{"id":"https://openalex.org/W7160960189","doi":"https://doi.org/10.48550/arxiv.2605.10501","title":"Accelerating Compound LLM Training Workloads with Maestro","display_name":"Accelerating Compound LLM Training Workloads with Maestro","publication_year":2026,"publication_date":"2026-05-11","ids":{"openalex":"https://openalex.org/W7160960189","doi":"https://doi.org/10.48550/arxiv.2605.10501"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.10501","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.10501","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,"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.2605.10501","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5064852098","display_name":"Xiulong Yuan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yuan, Xiulong","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135919884","display_name":"Hongqing Chen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Hongqing","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135972561","display_name":"Jiaxuan Peng","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Peng, Jiaxuan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135963605","display_name":"Fan Zhou","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhou, Fan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5034848916","display_name":"Zhixiang Ruan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ruan, Zhixiang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135964746","display_name":"Zekun Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Zekun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135970675","display_name":"Bo Zheng","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zheng, Bo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5004626105","display_name":"Rui Men","orcid":"https://orcid.org/0000-0002-4429-3461"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Men, Rui","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135925606","display_name":"Haiquan Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Haiquan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135909826","display_name":"Zhipeng Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Zhipeng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135929158","display_name":"Langshi Chen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Langshi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135924668","display_name":"Man Yuan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yuan, Man","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5050612924","display_name":"Jiaqi Gao","orcid":"https://orcid.org/0000-0003-3548-2030"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Gao, Jiaqi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101227501","display_name":"Zhengping Qian","orcid":"https://orcid.org/0000-0002-0741-9918"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Qian, Zhengping","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135952742","display_name":"Junyang Lin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lin, Junyang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135910385","display_name":"Yong Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Yong","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135912868","display_name":"Wei Lin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lin, Wei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135912295","display_name":"Junhua Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Junhua","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5135966855","display_name":"Jingren Zhou","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhou, Jingren","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/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.6643000245094299,"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":0.6643000245094299,"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":0.059300001710653305,"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":0.024299999698996544,"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/boosting","display_name":"Boosting (machine learning)","score":0.5971999764442444},{"id":"https://openalex.org/keywords/workload","display_name":"Workload","score":0.5906000137329102},{"id":"https://openalex.org/keywords/scheduling","display_name":"Scheduling (production processes)","score":0.5776000022888184},{"id":"https://openalex.org/keywords/throughput","display_name":"Throughput","score":0.436599999666214},{"id":"https://openalex.org/keywords/parallelism","display_name":"Parallelism (grammar)","score":0.4074000120162964},{"id":"https://openalex.org/keywords/multi-core-processor","display_name":"Multi-core processor","score":0.3971000015735626},{"id":"https://openalex.org/keywords/component","display_name":"Component (thermodynamics)","score":0.3953999876976013},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.36390000581741333},{"id":"https://openalex.org/keywords/resource-consumption","display_name":"Resource consumption","score":0.3431999981403351}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8783000111579895},{"id":"https://openalex.org/C46686674","wikidata":"https://www.wikidata.org/wiki/Q466303","display_name":"Boosting (machine learning)","level":2,"score":0.5971999764442444},{"id":"https://openalex.org/C2778476105","wikidata":"https://www.wikidata.org/wiki/Q628539","display_name":"Workload","level":2,"score":0.5906000137329102},{"id":"https://openalex.org/C206729178","wikidata":"https://www.wikidata.org/wiki/Q2271896","display_name":"Scheduling (production processes)","level":2,"score":0.5776000022888184},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.5516999959945679},{"id":"https://openalex.org/C157764524","wikidata":"https://www.wikidata.org/wiki/Q1383412","display_name":"Throughput","level":3,"score":0.436599999666214},{"id":"https://openalex.org/C2781172179","wikidata":"https://www.wikidata.org/wiki/Q853109","display_name":"Parallelism (grammar)","level":2,"score":0.4074000120162964},{"id":"https://openalex.org/C78766204","wikidata":"https://www.wikidata.org/wiki/Q555032","display_name":"Multi-core processor","level":2,"score":0.3971000015735626},{"id":"https://openalex.org/C168167062","wikidata":"https://www.wikidata.org/wiki/Q1117970","display_name":"Component (thermodynamics)","level":2,"score":0.3953999876976013},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.39340001344680786},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.36390000581741333},{"id":"https://openalex.org/C2777480716","wikidata":"https://www.wikidata.org/wiki/Q23582796","display_name":"Resource consumption","level":2,"score":0.3431999981403351},{"id":"https://openalex.org/C147764199","wikidata":"https://www.wikidata.org/wiki/Q6865248","display_name":"Minification","level":2,"score":0.3312999904155731},{"id":"https://openalex.org/C2989134064","wikidata":"https://www.wikidata.org/wiki/Q288510","display_name":"Execution time","level":2,"score":0.3116999864578247},{"id":"https://openalex.org/C113843644","wikidata":"https://www.wikidata.org/wiki/Q901882","display_name":"Interface (matter)","level":4,"score":0.3111000061035156},{"id":"https://openalex.org/C172430144","wikidata":"https://www.wikidata.org/wiki/Q17111997","display_name":"Symmetric multiprocessor system","level":2,"score":0.3091999888420105},{"id":"https://openalex.org/C61483411","wikidata":"https://www.wikidata.org/wiki/Q3124522","display_name":"Data parallelism","level":3,"score":0.30889999866485596},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.3046000003814697},{"id":"https://openalex.org/C149635348","wikidata":"https://www.wikidata.org/wiki/Q193040","display_name":"Embedded system","level":1,"score":0.30410000681877136},{"id":"https://openalex.org/C165696696","wikidata":"https://www.wikidata.org/wiki/Q11287","display_name":"Exploit","level":2,"score":0.29899999499320984},{"id":"https://openalex.org/C83283714","wikidata":"https://www.wikidata.org/wiki/Q121117","display_name":"Supercomputer","level":2,"score":0.29420000314712524},{"id":"https://openalex.org/C118524514","wikidata":"https://www.wikidata.org/wiki/Q173212","display_name":"Computer architecture","level":1,"score":0.29030001163482666},{"id":"https://openalex.org/C111919701","wikidata":"https://www.wikidata.org/wiki/Q9135","display_name":"Operating system","level":1,"score":0.2847000062465668},{"id":"https://openalex.org/C173801870","wikidata":"https://www.wikidata.org/wiki/Q201413","display_name":"Heuristic","level":2,"score":0.27480000257492065},{"id":"https://openalex.org/C2780870223","wikidata":"https://www.wikidata.org/wiki/Q1004415","display_name":"Runtime system","level":2,"score":0.25850000977516174},{"id":"https://openalex.org/C2778119891","wikidata":"https://www.wikidata.org/wiki/Q477690","display_name":"CUDA","level":2,"score":0.2574000060558319},{"id":"https://openalex.org/C169590947","wikidata":"https://www.wikidata.org/wiki/Q47506","display_name":"Compiler","level":2,"score":0.25189998745918274}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.10501","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.10501","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,"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.2605.10501","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.10501","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,"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":{"Compound":[0],"LLM":[1,9],"training":[2,77,101],"workloads-such":[3],"as":[4],"knowledge":[5,189],"distillation":[6,190],"and":[7,29,54,71,81,89,126,166,191],"multimodal":[8],"(MLLM)":[10],"training-are":[11],"gaining":[12],"prominence.":[13],"These":[14],"typically":[15],"comprise":[16],"heterogeneous":[17],"components":[18,80],"differing":[19],"in":[20,38,173],"parameter":[21],"scale,":[22],"execution":[23,156],"mode":[24],"(forward-only":[25],"or":[26],"full":[27],"forward-backward),":[28],"sequence":[30],"length.":[31],"Besides,":[32],"component":[33],"activation":[34],"can":[35],"be":[36],"data-dependent:":[37],"MLLM":[39,192],"training,":[40],"modality-specific":[41],"parts":[42],"activate":[43],"only":[44],"when":[45],"inputs":[46],"contain":[47],"corresponding":[48],"modalities,":[49],"causing":[50],"dynamic":[51,72],"computational":[52],"paths":[53],"irregular":[55],"runtime":[56,139],"workloads.":[57],"Conventional":[58],"frameworks,":[59],"designed":[60],"for":[61,175],"monolithic":[62],"models,":[63],"cannot":[64],"handle":[65],"the":[66,110],"dual":[67],"heterogeneity-static":[68],"(across":[69],"components)":[70],"(runtime).":[73],"By":[74],"enforcing":[75],"one-size-fits-all":[76],"configurations":[78],"across":[79],"ignoring":[82],"input-induced":[83],"variations,":[84],"they":[85],"suffer":[86],"suboptimal":[87],"throughput":[88],"poor":[90],"GPU":[91,178,182],"utilization.":[92,171],"In":[93],"this":[94],"paper,":[95],"we":[96],"introduce":[97],"Maestro,":[98],"a":[99,113,143],"section-centric":[100],"framework":[102],"that":[103,147],"addresses":[104],"both":[105],"challenges.":[106],"Maestro":[107,141,180],"first":[108],"restructures":[109],"workload":[111],"into":[112],"coarse-grained":[114],"section":[115,118,155],"graph.":[116],"Each":[117],"independently":[119],"configures":[120],"its":[121,194],"parallelism":[122,165],"strategy,":[123],"micro-batch":[124],"size,":[125],"data-parallel":[127],"degree-enabling":[128],"fine-grained,":[129],"component-aware":[130],"resource":[131],"allocation":[132],"to":[133,152],"tackle":[134,138],"static":[135],"heterogeneity.":[136],"To":[137],"irregularity,":[140],"introduces":[142],"wavefront":[144],"scheduling":[145],"algorithm":[146],"dynamically":[148],"reorders":[149],"input":[150],"samples":[151],"orchestrate":[153],"concurrent":[154],"while":[157],"preserving":[158],"cross-section":[159],"data":[160],"dependencies.":[161],"This":[162],"maximizes":[163],"inter-section":[164],"minimizes":[167],"stalls,":[168],"boosting":[169],"hardware":[170],"Deployed":[172],"production":[174],"millions":[176],"of":[177],"hours,":[179],"reduces":[181],"consumption":[183],"by":[184],"~40%":[185],"on":[186],"key":[187],"workloads-including":[188],"training-validating":[193],"real-world":[195],"impact.":[196]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-05-13T00:00:00"}