{"id":"https://openalex.org/W4416517630","doi":"https://doi.org/10.1145/3649601.3698722","title":"Towards Building Private LLMs: Exploring Multi-Node Expert Parallelism on Apple Silicon for Mixture-of-Experts Large Language Model","display_name":"Towards Building Private LLMs: Exploring Multi-Node Expert Parallelism on Apple Silicon for Mixture-of-Experts Large Language Model","publication_year":2025,"publication_date":"2025-06-30","ids":{"openalex":"https://openalex.org/W4416517630","doi":"https://doi.org/10.1145/3649601.3698722"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2506.23635","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2506.23635","pdf_url":"https://arxiv.org/pdf/2506.23635","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"},"type":"preprint","indexed_in":["arxiv"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2506.23635","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5029870143","display_name":"Mu-Chi Chen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Mu-Chi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5033413049","display_name":"Po-Yu Huang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Huang, Po-Hsuan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5120419728","display_name":"Xiangrui Ke","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ke, Xiangrui","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5012739047","display_name":"Chia-Heng Tu","orcid":"https://orcid.org/0000-0001-8967-1385"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tu, Chia-Heng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101441768","display_name":"Chun Jason Xue","orcid":"https://orcid.org/0000-0002-6431-9868"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xue, Chun Jason","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5020028710","display_name":"Shih\u2010Hao Hung","orcid":"https://orcid.org/0000-0003-2043-2663"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hung, Shih-Hao","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/T11704","display_name":"Mobile Crowdsensing and Crowdsourcing","score":0.22349999845027924,"subfield":{"id":"https://openalex.org/subfields/1706","display_name":"Computer Science Applications"},"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/T11704","display_name":"Mobile Crowdsensing and Crowdsourcing","score":0.22349999845027924,"subfield":{"id":"https://openalex.org/subfields/1706","display_name":"Computer Science Applications"},"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/T11636","display_name":"Artificial Intelligence in Healthcare and Education","score":0.17990000545978546,"subfield":{"id":"https://openalex.org/subfields/2718","display_name":"Health Informatics"},"field":{"id":"https://openalex.org/fields/27","display_name":"Medicine"},"domain":{"id":"https://openalex.org/domains/4","display_name":"Health Sciences"}},{"id":"https://openalex.org/T14347","display_name":"Big Data and Digital Economy","score":0.07959999889135361,"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/scalability","display_name":"Scalability","score":0.7093999981880188},{"id":"https://openalex.org/keywords/construct","display_name":"Construct (python library)","score":0.6197999715805054},{"id":"https://openalex.org/keywords/host","display_name":"Host (biology)","score":0.5527999997138977},{"id":"https://openalex.org/keywords/supercomputer","display_name":"Supercomputer","score":0.5220999717712402},{"id":"https://openalex.org/keywords/overhead","display_name":"Overhead (engineering)","score":0.5152000188827515},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.4984000027179718},{"id":"https://openalex.org/keywords/software","display_name":"Software","score":0.474700003862381},{"id":"https://openalex.org/keywords/latency","display_name":"Latency (audio)","score":0.44929999113082886},{"id":"https://openalex.org/keywords/gpu-cluster","display_name":"GPU cluster","score":0.38850000500679016}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8345000147819519},{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.7093999981880188},{"id":"https://openalex.org/C2780801425","wikidata":"https://www.wikidata.org/wiki/Q5164392","display_name":"Construct (python library)","level":2,"score":0.6197999715805054},{"id":"https://openalex.org/C126831891","wikidata":"https://www.wikidata.org/wiki/Q221673","display_name":"Host (biology)","level":2,"score":0.5527999997138977},{"id":"https://openalex.org/C83283714","wikidata":"https://www.wikidata.org/wiki/Q121117","display_name":"Supercomputer","level":2,"score":0.5220999717712402},{"id":"https://openalex.org/C2779960059","wikidata":"https://www.wikidata.org/wiki/Q7113681","display_name":"Overhead (engineering)","level":2,"score":0.5152000188827515},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.4984000027179718},{"id":"https://openalex.org/C2777904410","wikidata":"https://www.wikidata.org/wiki/Q7397","display_name":"Software","level":2,"score":0.474700003862381},{"id":"https://openalex.org/C82876162","wikidata":"https://www.wikidata.org/wiki/Q17096504","display_name":"Latency (audio)","level":2,"score":0.44929999113082886},{"id":"https://openalex.org/C2781335571","wikidata":"https://www.wikidata.org/wiki/Q2633544","display_name":"GPU cluster","level":3,"score":0.38850000500679016},{"id":"https://openalex.org/C29140674","wikidata":"https://www.wikidata.org/wiki/Q206637","display_name":"Computer cluster","level":2,"score":0.3723999857902527},{"id":"https://openalex.org/C46637626","wikidata":"https://www.wikidata.org/wiki/Q6693015","display_name":"Low latency (capital markets)","level":2,"score":0.3668999969959259},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.35580000281333923},{"id":"https://openalex.org/C34165917","wikidata":"https://www.wikidata.org/wiki/Q188267","display_name":"Programming paradigm","level":2,"score":0.3441999852657318},{"id":"https://openalex.org/C115903868","wikidata":"https://www.wikidata.org/wiki/Q80993","display_name":"Software engineering","level":1,"score":0.336899995803833},{"id":"https://openalex.org/C60832428","wikidata":"https://www.wikidata.org/wiki/Q945818","display_name":"Partitioned global address space","level":3,"score":0.33329999446868896},{"id":"https://openalex.org/C45012715","wikidata":"https://www.wikidata.org/wiki/Q746628","display_name":"Studio","level":2,"score":0.3147999942302704},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.31220000982284546},{"id":"https://openalex.org/C118524514","wikidata":"https://www.wikidata.org/wiki/Q173212","display_name":"Computer architecture","level":1,"score":0.31130000948905945},{"id":"https://openalex.org/C12186640","wikidata":"https://www.wikidata.org/wiki/Q6815743","display_name":"Memory model","level":3,"score":0.31119999289512634},{"id":"https://openalex.org/C133875982","wikidata":"https://www.wikidata.org/wiki/Q764810","display_name":"Shared memory","level":2,"score":0.3102000057697296},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.30550000071525574},{"id":"https://openalex.org/C111919701","wikidata":"https://www.wikidata.org/wiki/Q9135","display_name":"Operating system","level":1,"score":0.2930999994277954},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.28540000319480896},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.28369998931884766},{"id":"https://openalex.org/C854659","wikidata":"https://www.wikidata.org/wiki/Q1859284","display_name":"Message passing","level":2,"score":0.2768999934196472},{"id":"https://openalex.org/C2781172179","wikidata":"https://www.wikidata.org/wiki/Q853109","display_name":"Parallelism (grammar)","level":2,"score":0.2768000066280365},{"id":"https://openalex.org/C202491316","wikidata":"https://www.wikidata.org/wiki/Q272683","display_name":"Instruction set","level":2,"score":0.2743000090122223},{"id":"https://openalex.org/C58328972","wikidata":"https://www.wikidata.org/wiki/Q184609","display_name":"Expert system","level":2,"score":0.26600000262260437},{"id":"https://openalex.org/C176649486","wikidata":"https://www.wikidata.org/wiki/Q2308807","display_name":"Memory management","level":3,"score":0.25540000200271606}],"mesh":[],"locations_count":1,"locations":[{"id":"pmh:oai:arXiv.org:2506.23635","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2506.23635","pdf_url":"https://arxiv.org/pdf/2506.23635","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":"pmh:oai:arXiv.org:2506.23635","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2506.23635","pdf_url":"https://arxiv.org/pdf/2506.23635","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"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Large":[0],"Language":[1],"Models":[2],"(LLMs)":[3],"have":[4],"revolutionized":[5],"Artificial":[6],"Intelligence":[7],"(AI)":[8],"with":[9,50,69,166],"significant":[10,124],"advancements":[11],"such":[12],"as":[13,41,57],"OpenAI's":[14],"ChatGPT,":[15],"Meta's":[16],"Llama,":[17],"and":[18,26,63,184],"Databricks'":[19],"DBRX.":[20],"This":[21],"paper":[22],"addresses":[23],"the":[24,65,70,82,101,106,114,145,152,162,185],"cost":[25],"scalability":[27],"challenges":[28],"encountered":[29],"when":[30],"constructing":[31],"private":[32,192],"LLM":[33,193],"systems":[34],"for":[35,100,109,190],"personal":[36],"or":[37],"small":[38],"group":[39],"services,":[40],"aimed":[42],"by":[43],"Apple":[44,129],"Intelligence.":[45],"A":[46],"Mac":[47,153],"Studio":[48,154],"cluster":[49,155],"Apple's":[51],"M2":[52],"Ultra":[53],"chips":[54],"is":[55,103,156],"established":[56],"a":[58,150,174],"cost-efficient":[59,160],"solution":[60],"to":[61,87,105,128,143,177],"host":[62],"accelerate":[64],"pretrained":[66],"DBRX":[67],"model":[68,176,186],"Mixture-of-Experts":[71],"(MoE)":[72],"architecture.":[73],"Our":[74],"performance":[75,175,180],"analysis":[76],"reveal":[77],"that":[78,97],"parallel":[79],"execution":[80],"of":[81,116],"model's":[83],"experts":[84,102],"across":[85],"two":[86],"four":[88],"machine":[89],"nodes":[90],"significantly":[91],"reduces":[92],"inference":[93],"time.":[94],"We":[95,121],"find":[96],"computation":[98],"time":[99,108],"comparable":[104],"communication":[107],"exchanging":[110],"their":[111],"outputs,":[112],"emphasizing":[113],"importance":[115],"network":[117],"latency":[118],"over":[119],"bandwidth.":[120],"also":[122],"observe":[123],"management":[125,133,147],"overhead":[126],"due":[127],"software":[130],"stack's":[131],"memory":[132,146],"logic.":[134],"Based":[135],"on":[136],"these":[137],"findings,":[138],"we":[139,172],"develop":[140],"optimization":[141],"schemes":[142],"eliminate":[144],"overhead.":[148],"As":[149],"result,":[151],"1.15":[157],"times":[158],"more":[159],"than":[161],"state-of-the-art":[163],"AI":[164],"supercomputer":[165],"NVIDIA":[167],"H100":[168],"GPUs.":[169],"In":[170],"addition,":[171],"construct":[173],"estimate":[178],"system":[179],"under":[181],"varying":[182],"configurations,":[183],"provides":[187],"valuable":[188],"insights":[189],"designing":[191],"systems.":[194]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2025-10-10T00:00:00"}