657","display_name":"Operations management","level":1,"score":0.0}],"mesh":[],"locations_count":2,"locations":[{"id":"doi:10.1145/3588964","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3588964","pdf_url":"https://dl.acm.org/doi/pdf/10.1145/3588964","source":{"id":"https://openalex.org/S4387289859","display_name":"Proceedings of the ACM on Management of Data","issn_l":"2836-6573","issn":["2836-6573"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":"https://openalex.org/P4310319798","host_organization_name":"Association for Computing Machinery","host_organization_lineage":["https://openalex.org/P4310319798"],"host_organization_lineage_names":["Association for Computing Machinery"],"type":"journal"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the ACM on Management of Data","raw_type":"journal-article"},{"id":"pmh:oai:arXiv.org:2304.03946","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2304.03946","pdf_url":"https://arxiv.org/pdf/2304.03946","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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"}],"best_oa_location":{"id":"doi:10.1145/3588964","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3588964","pdf_url":"https://dl.acm.org/doi/pdf/10.1145/3588964","source":{"id":"https://openalex.org/S4387289859","display_name":"Proceedings of the ACM on Management of Data","issn_l":"2836-6573","issn":["2836-6573"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":"https://openalex.org/P4310319798","host_organization_name":"Association for Computing Machinery","host_organization_lineage":["https://openalex.org/P4310319798"],"host_organization_lineage_names":["Association for Computing Machinery"],"type":"journal"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the ACM on Management of Data","raw_type":"journal-article"},"sustainable_development_goals":[{"display_name":"Responsible consumption and production","id":"https://metadata.un.org/sdg/12","score":0.550000011920929}],"awards":[{"id":"https://openalex.org/G1691401027","display_name":null,"funder_award_id":"U22B2037","funder_id":"https://openalex.org/F4320321001","funder_display_name":"National Natural Science Foundation of China"},{"id":"https://openalex.org/G4253300795","display_name":null,"funder_award_id":"61832001","funder_id":"https://openalex.org/F4320321001","funder_display_name":"National Natural Science Foundation of China"},{"id":"https://openalex.org/G8361467185","display_name":null,"funder_award_id":"2020AAA0105200","funder_id":"https://openalex.org/F4320335777","funder_display_name":"National Key Research and Development Program of China"}],"funders":[{"id":"https://openalex.org/F4320316083","display_name":"Tencent","ror":"https://ror.org/00hhjss72"},{"id":"https://openalex.org/F4320321001","display_name":"National Natural Science Foundation of China","ror":"https://ror.org/01h0zpd94"},{"id":"https://openalex.org/F4320335777","display_name":"National Key Research and Development Program of China","ror":null}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4364382874.pdf","grobid_xml":"https://content.openalex.org/works/W4364382874.grobid-xml"},"referenced_works_count":51,"referenced_works":["https://openalex.org/W2781528640","https://openalex.org/W2896457183","https://openalex.org/W2943552823","https://openalex.org/W2954558120","https://openalex.org/W2969388332","https://openalex.org/W2981852735","https://openalex.org/W2991040477","https://openalex.org/W3001279689","https://openalex.org/W3034608272","https://openalex.org/W3037585619","https://openalex.org/W3040573126","https://openalex.org/W3086105743","https://openalex.org/W3094502228","https://openalex.org/W3119866685","https://openalex.org/W3124190522","https://openalex.org/W3138516171","https://openalex.org/W3147874613","https://openalex.org/W3155426438","https://openalex.org/W3158196591","https://openalex.org/W3170796112","https://openalex.org/W3189646782","https://openalex.org/W3204998121","https://openalex.org/W4200634402","https://openalex.org/W4220838824","https://openalex.org/W4220967350","https://openalex.org/W4221150592","https://openalex.org/W4224299101","https://openalex.org/W4225390052","https://openalex.org/W4226328099","https://openalex.org/W4226399778","https://openalex.org/W4226515448","https://openalex.org/W4281922990","https://openalex.org/W4282576620","https://openalex.org/W4283073926","https://openalex.org/W4287121196","https://openalex.org/W4287319715","https://openalex.org/W4287391717","https://openalex.org/W4288089799","https://openalex.org/W4289446312","https://openalex.org/W4289533987","https://openalex.org/W4292779060","https://openalex.org/W4293718192","https://openalex.org/W4295312788","https://openalex.org/W4296143854","https://openalex.org/W4307156798","https://openalex.org/W4310282800","https://openalex.org/W4311209912","https://openalex.org/W4385245566","https://openalex.org/W4386768680","https://openalex.org/W6796487566","https://openalex.org/W6796854725"],"related_works":["https://openalex.org/W2293118914","https://openalex.org/W2998381397","https://openalex.org/W4236419692","https://openalex.org/W3167919718","https://openalex.org/W4251718783","https://openalex.org/W2171015181","https://openalex.org/W4239447582","https://openalex.org/W2150836909","https://openalex.org/W2783065147","https://openalex.org/W4312723892"],"abstract_inverted_index":{"With":[0],"the":[1,16,79,95,128,141,154,177,183,187,193,198,213],"increasing":[2],"data":[3,91,96,184,200],"volume,":[4],"there":[5],"is":[6,30,208],"a":[7,38,64,89,117,161,172],"trend":[8],"of":[9,22,27,32,34,41,145],"using":[10],"large-scale":[11],"pre-trained":[12],"models":[13,29,225,232],"to":[14,78,152,181,210,260,271],"store":[15],"knowledge":[17],"into":[18],"an":[19,137],"enormous":[20],"number":[21],"model":[23],"parameters.":[24],"The":[25],"training":[26,120,146],"these":[28],"composed":[31],"lots":[33],"dense":[35],"algebras,":[36],"requiring":[37],"huge":[39],"amount":[40],"hardware":[42],"resources.":[43],"Recently,":[44],"sparsely-gated":[45],"Mixture-of-Experts":[46],"(MoEs)":[47],"are":[48,87],"becoming":[49,88],"more":[50],"popular":[51],"and":[52,82,99,107,125,143,157,165,190,228,230,258,262,269],"have":[53,219],"demonstrated":[54],"impressive":[55],"pretraining":[56],"scalability":[57],"in":[58,74,94],"various":[59],"downstream":[60],"tasks.":[61],"However,":[62],"such":[63],"sparse":[65],"conditional":[66],"computation":[67],"may":[68],"not":[69],"be":[70],"effective":[71],"as":[72],"expected":[73],"practical":[75],"systems":[76,246],"due":[77],"routing":[80,155],"imbalance":[81,156],"fluctuation":[83,158],"problems.":[84],"Generally,":[85],"MoEs":[86],"new":[90],"analytics":[92],"paradigm":[93],"life":[97],"cycle":[98],"suffering":[100],"from":[101],"unique":[102],"challenges":[103],"at":[104],"scales,":[105],"complexities,":[106],"granularities":[108],"never":[109],"before":[110],"possible.":[111],"In":[112],"this":[113],"paper,":[114],"we":[115,170],"propose":[116],"novel":[118,173],"DNN":[119,179],"framework,":[121],"FlexMoE,":[122],"which":[123,149],"systematically":[124],"transparently":[126],"address":[127],"inefficiency":[129],"caused":[130],"by":[131,160,197,254,265],"dynamic":[132,162],"dataflow.":[133],"We":[134,218],"first":[135],"present":[136],"empirical":[138],"analysis":[139],"on":[140,222,247,256,267],"problems":[142,159],"opportunities":[144],"MoE":[147],"models,":[148],"motivates":[150],"us":[151],"overcome":[153],"expert":[163],"management":[164],"device":[166,214],"placement":[167,215],"mechanism.":[168],"Then":[169],"introduce":[171],"scheduling":[174,188],"module":[175],"over":[176],"existing":[178,245],"runtime":[180],"monitor":[182],"flow,":[185],"make":[186],"plans,":[189],"dynamically":[191,211],"adjust":[192],"model-to-hardware":[194],"mapping":[195],"guided":[196],"real-time":[199],"traffic.":[201],"A":[202],"simple":[203],"but":[204],"efficient":[205],"heuristic":[206],"algorithm":[207],"exploited":[209],"optimize":[212],"during":[216],"training.":[217],"conducted":[220],"experiments":[221],"both":[223],"NLP":[224],"(e.g.,":[226,233],"BERT":[227],"GPT)":[229],"vision":[231],"Swin).":[234],"And":[235],"results":[236],"show":[237],"FlexMoE":[238,251],"can":[239],"achieve":[240],"superior":[241],"performance":[242],"compared":[243],"with":[244],"real-world":[248],"workloads":[249],"---":[250],"outperforms":[252,263],"DeepSpeed":[253],"1.70x":[255],"average":[257,268],"up":[259,270],"2.10x,":[261],"FasterMoE":[264],"1.30x":[266],"1.45x.":[272]},"counts_by_year":[{"year":2026,"cited_by_count":4},{"year":2025,"cited_by_count":22},{"year":2024,"cited_by_count":5},{"year":2023,"cited_by_count":5}],"updated_date":"2026-07-31T08:31:51.225901","created_date":"2023-04-12T00:00:00"}