{"id":"https://openalex.org/W4417145508","doi":"https://doi.org/10.48550/arxiv.2503.11367","title":"Efficient Distributed MLLM Training with Cornstarch","display_name":"Efficient Distributed MLLM Training with Cornstarch","publication_year":2025,"publication_date":"2025-03-14","ids":{"openalex":"https://openalex.org/W4417145508","doi":"https://doi.org/10.48550/arxiv.2503.11367"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2503.11367","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2503.11367","pdf_url":"https://arxiv.org/pdf/2503.11367","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,"listed_in":[],"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"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2503.11367","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5064139539","display_name":"Insu Jang","orcid":"https://orcid.org/0009-0007-5206-2333","observed_orcids":["https://orcid.org/0009-0007-5206-2333"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jang, Insu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5120728213","display_name":"Runyu Lu","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lu, Runyu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5008689083","display_name":"Nikhil Bansal","orcid":"https://orcid.org/0000-0002-6290-0894","observed_orcids":["https://orcid.org/0000-0002-6290-0894"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Bansal, Nikhil","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100683541","display_name":"Ang Chen","orcid":"https://orcid.org/0000-0002-1352-7384","observed_orcids":["https://orcid.org/0000-0002-1352-7384"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Ang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5013180923","display_name":"Mosharaf Chowdhury","orcid":"https://orcid.org/0000-0003-0884-6740","observed_orcids":["https://orcid.org/0000-0003-0884-6740"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chowdhury, Mosharaf","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":true,"cited_by_count":8,"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/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.6244999766349792,"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":0.6244999766349792,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.11789999902248383,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.1177000030875206,"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/multimodal-large-language-models","display_name":"multimodal large language models","score":0.9980000257492065},{"id":"https://openalex.org/keywords/distributed-training","display_name":"distributed training","score":0.9620000123977661},{"id":"https://openalex.org/keywords/pipeline-parallelism","display_name":"pipeline parallelism","score":0.8830000162124634},{"id":"https://openalex.org/keywords/training-throughput","display_name":"training throughput","score":0.7739999890327454},{"id":"https://openalex.org/keywords/model-heterogeneity","display_name":"model heterogeneity","score":0.6690000295639038},{"id":"https://openalex.org/keywords/data-parallelism","display_name":"data parallelism","score":0.5920000076293945},{"id":"https://openalex.org/keywords/training-framework","display_name":"training framework","score":0.699999988079071}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8355000019073486},{"id":"https://openalex.org/C43521106","wikidata":"https://www.wikidata.org/wiki/Q2165493","display_name":"Pipeline (software)","level":2,"score":0.6967999935150146},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.6169000267982483},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.5885000228881836},{"id":"https://openalex.org/C2781172179","wikidata":"https://www.wikidata.org/wiki/Q853109","display_name":"Parallelism (grammar)","level":2,"score":0.5042999982833862},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.4560999870300293},{"id":"https://openalex.org/C2779903281","wikidata":"https://www.wikidata.org/wiki/Q6888026","display_name":"Modalities","level":2,"score":0.4408999979496002},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.43540000915527344},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.4117000102996826},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.38659998774528503},{"id":"https://openalex.org/C124304363","wikidata":"https://www.wikidata.org/wiki/Q673661","display_name":"Abstraction","level":2,"score":0.38429999351501465},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3686000108718872},{"id":"https://openalex.org/C67186912","wikidata":"https://www.wikidata.org/wiki/Q367664","display_name":"Data modeling","level":2,"score":0.3431999981403351},{"id":"https://openalex.org/C61483411","wikidata":"https://www.wikidata.org/wiki/Q3124522","display_name":"Data parallelism","level":3,"score":0.337799996137619},{"id":"https://openalex.org/C183322885","wikidata":"https://www.wikidata.org/wiki/Q17007702","display_name":"Context model","level":3,"score":0.31610000133514404},{"id":"https://openalex.org/C162319229","wikidata":"https://www.wikidata.org/wiki/Q175263","display_name":"Data structure","level":2,"score":0.2962999939918518},{"id":"https://openalex.org/C77618280","wikidata":"https://www.wikidata.org/wiki/Q1155772","display_name":"Scheme (mathematics)","level":2,"score":0.2962000072002411},{"id":"https://openalex.org/C115903868","wikidata":"https://www.wikidata.org/wiki/Q80993","display_name":"Software engineering","level":1,"score":0.2849999964237213},{"id":"https://openalex.org/C95203288","wikidata":"https://www.wikidata.org/wiki/Q221682","display_name":"Semaphore","level":2,"score":0.28439998626708984},{"id":"https://openalex.org/C195324797","wikidata":"https://www.wikidata.org/wiki/Q33742","display_name":"Natural language","level":2,"score":0.27799999713897705},{"id":"https://openalex.org/C2779227376","wikidata":"https://www.wikidata.org/wiki/Q6505497","display_name":"Layer (electronics)","level":2,"score":0.275299996137619},{"id":"https://openalex.org/C138958017","wikidata":"https://www.wikidata.org/wiki/Q190087","display_name":"Data type","level":2,"score":0.2567000091075897},{"id":"https://openalex.org/C2778562939","wikidata":"https://www.wikidata.org/wiki/Q1298791","display_name":"Synchronization (alternating current)","level":3,"score":0.2556999921798706},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.2556000053882599}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2503.11367","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2503.11367","pdf_url":"https://arxiv.org/pdf/2503.11367","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,"listed_in":[],"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"},{"id":"doi:10.48550/arxiv.2503.11367","doi":"https://doi.org/10.48550/arxiv.2503.11367","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2503.11367","pdf_url":null,"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,"listed_in":[],"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"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2503.11367","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2503.11367","pdf_url":"https://arxiv.org/pdf/2503.11367","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,"listed_in":[],"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":[],"study_designs":[],"awards":[{"id":"https://openalex.org/G1171448932","display_name":"Collaborative Research: CNS Core: Large: Runtime Programmable Networks","funder_award_id":"2535540","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"},{"id":"https://openalex.org/G1614532030","display_name":"Collaborative Research: CNS Core: Medium: Movement of Computation and Data in Splitkernel-disaggregated, Data-intensive Systems","funder_award_id":"2406598","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"},{"id":"https://openalex.org/G2607445930","display_name":"Collaborative Research: AF: Medium: Fundamental Challenges in Discrete and Continuous Optimization","funder_award_id":"2504995","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"},{"id":"https://openalex.org/G3695048862","display_name":"Collaborative Research: AF: Small: New Directions and Approaches in Discrepancy Theory","funder_award_id":"2327011","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"},{"id":"https://openalex.org/G3700094867","display_name":"Collaborative Research: CNS Core: Medium: Systems Support for Federated Learning","funder_award_id":"2106184","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"},{"id":"https://openalex.org/G4315580007","display_name":"Collaborative Research: DESC: Type II: Enhancing Data Center Efficiency through Intelligent Tiered Memory Management","funder_award_id":"2450085","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"}],"funders":[{"id":"https://openalex.org/F4320306076","display_name":"National Science Foundation","ror":"https://ror.org/021nxhr62"},{"id":"https://openalex.org/F4320307791","display_name":"Cisco Systems","ror":"https://ror.org/03yt1ez60"},{"id":"https://openalex.org/F4320327705","display_name":"Mozilla Foundation","ror":"https://ror.org/01y8r3379"},{"id":"https://openalex.org/F4320337253","display_name":"Instituto de Ciencias del Mar y Limnolog\u00eda, Universidad Nacional Aut\u00f3noma de M\u00e9xico","ror":"https://ror.org/040wdye98"}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4417145508.pdf","grobid_xml":"https://content.openalex.org/works/W4417145508.grobid-xml"},"referenced_works_count":22,"referenced_works":["https://openalex.org/W4310282800","https://openalex.org/W4361194507","https://openalex.org/W4386768656","https://openalex.org/W4387561528","https://openalex.org/W4390723197","https://openalex.org/W4392131573","https://openalex.org/W4393284539","https://openalex.org/W4395117922","https://openalex.org/W4395474395","https://openalex.org/W4399452094","https://openalex.org/W4400105114","https://openalex.org/W4400717319","https://openalex.org/W4401306886","https://openalex.org/W4401660192","https://openalex.org/W4402711432","https://openalex.org/W4403345778","https://openalex.org/W4403853618","https://openalex.org/W4404784276","https://openalex.org/W4407197044","https://openalex.org/W4415796658","https://openalex.org/W6947991961","https://openalex.org/W7134986434"],"related_works":["https://openalex.org/W4416257518","https://openalex.org/W4410454663","https://openalex.org/W4403965219","https://openalex.org/W4416800422","https://openalex.org/W4410356774","https://openalex.org/W7161597045","https://openalex.org/W4410363086","https://openalex.org/W4417237984","https://openalex.org/W7162456501","https://openalex.org/W6891969918"],"abstract_inverted_index":{"Multimodal":[0],"large":[1,9],"language":[2,10],"models":[3,11],"(MLLMs)":[4],"extend":[5],"the":[6,61,74],"capabilities":[7],"of":[8,76,131],"(LLMs)":[12],"by":[13,125],"combining":[14],"heterogeneous":[15],"model":[16,32,97],"architectures":[17],"to":[18,40,59,70,111],"handle":[19],"diverse":[20],"modalities":[21],"like":[22],"images":[23],"and":[24,34,98,106],"audio.":[25],"However,":[26],"this":[27,79],"inherent":[28],"heterogeneity":[29,62],"in":[30,63,95,129],"MLLM":[31,48,64,87,113,132],"structure":[33],"data":[35,99],"types":[36],"makes":[37],"makeshift":[38],"extensions":[39],"existing":[41],"LLM":[42],"training":[43,88,114,133],"frameworks":[44],"unsuitable":[45],"for":[46],"efficient":[47,85],"training.":[49],"While":[50],"there":[51],"are":[52,68],"a":[53],"few":[54],"works":[55],"that":[56,90,120],"have":[57],"attempted":[58],"address":[60],"training,":[65],"their":[66],"approaches":[67],"limited":[69],"only":[71],"superficially":[72],"considering":[73],"characteristics":[75,94],"MLLMs.":[77],"In":[78],"paper,":[80],"we":[81],"present":[82],"Cornstarch,":[83],"an":[84,137],"distributed":[86],"framework":[89],"contemplates":[91],"MLLM's":[92],"unique":[93],"both":[96],"parallelization.":[100],"Cornstarch":[101,121,135],"introduces":[102],"frozen-aware":[103],"pipeline":[104],"parallelism":[105,110],"token":[107],"workload-balanced":[108],"context":[109],"improve":[112],"throughput.":[115,134],"Our":[116],"extensive":[117],"evaluation":[118],"shows":[119],"outperforms":[122],"state-of-the-art":[123],"solutions":[124],"$2.26\\times$":[126],"on":[127],"average":[128],"terms":[130],"is":[136],"open-source":[138],"project":[139],"available":[140],"at":[141],"https://github.com/cornstarch-org/Cornstarch.":[142]},"counts_by_year":[{"year":2026,"cited_by_count":8}],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2025-10-10T00:00:00"}