{"id":"https://openalex.org/W7160906116","doi":"https://doi.org/10.48550/arxiv.2605.08962","title":"MegaScale-Omni: A Hyper-Scale, Workload-Resilient System for MultiModal LLM Training in Production","display_name":"MegaScale-Omni: A Hyper-Scale, Workload-Resilient System for MultiModal LLM Training in Production","publication_year":2026,"publication_date":"2026-05-09","ids":{"openalex":"https://openalex.org/W7160906116","doi":"https://doi.org/10.48550/arxiv.2605.08962"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.08962","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.08962","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","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.2605.08962","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5111875369","display_name":"Chunyu Xue","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xue, Chunyu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135933259","display_name":"Yangrui Chen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Yangrui","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5091837703","display_name":"Jianyu Jiang","orcid":"https://orcid.org/0000-0002-8684-8509"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jiang, Jianyu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5018118687","display_name":"Ningxin Zheng","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zheng, Ningxin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5017329016","display_name":"Junda Feng","orcid":"https://orcid.org/0000-0003-3664-6615"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Feng, Junda","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135978044","display_name":"Jingji Chen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Jingji","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135997215","display_name":"Shixiong Zhao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhao, Shixiong","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135972394","display_name":"Shen Yan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yan, Shen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135978967","display_name":"Yi Lin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lin, Yi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135921380","display_name":"Lei Shi","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shi, Lei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5053078653","display_name":"Zanbo Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Zanbo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5102431101","display_name":"Lishu Luo","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Luo, Lishu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135917975","display_name":"Faming Wu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wu, Faming","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135984883","display_name":"Haibin Lin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lin, Haibin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135921993","display_name":"Xin Liu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Xin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135926752","display_name":"Yanghua Peng","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Peng, Yanghua","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5135979122","display_name":"Quan Chen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Quan","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/T10181","display_name":"Natural Language Processing Techniques","score":0.23479999601840973,"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"}},"topics":[{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.23479999601840973,"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/T10028","display_name":"Topic Modeling","score":0.15459999442100525,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.09049999713897705,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/workload","display_name":"Workload","score":0.7305999994277954},{"id":"https://openalex.org/keywords/pipeline","display_name":"Pipeline (software)","score":0.6801000237464905},{"id":"https://openalex.org/keywords/encoder","display_name":"Encoder","score":0.6054999828338623},{"id":"https://openalex.org/keywords/process","display_name":"Process (computing)","score":0.5250999927520752},{"id":"https://openalex.org/keywords/throughput","display_name":"Throughput","score":0.5210999846458435},{"id":"https://openalex.org/keywords/parallelism","display_name":"Parallelism (grammar)","score":0.4867999851703644},{"id":"https://openalex.org/keywords/training","display_name":"Training (meteorology)","score":0.4472000002861023},{"id":"https://openalex.org/keywords/exploit","display_name":"Exploit","score":0.4449000060558319},{"id":"https://openalex.org/keywords/component","display_name":"Component (thermodynamics)","score":0.4194999933242798}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8172000050544739},{"id":"https://openalex.org/C2778476105","wikidata":"https://www.wikidata.org/wiki/Q628539","display_name":"Workload","level":2,"score":0.7305999994277954},{"id":"https://openalex.org/C43521106","wikidata":"https://www.wikidata.org/wiki/Q2165493","display_name":"Pipeline (software)","level":2,"score":0.6801000237464905},{"id":"https://openalex.org/C118505674","wikidata":"https://www.wikidata.org/wiki/Q42586063","display_name":"Encoder","level":2,"score":0.6054999828338623},{"id":"https://openalex.org/C98045186","wikidata":"https://www.wikidata.org/wiki/Q205663","display_name":"Process (computing)","level":2,"score":0.5250999927520752},{"id":"https://openalex.org/C157764524","wikidata":"https://www.wikidata.org/wiki/Q1383412","display_name":"Throughput","level":3,"score":0.5210999846458435},{"id":"https://openalex.org/C2781172179","wikidata":"https://www.wikidata.org/wiki/Q853109","display_name":"Parallelism (grammar)","level":2,"score":0.4867999851703644},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.4472000002861023},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.4458000063896179},{"id":"https://openalex.org/C165696696","wikidata":"https://www.wikidata.org/wiki/Q11287","display_name":"Exploit","level":2,"score":0.4449000060558319},{"id":"https://openalex.org/C168167062","wikidata":"https://www.wikidata.org/wiki/Q1117970","display_name":"Component (thermodynamics)","level":2,"score":0.4194999933242798},{"id":"https://openalex.org/C77618280","wikidata":"https://www.wikidata.org/wiki/Q1155772","display_name":"Scheme (mathematics)","level":2,"score":0.41040000319480896},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3750999867916107},{"id":"https://openalex.org/C29202148","wikidata":"https://www.wikidata.org/wiki/Q287260","display_name":"Resource allocation","level":2,"score":0.35749998688697815},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.35089999437332153},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.32989999651908875},{"id":"https://openalex.org/C206345919","wikidata":"https://www.wikidata.org/wiki/Q20380951","display_name":"Resource (disambiguation)","level":2,"score":0.3294999897480011},{"id":"https://openalex.org/C2780226545","wikidata":"https://www.wikidata.org/wiki/Q6888030","display_name":"Modality (human\u2013computer interaction)","level":2,"score":0.325300008058548},{"id":"https://openalex.org/C78766204","wikidata":"https://www.wikidata.org/wiki/Q555032","display_name":"Multi-core processor","level":2,"score":0.32170000672340393},{"id":"https://openalex.org/C79403827","wikidata":"https://www.wikidata.org/wiki/Q3988","display_name":"Real-time computing","level":1,"score":0.31779998540878296},{"id":"https://openalex.org/C61483411","wikidata":"https://www.wikidata.org/wiki/Q3124522","display_name":"Data parallelism","level":3,"score":0.3172000050544739},{"id":"https://openalex.org/C118524514","wikidata":"https://www.wikidata.org/wiki/Q173212","display_name":"Computer architecture","level":1,"score":0.31610000133514404},{"id":"https://openalex.org/C2778112365","wikidata":"https://www.wikidata.org/wiki/Q3511065","display_name":"Sequence (biology)","level":2,"score":0.2976999878883362},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.2953000068664551},{"id":"https://openalex.org/C19275194","wikidata":"https://www.wikidata.org/wiki/Q222903","display_name":"Multiplexing","level":2,"score":0.2935999929904938},{"id":"https://openalex.org/C106515295","wikidata":"https://www.wikidata.org/wiki/Q26806595","display_name":"Parallel processing","level":2,"score":0.29159998893737793},{"id":"https://openalex.org/C2778915421","wikidata":"https://www.wikidata.org/wiki/Q3643177","display_name":"Performance improvement","level":2,"score":0.29109999537467957},{"id":"https://openalex.org/C18555067","wikidata":"https://www.wikidata.org/wiki/Q8375051","display_name":"Joint (building)","level":2,"score":0.2851000130176544},{"id":"https://openalex.org/C126042441","wikidata":"https://www.wikidata.org/wiki/Q1324888","display_name":"Frame (networking)","level":2,"score":0.28380000591278076},{"id":"https://openalex.org/C2776857766","wikidata":"https://www.wikidata.org/wiki/Q7832987","display_name":"Training system","level":2,"score":0.2800000011920929},{"id":"https://openalex.org/C67186912","wikidata":"https://www.wikidata.org/wiki/Q367664","display_name":"Data modeling","level":2,"score":0.27459999918937683},{"id":"https://openalex.org/C2780609101","wikidata":"https://www.wikidata.org/wiki/Q17156588","display_name":"Resource management (computing)","level":2,"score":0.2720000147819519},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.25679999589920044},{"id":"https://openalex.org/C2778348673","wikidata":"https://www.wikidata.org/wiki/Q739302","display_name":"Production (economics)","level":2,"score":0.25519999861717224}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.08962","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.08962","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2605.08962","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.08962","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[{"id":"https://metadata.un.org/sdg/8","score":0.4822755753993988,"display_name":"Decent work and economic growth"}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"As":[0],"the":[1,51,75,104,156],"foundational":[2],"component":[3],"of":[4,42,78,126,158,167],"versatile":[5],"AI":[6],"applications,":[7],"training":[8,61,76,163],"an":[9,58],"multimodal":[10,17],"large":[11],"language":[12],"model":[13,46],"(MLLM)":[14],"relies":[15],"on":[16],"datasets":[18],"with":[19,81,89,130,165],"dynamic":[20,35,65,178],"modality":[21],"mixture":[22],"proportions":[23],"and":[24,45,50,68,99,122,144],"sample":[25],"length":[26],"distributions.":[27],"However,":[28],"existing":[29],"MLLM":[30,60,162],"systems":[31],"remain":[32],"inefficient":[33],"under":[34,109,176],"workloads,":[36,179],"due":[37],"to":[38,95,149,182],"statically":[39],"coupled":[40],"decisions":[41],"resource":[43],"allocation":[44],"parallelization":[47,112],"between":[48],"encoders":[49,94],"LLM":[52,105,150],"backbone.":[53],"This":[54],"paper":[55],"presents":[56],"MegaScale-Omni,":[57],"industrial-grade":[59],"system":[62],"tailored":[63],"for":[64,93,103,118],"workload":[66,131],"adaption":[67],"hyper-scale":[69],"deployment.":[70],"MegaScale-Omni":[71,152],"is":[72,153],"built":[73],"upon":[74],"scheme":[77],"encoder-LLM":[79,116,127],"multiplexing":[80],"three":[82],"key":[83],"innovations:":[84],"(1)":[85],"Decoupled":[86],"parallelism":[87,92,102],"strategies":[88],"long-short":[90],"sequence":[91],"process":[96],"variable-length":[97],"samples,":[98],"full-fledged":[100],"5D":[101],"backbone,":[106],"both":[107],"organized":[108],"a":[110,123],"communication-efficient":[111],"layout.":[113],"(2)":[114],"Unified":[115],"representations":[117],"flexible,":[119],"extensible":[120],"colocation,":[121],"new":[124],"paradigm":[125],"joint":[128],"pipeline":[129],"resilience.":[132],"(3)":[133],"Workload":[134],"balancing":[135],"techniques":[136],"via":[137],"decentralized":[138],"grouped":[139],"reordering":[140],"in":[141],"data":[142],"loaders":[143],"adaptive":[145],"resharding":[146],"from":[147],"encoder":[148],"ranks.":[151],"deployed":[154],"as":[155,180],"foundation":[157],"our":[159],"in-house":[160],"large-scale":[161],"tasks":[164],"thousands":[166],"GPUs.":[168],"Our":[169],"experimental":[170],"results":[171],"demonstrate":[172],"$1.27\\times$-$7.57\\times$":[173],"throughput":[174],"improvement":[175],"production-grade":[177],"compared":[181],"four":[183],"state-of-the-art":[184],"systems.":[185]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-05-13T00:00:00"}