{"id":"https://openalex.org/W7162334674","doi":"https://doi.org/10.48550/arxiv.2605.23764","title":"HyperParallel-MoE: Multi-Core Interleaved Scheduling for Fast MoE Training on Ascend NPUs","display_name":"HyperParallel-MoE: Multi-Core Interleaved Scheduling for Fast MoE Training on Ascend NPUs","publication_year":2026,"publication_date":"2026-05-22","ids":{"openalex":"https://openalex.org/W7162334674","doi":"https://doi.org/10.48550/arxiv.2605.23764"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.23764","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.23764","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":false,"raw_source_name":null,"raw_type":"Preprint"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2605.23764","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5102982554","display_name":"Zewen Jin","orcid":"https://orcid.org/0000-0002-4762-6559"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jin, Zewen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136933345","display_name":"Congkun Ai","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ai, Congkun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136931625","display_name":"Guangpeng Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Guangpeng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136971495","display_name":"Hanbo Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Hanbo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136941745","display_name":"Haoran Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Haoran","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5040041960","display_name":"Shihan Xiao","orcid":"https://orcid.org/0000-0001-9702-9130"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xiao, Shihan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136948560","display_name":"Da Lei","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lei, Da","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136905903","display_name":"Xuefeng Jin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jin, Xuefeng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136931308","display_name":"Teng Su","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Su, Teng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5136901850","display_name":"Cheng Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Cheng","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/T10036","display_name":"Advanced Neural Network Applications","score":0.16300000250339508,"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":0.16300000250339508,"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/T11612","display_name":"Stochastic Gradient Optimization Techniques","score":0.15440000593662262,"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"}},{"id":"https://openalex.org/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.07000000029802322,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/scheduling","display_name":"Scheduling (production processes)","score":0.6474000215530396},{"id":"https://openalex.org/keywords/computation","display_name":"Computation","score":0.5785999894142151},{"id":"https://openalex.org/keywords/symmetric-multiprocessor-system","display_name":"Symmetric multiprocessor system","score":0.46129998564720154},{"id":"https://openalex.org/keywords/latency","display_name":"Latency (audio)","score":0.40380001068115234},{"id":"https://openalex.org/keywords/synchronization","display_name":"Synchronization (alternating current)","score":0.3905999958515167},{"id":"https://openalex.org/keywords/runtime-system","display_name":"Runtime system","score":0.3864000141620636},{"id":"https://openalex.org/keywords/task","display_name":"Task (project management)","score":0.35760000348091125}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8605999946594238},{"id":"https://openalex.org/C206729178","wikidata":"https://www.wikidata.org/wiki/Q2271896","display_name":"Scheduling (production processes)","level":2,"score":0.6474000215530396},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.5785999894142151},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.5157999992370605},{"id":"https://openalex.org/C172430144","wikidata":"https://www.wikidata.org/wiki/Q17111997","display_name":"Symmetric multiprocessor system","level":2,"score":0.46129998564720154},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.4122999906539917},{"id":"https://openalex.org/C82876162","wikidata":"https://www.wikidata.org/wiki/Q17096504","display_name":"Latency (audio)","level":2,"score":0.40380001068115234},{"id":"https://openalex.org/C2778562939","wikidata":"https://www.wikidata.org/wiki/Q1298791","display_name":"Synchronization (alternating current)","level":3,"score":0.3905999958515167},{"id":"https://openalex.org/C2780870223","wikidata":"https://www.wikidata.org/wiki/Q1004415","display_name":"Runtime system","level":2,"score":0.3864000141620636},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.35760000348091125},{"id":"https://openalex.org/C74193536","wikidata":"https://www.wikidata.org/wiki/Q574844","display_name":"Kernel (algebra)","level":2,"score":0.32820001244544983},{"id":"https://openalex.org/C17349429","wikidata":"https://www.wikidata.org/wiki/Q1049914","display_name":"Matrix multiplication","level":3,"score":0.3160000145435333},{"id":"https://openalex.org/C2777655698","wikidata":"https://www.wikidata.org/wiki/Q1209358","display_name":"Versa","level":2,"score":0.2913999855518341},{"id":"https://openalex.org/C175154964","wikidata":"https://www.wikidata.org/wiki/Q380077","display_name":"Task analysis","level":3,"score":0.2623000144958496},{"id":"https://openalex.org/C111919701","wikidata":"https://www.wikidata.org/wiki/Q9135","display_name":"Operating system","level":1,"score":0.26010000705718994},{"id":"https://openalex.org/C206952183","wikidata":"https://www.wikidata.org/wiki/Q1193100","display_name":"Preemption","level":2,"score":0.2542000114917755},{"id":"https://openalex.org/C96147967","wikidata":"https://www.wikidata.org/wiki/Q190686","display_name":"Subroutine","level":2,"score":0.25360000133514404},{"id":"https://openalex.org/C201410400","wikidata":"https://www.wikidata.org/wiki/Q1064412","display_name":"Multithreading","level":3,"score":0.2531000077724457}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.23764","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.23764","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":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2605.23764","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.23764","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":false,"raw_source_name":null,"raw_type":"Preprint"},"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":{"Modern":[0],"Mixture-of-Experts":[1],"(MoE)":[2],"models":[3,172],"increasingly":[4],"rely":[5],"on":[6,64,173,201],"large-scale":[7],"AI":[8],"accelerator":[9],"clusters":[10],"for":[11,61],"efficient":[12],"training.":[13],"Ascend":[14,65,174],"NPUs":[15],"expose":[16],"heterogeneous":[17,49,77,193],"on-chip":[18],"compute":[19],"resources,":[20],"including":[21],"matrix-oriented":[22],"AIC":[23,80,133],"units":[24,28],"and":[25,58,81,101,108,134,149,163,166],"vector-oriented":[26],"AIV":[27,82,135],"with":[29,116],"explicit":[30],"cross-queue":[31,114],"synchronization":[32],"support.":[33],"However,":[34],"existing":[35,154],"training":[36,63,199],"frameworks":[37],"largely":[38],"execute":[39],"MoE":[40,62,70,171,198],"operators":[41],"in":[42,160],"a":[43,56,73,104,127,138],"serialized":[44],"kernel-by-kernel":[45],"manner,":[46],"leaving":[47],"substantial":[48],"parallelism":[50],"underutilized.":[51],"This":[52],"paper":[53],"presents":[54],"HyperParallel-MoE,":[55],"compilation":[57],"scheduling":[59,111,194],"framework":[60],"NPUs.":[66,203],"HyperParallel-MoE":[67,120,159,181],"transforms":[68],"operator-level":[69],"execution":[71,115],"into":[72],"statically":[74],"scheduled":[75],"tile-level":[76,192],"taskflow":[78,125],"spanning":[79],"resources.":[83],"It":[84],"introduces":[85],"AIV-driven":[86],"one-sided":[87],"communication":[88,100],"to":[89,98,112,188],"eliminate":[90],"host-side":[91],"collective":[92],"synchronization,":[93],"dependency-preserving":[94],"tile":[95],"task":[96,106],"generation":[97],"unify":[99],"computation":[102,151],"under":[103],"common":[105],"abstraction,":[107],"event-driven":[109],"static":[110],"coordinate":[113],"low":[117],"runtime":[118,129],"overhead.":[119],"further":[121],"executes":[122],"the":[123,161],"compiled":[124],"within":[126],"unified":[128],"that":[130,191],"concurrently":[131],"drives":[132],"workers":[136],"inside":[137],"single":[139],"kernel":[140],"launch,":[141],"enabling":[142],"fine-grained":[143],"overlap":[144],"among":[145],"communication,":[146],"matrix":[147],"computation,":[148],"vector":[150],"while":[152],"preserving":[153],"optimized":[155],"operators.":[156],"We":[157],"implement":[158],"MindSpore":[162],"MindFormers":[164],"stack":[165],"evaluate":[167],"it":[168],"using":[169],"DeepSeek-style":[170],"A3":[175],"clusters.":[176],"Across":[177],"multiple":[178],"expert-parallel":[179],"configurations,":[180],"reduces":[182],"Dispatch-to-Combine":[183],"MoE-FFN":[184],"latency":[185],"by":[186],"up":[187],"1.58x,":[189],"demonstrating":[190],"can":[195],"substantially":[196],"improve":[197],"efficiency":[200],"modern":[202],"The":[204],"source":[205],"code":[206],"is":[207],"available":[208],"at":[209],"https://gitcode.com/mindspore/hyper-parallel/tree/master/hyper_parallel/core/multicore":[210]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-05-26T00:00:00"}