{"id":"https://openalex.org/W4407197341","doi":"https://doi.org/10.1145/3669940.3707267","title":"MoE-L ightning : High-Throughput MoE Inference on Memory-constrained GPUs","display_name":"MoE-L ightning : High-Throughput MoE Inference on Memory-constrained GPUs","publication_year":2025,"publication_date":"2025-02-06","ids":{"openalex":"https://openalex.org/W4407197341","doi":"https://doi.org/10.1145/3669940.3707267"},"language":"en","primary_location":{"id":"doi:10.1145/3669940.3707267","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3669940.3707267","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 30th ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 1","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://doi.org/10.1145/3669940.3707267","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5108829669","display_name":"Shiyi Cao","orcid":"https://orcid.org/0009-0005-6834-375X"},"institutions":[{"id":"https://openalex.org/I134446601","display_name":"Berkeley College","ror":"https://ror.org/02xewxa75","country_code":"US","type":"education","lineage":["https://openalex.org/I134446601"]},{"id":"https://openalex.org/I95457486","display_name":"University of California, Berkeley","ror":"https://ror.org/01an7q238","country_code":"US","type":"education","lineage":["https://openalex.org/I95457486"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Shiyi Cao","raw_affiliation_strings":["UC Berkeley, Berkeley, CA, USA"],"raw_orcid":"https://orcid.org/0009-0005-6834-375X","affiliations":[{"raw_affiliation_string":"UC Berkeley, Berkeley, CA, USA","institution_ids":["https://openalex.org/I134446601","https://openalex.org/I95457486"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5102028268","display_name":"Shu Liu","orcid":"https://orcid.org/0009-0009-0825-6178"},"institutions":[{"id":"https://openalex.org/I134446601","display_name":"Berkeley College","ror":"https://ror.org/02xewxa75","country_code":"US","type":"education","lineage":["https://openalex.org/I134446601"]},{"id":"https://openalex.org/I95457486","display_name":"University of California, Berkeley","ror":"https://ror.org/01an7q238","country_code":"US","type":"education","lineage":["https://openalex.org/I95457486"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Shu Liu","raw_affiliation_strings":["UC Berkeley, Berkeley, CA, USA"],"raw_orcid":"https://orcid.org/0009-0009-0825-6178","affiliations":[{"raw_affiliation_string":"UC Berkeley, Berkeley, CA, USA","institution_ids":["https://openalex.org/I134446601","https://openalex.org/I95457486"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5095886747","display_name":"Tyler Griggs","orcid":null},"institutions":[{"id":"https://openalex.org/I134446601","display_name":"Berkeley College","ror":"https://ror.org/02xewxa75","country_code":"US","type":"education","lineage":["https://openalex.org/I134446601"]},{"id":"https://openalex.org/I95457486","display_name":"University of California, Berkeley","ror":"https://ror.org/01an7q238","country_code":"US","type":"education","lineage":["https://openalex.org/I95457486"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Tyler Griggs","raw_affiliation_strings":["UC Berkeley, Berkeley, CA, USA"],"raw_orcid":"https://orcid.org/0009-0006-4655-0229","affiliations":[{"raw_affiliation_string":"UC Berkeley, Berkeley, CA, USA","institution_ids":["https://openalex.org/I134446601","https://openalex.org/I95457486"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5070342607","display_name":"Peter Schafhalter","orcid":null},"institutions":[{"id":"https://openalex.org/I134446601","display_name":"Berkeley College","ror":"https://ror.org/02xewxa75","country_code":"US","type":"education","lineage":["https://openalex.org/I134446601"]},{"id":"https://openalex.org/I95457486","display_name":"University of California, Berkeley","ror":"https://ror.org/01an7q238","country_code":"US","type":"education","lineage":["https://openalex.org/I95457486"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Peter Schafhalter","raw_affiliation_strings":["UC Berkeley, Berkeley, USA"],"raw_orcid":"https://orcid.org/0009-0007-9865-0456","affiliations":[{"raw_affiliation_string":"UC Berkeley, Berkeley, USA","institution_ids":["https://openalex.org/I134446601","https://openalex.org/I95457486"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5027378038","display_name":"Xiaoxuan Liu","orcid":null},"institutions":[{"id":"https://openalex.org/I134446601","display_name":"Berkeley College","ror":"https://ror.org/02xewxa75","country_code":"US","type":"education","lineage":["https://openalex.org/I134446601"]},{"id":"https://openalex.org/I95457486","display_name":"University of California, Berkeley","ror":"https://ror.org/01an7q238","country_code":"US","type":"education","lineage":["https://openalex.org/I95457486"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Xiaoxuan Liu","raw_affiliation_strings":["UC Berkeley, Berkeley, USA"],"raw_orcid":"https://orcid.org/0009-0004-4968-1027","affiliations":[{"raw_affiliation_string":"UC Berkeley, Berkeley, USA","institution_ids":["https://openalex.org/I134446601","https://openalex.org/I95457486"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5083402013","display_name":"Ying Sheng","orcid":null},"institutions":[{"id":"https://openalex.org/I1743320","display_name":"Palo Alto University","ror":"https://ror.org/04f812k67","country_code":"US","type":"education","lineage":["https://openalex.org/I1743320"]},{"id":"https://openalex.org/I97018004","display_name":"Stanford University","ror":"https://ror.org/00f54p054","country_code":"US","type":"education","lineage":["https://openalex.org/I97018004"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Ying Sheng","raw_affiliation_strings":["Stanford University, Palo Alto, USA"],"raw_orcid":"https://orcid.org/0000-0002-1883-2126","affiliations":[{"raw_affiliation_string":"Stanford University, Palo Alto, USA","institution_ids":["https://openalex.org/I1743320","https://openalex.org/I97018004"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5072427753","display_name":"Joseph E. Gonzalez","orcid":"https://orcid.org/0000-0003-2921-956X"},"institutions":[{"id":"https://openalex.org/I134446601","display_name":"Berkeley College","ror":"https://ror.org/02xewxa75","country_code":"US","type":"education","lineage":["https://openalex.org/I134446601"]},{"id":"https://openalex.org/I95457486","display_name":"University of California, Berkeley","ror":"https://ror.org/01an7q238","country_code":"US","type":"education","lineage":["https://openalex.org/I95457486"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Joseph E. Gonzalez","raw_affiliation_strings":["UC Berkeley, Berkeley, USA"],"raw_orcid":"https://orcid.org/0000-0003-2921-956X","affiliations":[{"raw_affiliation_string":"UC Berkeley, Berkeley, USA","institution_ids":["https://openalex.org/I134446601","https://openalex.org/I95457486"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5005554337","display_name":"Matei Zaharia","orcid":"https://orcid.org/0000-0002-7547-7204"},"institutions":[{"id":"https://openalex.org/I134446601","display_name":"Berkeley College","ror":"https://ror.org/02xewxa75","country_code":"US","type":"education","lineage":["https://openalex.org/I134446601"]},{"id":"https://openalex.org/I95457486","display_name":"University of California, Berkeley","ror":"https://ror.org/01an7q238","country_code":"US","type":"education","lineage":["https://openalex.org/I95457486"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Matei Zaharia","raw_affiliation_strings":["UC Berkeley, Berkeley, USA"],"raw_orcid":"https://orcid.org/0000-0002-7547-7204","affiliations":[{"raw_affiliation_string":"UC Berkeley, Berkeley, USA","institution_ids":["https://openalex.org/I134446601","https://openalex.org/I95457486"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5041920173","display_name":"Ion Stoica","orcid":"https://orcid.org/0000-0002-5373-0088"},"institutions":[{"id":"https://openalex.org/I134446601","display_name":"Berkeley College","ror":"https://ror.org/02xewxa75","country_code":"US","type":"education","lineage":["https://openalex.org/I134446601"]},{"id":"https://openalex.org/I95457486","display_name":"University of California, Berkeley","ror":"https://ror.org/01an7q238","country_code":"US","type":"education","lineage":["https://openalex.org/I95457486"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Ion Stoica","raw_affiliation_strings":["UC Berkeley, Berkeley, USA"],"raw_orcid":"https://orcid.org/0000-0002-5373-0088","affiliations":[{"raw_affiliation_string":"UC Berkeley, Berkeley, USA","institution_ids":["https://openalex.org/I134446601","https://openalex.org/I95457486"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":4,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":14.8912,"has_fulltext":false,"cited_by_count":12,"citation_normalized_percentile":{"value":0.99463327,"is_in_top_1_percent":true,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":95,"max":100},"biblio":{"volume":null,"issue":null,"first_page":"715","last_page":"730"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.9998999834060669,"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":0.9998999834060669,"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":0.9972000122070312,"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/T11181","display_name":"Advanced Data Storage Technologies","score":0.9959999918937683,"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":0.6605454087257385},{"id":"https://openalex.org/keywords/throughput","display_name":"Throughput","score":0.6203064918518066},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.5543453693389893},{"id":"https://openalex.org/keywords/parallel-computing","display_name":"Parallel computing","score":0.4196258783340454},{"id":"https://openalex.org/keywords/operating-system","display_name":"Operating system","score":0.19492453336715698},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.1267811357975006}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6605454087257385},{"id":"https://openalex.org/C157764524","wikidata":"https://www.wikidata.org/wiki/Q1383412","display_name":"Throughput","level":3,"score":0.6203064918518066},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5543453693389893},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.4196258783340454},{"id":"https://openalex.org/C111919701","wikidata":"https://www.wikidata.org/wiki/Q9135","display_name":"Operating system","level":1,"score":0.19492453336715698},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.1267811357975006},{"id":"https://openalex.org/C555944384","wikidata":"https://www.wikidata.org/wiki/Q249","display_name":"Wireless","level":2,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1145/3669940.3707267","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3669940.3707267","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 30th ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 1","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.1145/3669940.3707267","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3669940.3707267","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 30th ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 1","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":42,"referenced_works":["https://openalex.org/W569478347","https://openalex.org/W2002555321","https://openalex.org/W2025653905","https://openalex.org/W2150884987","https://openalex.org/W2890152612","https://openalex.org/W2901299405","https://openalex.org/W2969388332","https://openalex.org/W2970971581","https://openalex.org/W3126805032","https://openalex.org/W3132107458","https://openalex.org/W3204998121","https://openalex.org/W4200634402","https://openalex.org/W4281758439","https://openalex.org/W4289302788","https://openalex.org/W4308760184","https://openalex.org/W4310561894","https://openalex.org/W4310895557","https://openalex.org/W4319166707","https://openalex.org/W4321636575","https://openalex.org/W4323654151","https://openalex.org/W4324297016","https://openalex.org/W4377864080","https://openalex.org/W4380353763","https://openalex.org/W4382319938","https://openalex.org/W4385571586","https://openalex.org/W4387321091","https://openalex.org/W4390041201","https://openalex.org/W4390091485","https://openalex.org/W4390438324","https://openalex.org/W4390833687","https://openalex.org/W4391272939","https://openalex.org/W4391670445","https://openalex.org/W4391800686","https://openalex.org/W4392222465","https://openalex.org/W4392736881","https://openalex.org/W4392974063","https://openalex.org/W4396815596","https://openalex.org/W4404401018","https://openalex.org/W6761628794","https://openalex.org/W6846659131","https://openalex.org/W6852874933","https://openalex.org/W6862282697"],"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":{"Efficient":[0],"deployment":[1],"of":[2,8,20],"large":[3,56],"language":[4],"models,":[5,11],"particularly":[6],"Mixture":[7],"Experts":[9],"(MoE)":[10],"on":[12,107,141,189],"resource-constrained":[13],"platforms":[14],"presents":[15],"significant":[16],"challenges":[17],"in":[18,41],"terms":[19],"computational":[21],"efficiency":[22],"and":[23,101,187],"memory":[24],"utilization.":[25,173],"The":[26],"MoE":[27,60,75],"architecture,":[28],"renowned":[29],"for":[30,138,180],"its":[31],"ability":[32],"to":[33,63,96,114,128],"increase":[34,40],"model":[35,57],"capacity":[36],"without":[37,65],"a":[38,73,87,102,108,142],"proportional":[39],"inference":[42,77,136,179],"cost,":[43],"greatly":[44],"reduces":[45],"the":[46,55,148,155,161],"token":[47],"generation":[48],"latency":[49],"compared":[50],"with":[51,93,118,165],"dense":[52],"models.":[53],"However,":[54],"size":[58],"makes":[59],"models":[61],"inaccessible":[62],"individuals":[64],"high-end":[66],"GPUs.":[67],"In":[68],"this":[69],"paper,":[70],"we":[71,112],"propose":[72],"high-throughput":[74],"batch":[76,178],"system,":[78],"MoE-Lightning,":[79],"that":[80],"significantly":[81,170],"outperforms":[82],"past":[83],"work.":[84],"MoE-Lightning":[85,124,158,174],"introduces":[86],"novel":[88],"CPU-GPU-I/O":[89],"pipelining":[90],"schedule,":[91],"CGOPipe,":[92],"paged":[94],"weights":[95],"achieve":[97,126],"high":[98],"resource":[99,172],"utilization,":[100],"performance":[103],"model,":[104],"HRM,":[105],"based":[106],"Hierarchical":[109],"Roofline":[110],"Model":[111],"introduce":[113],"help":[115],"find":[116],"policies":[117],"higher":[119,130],"throughput":[120,131,151,162],"than":[121,132],"existing":[122],"systems.":[123],"can":[125,159],"up":[127],"(10.3x)":[129],"state-of-the-art":[133],"offloading-enabled":[134],"LLM":[135],"systems":[137],"Mixtral":[139,185],"8x7B":[140],"single":[143],"T4":[144],"GPU":[145,156],"(16GB).":[146],"When":[147],"theoretical":[149],"system":[150],"is":[152],"bounded":[153],"by":[154],"memory,":[157,169],"reach":[160],"upper":[163],"bound":[164],"2-3x":[166],"less":[167],"CPU":[168],"increasing":[171],"also":[175],"supports":[176],"efficient":[177],"much":[181],"larger":[182],"MoEs":[183],"(e.g.,":[184,193],"8x22B":[186],"DBRX)":[188],"multiple":[190],"low-cost":[191],"GPUs":[192],"2--4":[194],"T4s).":[195]},"counts_by_year":[{"year":2026,"cited_by_count":10},{"year":2025,"cited_by_count":2}],"updated_date":"2026-08-21T09:56:20.448147","created_date":"2025-10-10T00:00:00"}
