{"id":"https://openalex.org/W7160645524","doi":"https://doi.org/10.48550/arxiv.2605.06374","title":"ResiHP: Taming LLM Training Failures with Dynamic Hybrid Parallelism","display_name":"ResiHP: Taming LLM Training Failures with Dynamic Hybrid Parallelism","publication_year":2026,"publication_date":"2026-05-07","ids":{"openalex":"https://openalex.org/W7160645524","doi":"https://doi.org/10.48550/arxiv.2605.06374"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.06374","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.06374","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.2605.06374","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5135667586","display_name":"Tenghui Ma","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ma, Tenghui","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5102966178","display_name":"Jihu Guo","orcid":"https://orcid.org/0009-0002-9811-0234"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Guo, Jihu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135655395","display_name":"Wei Gao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Gao, Wei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135651315","display_name":"Sitian Lu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lu, Sitian","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135638704","display_name":"Zhisheng Ye","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ye, Zhisheng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128831402","display_name":"Hanjing Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Hanjing","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5135641264","display_name":"Dahua Lin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lin, Dahua","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":0.2953999936580658,"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":0.2953999936580658,"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/T10101","display_name":"Cloud Computing and Resource Management","score":0.15919999778270721,"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"}},{"id":"https://openalex.org/T10772","display_name":"Distributed systems and fault tolerance","score":0.13330000638961792,"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/workload","display_name":"Workload","score":0.6938999891281128},{"id":"https://openalex.org/keywords/skew","display_name":"Skew","score":0.6590999960899353},{"id":"https://openalex.org/keywords/spurious-relationship","display_name":"Spurious relationship","score":0.6017000079154968},{"id":"https://openalex.org/keywords/training","display_name":"Training (meteorology)","score":0.5382000207901001},{"id":"https://openalex.org/keywords/scheduling","display_name":"Scheduling (production processes)","score":0.5376999974250793},{"id":"https://openalex.org/keywords/overhead","display_name":"Overhead (engineering)","score":0.5152000188827515},{"id":"https://openalex.org/keywords/throughput","display_name":"Throughput","score":0.4052000045776367},{"id":"https://openalex.org/keywords/adaptation","display_name":"Adaptation (eye)","score":0.38940000534057617},{"id":"https://openalex.org/keywords/parallelism","display_name":"Parallelism (grammar)","score":0.3626999855041504}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7275999784469604},{"id":"https://openalex.org/C2778476105","wikidata":"https://www.wikidata.org/wiki/Q628539","display_name":"Workload","level":2,"score":0.6938999891281128},{"id":"https://openalex.org/C43711488","wikidata":"https://www.wikidata.org/wiki/Q7534783","display_name":"Skew","level":2,"score":0.6590999960899353},{"id":"https://openalex.org/C97256817","wikidata":"https://www.wikidata.org/wiki/Q1462316","display_name":"Spurious relationship","level":2,"score":0.6017000079154968},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.5382000207901001},{"id":"https://openalex.org/C206729178","wikidata":"https://www.wikidata.org/wiki/Q2271896","display_name":"Scheduling (production processes)","level":2,"score":0.5376999974250793},{"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/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.45179998874664307},{"id":"https://openalex.org/C79403827","wikidata":"https://www.wikidata.org/wiki/Q3988","display_name":"Real-time computing","level":1,"score":0.42669999599456787},{"id":"https://openalex.org/C157764524","wikidata":"https://www.wikidata.org/wiki/Q1383412","display_name":"Throughput","level":3,"score":0.4052000045776367},{"id":"https://openalex.org/C139807058","wikidata":"https://www.wikidata.org/wiki/Q352374","display_name":"Adaptation (eye)","level":2,"score":0.38940000534057617},{"id":"https://openalex.org/C2781172179","wikidata":"https://www.wikidata.org/wiki/Q853109","display_name":"Parallelism (grammar)","level":2,"score":0.3626999855041504},{"id":"https://openalex.org/C165696696","wikidata":"https://www.wikidata.org/wiki/Q11287","display_name":"Exploit","level":2,"score":0.36000001430511475},{"id":"https://openalex.org/C19012869","wikidata":"https://www.wikidata.org/wiki/Q578372","display_name":"Response time","level":2,"score":0.311599999666214},{"id":"https://openalex.org/C50897621","wikidata":"https://www.wikidata.org/wiki/Q2665508","display_name":"Hybrid system","level":2,"score":0.2921999990940094},{"id":"https://openalex.org/C63540848","wikidata":"https://www.wikidata.org/wiki/Q3140932","display_name":"Fault tolerance","level":2,"score":0.2759000062942505},{"id":"https://openalex.org/C193702766","wikidata":"https://www.wikidata.org/wiki/Q1414548","display_name":"Concurrency","level":2,"score":0.2743000090122223},{"id":"https://openalex.org/C200601418","wikidata":"https://www.wikidata.org/wiki/Q2193887","display_name":"Reliability engineering","level":1,"score":0.2711000144481659},{"id":"https://openalex.org/C149635348","wikidata":"https://www.wikidata.org/wiki/Q193040","display_name":"Embedded system","level":1,"score":0.2646999955177307},{"id":"https://openalex.org/C2778112365","wikidata":"https://www.wikidata.org/wiki/Q3511065","display_name":"Sequence (biology)","level":2,"score":0.2628999948501587},{"id":"https://openalex.org/C26713055","wikidata":"https://www.wikidata.org/wiki/Q245962","display_name":"Implementation","level":2,"score":0.26190000772476196},{"id":"https://openalex.org/C176553487","wikidata":"https://www.wikidata.org/wiki/Q7855819","display_name":"Turnaround time","level":2,"score":0.26190000772476196},{"id":"https://openalex.org/C2776857766","wikidata":"https://www.wikidata.org/wiki/Q7832987","display_name":"Training system","level":2,"score":0.260699987411499},{"id":"https://openalex.org/C46686674","wikidata":"https://www.wikidata.org/wiki/Q466303","display_name":"Boosting (machine learning)","level":2,"score":0.2565000057220459},{"id":"https://openalex.org/C94915269","wikidata":"https://www.wikidata.org/wiki/Q1834857","display_name":"Detector","level":2,"score":0.2551000118255615},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.2540000081062317},{"id":"https://openalex.org/C77019957","wikidata":"https://www.wikidata.org/wiki/Q2689057","display_name":"Dependability","level":2,"score":0.2533000111579895}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.06374","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.06374","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.2605.06374","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.06374","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":{"Hybrid":[0],"parallelism":[1,143],"underpins":[2],"large-scale":[3],"LLM":[4],"training":[5,28,154,163,171],"across":[6,24],"tens":[7],"of":[8,10],"thousands":[9],"GPUs.":[11],"At":[12],"such":[13],"scale,":[14],"hardware":[15],"failures":[16,65,124],"on":[17],"individual":[18,69],"devices":[19],"lead":[20],"to":[21,75,109,152],"performance":[22,41],"skew":[23,42],"devices,":[25],"diminishing":[26],"overall":[27],"efficiency.":[29],"Existing":[30],"resilient":[31,81,90,170],"systems":[32,172],"overlook":[33],"sequence":[34,55],"length":[35,56],"variability":[36,57],"in":[37,71,177],"datasets":[38],"and":[39,63,79,97,148],"device":[40],"under":[43,156,173],"hybrid":[44,72,101],"parallelism.":[45],"As":[46],"a":[47,89,107,117,138,178],"result,":[48],"(1)":[49],"iteration":[50,126],"time":[51,120,127],"fluctuations":[52,128],"induced":[53],"by":[54,165],"can":[58],"trigger":[59],"spurious":[60],"fail-slow":[61],"detections,":[62],"(2)":[64],"are":[66],"mitigated":[67],"through":[68],"adaptations":[70],"parallelism,":[73],"leading":[74],"unnecessary":[76],"detection":[77,96],"overhead":[78],"inefficient":[80],"training.":[82,103],"To":[83],"respond,":[84],"this":[85],"paper":[86],"presents":[87],"ResiHP,":[88],"system":[91],"that":[92,122,140,160],"enables":[93],"robust":[94],"failure":[95,175],"fine-grained":[98],"adaptation":[99],"for":[100,132],"parallel":[102],"First,":[104],"we":[105,136],"develop":[106],"Detector":[108],"accurately":[110],"identify":[111],"failures.":[112,157],"In":[113],"particular,":[114],"it":[115],"employs":[116],"workload-aware":[118],"execution":[119],"predictor":[121],"disentangles":[123],"from":[125],"while":[129],"remaining":[130],"lightweight":[131],"online":[133],"detection.":[134],"Second,":[135],"design":[137],"Scheduler":[139],"dynamically":[141],"adapts":[142],"group":[144],"sizes,":[145],"model":[146],"partitioning,":[147],"workload":[149],"scheduling":[150],"policies":[151],"improve":[153],"efficiency":[155],"Experiments":[158],"show":[159],"ResiHP":[161],"improves":[162],"throughput":[164],"1.04-4.39$\\times$":[166],"compared":[167],"with":[168],"state-of-the-art":[169],"diverse":[174],"scenarios":[176],"256-GPU":[179],"cluster.":[180]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-05-09T00:00:00"}