{"id":"https://openalex.org/W4386768656","doi":"https://doi.org/10.14778/3611540.3611569","title":"PyTorch FSDP: Experiences on Scaling Fully Sharded Data Parallel","display_name":"PyTorch FSDP: Experiences on Scaling Fully Sharded Data Parallel","publication_year":2023,"publication_date":"2023-08-01","ids":{"openalex":"https://openalex.org/W4386768656","doi":"https://doi.org/10.14778/3611540.3611569"},"language":"en","primary_location":{"id":"doi:10.14778/3611540.3611569","is_oa":false,"landing_page_url":"https://doi.org/10.14778/3611540.3611569","pdf_url":null,"source":{"id":"https://openalex.org/S4210226185","display_name":"Proceedings of the VLDB Endowment","issn_l":"2150-8097","issn":["2150-8097"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":"https://openalex.org/P4310319798","host_organization_name":"Association for Computing Machinery","host_organization_lineage":["https://openalex.org/P4310319798"],"host_organization_lineage_names":["Association for Computing Machinery"],"type":"journal"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the VLDB Endowment","raw_type":"journal-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":false,"oa_status":"closed","oa_url":null,"any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5112754141","display_name":"Yanli Zhao","orcid":"https://orcid.org/0000-0002-9231-8360"},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Yanli Zhao","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5058043863","display_name":"Andrew Gu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Andrew Gu","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5014393716","display_name":"Rohan Varma","orcid":"https://orcid.org/0000-0002-4635-7334"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Rohan Varma","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101672911","display_name":"Liang Luo","orcid":"https://orcid.org/0009-0009-4403-5604"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liang Luo","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5109370237","display_name":"Chien-Chin Huang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chien-Chin Huang","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100413849","display_name":"Min Xu","orcid":"https://orcid.org/0000-0001-9581-8849"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Min Xu","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5035010141","display_name":"Less Wright","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Less Wright","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5049675042","display_name":"Hamid Shojanazeri","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hamid Shojanazeri","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5076248976","display_name":"Myle Ott","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Myle Ott","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5042449490","display_name":"Sam Shleifer","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sam Shleifer","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5027893749","display_name":"Alban Desmaison","orcid":"https://orcid.org/0009-0002-4359-1974"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Alban Desmaison","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5042895564","display_name":"Can Balioglu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Can Balioglu","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5003360849","display_name":"Pritam Damania","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Pritam Damania","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5068083482","display_name":"Bernard Nguyen","orcid":"https://orcid.org/0000-0003-4203-6805"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Bernard Nguyen","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5074075244","display_name":"Geeta Chauhan","orcid":"https://orcid.org/0009-0003-0830-7330"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Geeta Chauhan","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5102015406","display_name":"Yuchen Hao","orcid":"https://orcid.org/0009-0009-3794-7074"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yuchen Hao","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5045208071","display_name":"Ajit Mathews","orcid":"https://orcid.org/0009-0003-4199-0434"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ajit Mathews","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5104910007","display_name":"Shen Li","orcid":"https://orcid.org/0000-0003-2947-3787"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shen Li","raw_affiliation_strings":["Meta AI"],"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":[]}]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":18,"corresponding_author_ids":["https://openalex.org/A5112754141"],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":17.2996,"has_fulltext":false,"cited_by_count":145,"citation_normalized_percentile":{"value":0.99541955,"is_in_top_1_percent":true,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":99,"max":100},"biblio":{"volume":"16","issue":"12","first_page":"3848","last_page":"3860"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.9994999766349792,"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":0.9994999766349792,"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/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.9987999796867371,"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/T11181","display_name":"Advanced Data Storage Technologies","score":0.9932000041007996,"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/computer-science","display_name":"Computer science","score":0.80790776014328},{"id":"https://openalex.org/keywords/leverage","display_name":"Leverage (statistics)","score":0.73039710521698},{"id":"https://openalex.org/keywords/scalability","display_name":"Scalability","score":0.6026870012283325},{"id":"https://openalex.org/keywords/scaling","display_name":"Scaling","score":0.4774513840675354},{"id":"https://openalex.org/keywords/field","display_name":"Field (mathematics)","score":0.44738060235977173},{"id":"https://openalex.org/keywords/parallel-computing","display_name":"Parallel computing","score":0.39126238226890564},{"id":"https://openalex.org/keywords/distributed-computing","display_name":"Distributed computing","score":0.3393532633781433},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.3336564898490906},{"id":"https://openalex.org/keywords/operating-system","display_name":"Operating system","score":0.11390182375907898}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.80790776014328},{"id":"https://openalex.org/C153083717","wikidata":"https://www.wikidata.org/wiki/Q6535263","display_name":"Leverage (statistics)","level":2,"score":0.73039710521698},{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.6026870012283325},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.4774513840675354},{"id":"https://openalex.org/C9652623","wikidata":"https://www.wikidata.org/wiki/Q190109","display_name":"Field (mathematics)","level":2,"score":0.44738060235977173},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.39126238226890564},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.3393532633781433},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3336564898490906},{"id":"https://openalex.org/C111919701","wikidata":"https://www.wikidata.org/wiki/Q9135","display_name":"Operating system","level":1,"score":0.11390182375907898},{"id":"https://openalex.org/C202444582","wikidata":"https://www.wikidata.org/wiki/Q837863","display_name":"Pure mathematics","level":1,"score":0.0},{"id":"https://openalex.org/C2524010","wikidata":"https://www.wikidata.org/wiki/Q8087","display_name":"Geometry","level":1,"score":0.0},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.14778/3611540.3611569","is_oa":false,"landing_page_url":"https://doi.org/10.14778/3611540.3611569","pdf_url":null,"source":{"id":"https://openalex.org/S4210226185","display_name":"Proceedings of the VLDB Endowment","issn_l":"2150-8097","issn":["2150-8097"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":"https://openalex.org/P4310319798","host_organization_name":"Association for Computing Machinery","host_organization_lineage":["https://openalex.org/P4310319798"],"host_organization_lineage_names":["Association for Computing Machinery"],"type":"journal"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the VLDB Endowment","raw_type":"journal-article"}],"best_oa_location":null,"sustainable_development_goals":[{"score":0.5699999928474426,"id":"https://metadata.un.org/sdg/9","display_name":"Industry, innovation and infrastructure"}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":12,"referenced_works":["https://openalex.org/W2604553456","https://openalex.org/W2763421725","https://openalex.org/W2969388332","https://openalex.org/W3037847693","https://openalex.org/W3038098022","https://openalex.org/W3129831491","https://openalex.org/W3189259198","https://openalex.org/W3204998121","https://openalex.org/W3206636350","https://openalex.org/W4226343359","https://openalex.org/W4226479682","https://openalex.org/W4287756266"],"related_works":["https://openalex.org/W2302028273","https://openalex.org/W1525643724","https://openalex.org/W2067938758","https://openalex.org/W2364921833","https://openalex.org/W2385146268","https://openalex.org/W1569389315","https://openalex.org/W2503642292","https://openalex.org/W2380023786","https://openalex.org/W2546696010","https://openalex.org/W1992741870"],"abstract_inverted_index":{"It":[0],"is":[1,148],"widely":[2],"acknowledged":[3],"that":[4,146],"large":[5,41,89],"models":[6,164],"have":[7],"the":[8,21,26,36,64],"potential":[9],"to":[10,47,67,113,132,154],"deliver":[11],"superior":[12],"performance":[13,153],"across":[14,136],"a":[15,48,126,137],"broad":[16],"range":[17,127],"of":[18,28,40,51,128,139,150,170],"domains.":[19],"Despite":[20],"remarkable":[22],"progress":[23],"made":[24],"in":[25,58,168],"field":[27],"machine":[29],"learning":[30],"systems":[31],"research,":[32],"which":[33],"has":[34,93],"enabled":[35],"development":[37],"and":[38,54,69,108,118,130],"exploration":[39],"models,":[42],"such":[43],"abilities":[44],"remain":[45],"confined":[46],"small":[49],"group":[50],"advanced":[52],"users":[53],"industry":[55],"leaders,":[56],"resulting":[57],"an":[59,85],"implicit":[60],"technical":[61],"barrier":[62],"for":[63,88,161],"wider":[65],"community":[66],"access":[68],"leverage":[70],"these":[71],"technologies.":[72],"In":[73],"this":[74],"paper,":[75],"we":[76],"introduce":[77],"PyTorch":[78,100],"Fully":[79],"Sharded":[80],"Data":[81,156],"Parallel":[82,157],"(FSDP)":[83],"as":[84],"industry-grade":[86],"solution":[87],"model":[90],"training.":[91],"FSDP":[92,123,147],"been":[94],"closely":[95],"co-designed":[96],"with":[97,165],"several":[98],"key":[99],"core":[101],"components":[102],"including":[103],"Tensor":[104],"implementation,":[105],"dispatcher":[106],"system,":[107],"CUDA":[109],"memory":[110],"caching":[111],"allocator,":[112],"provide":[114],"non-intrusive":[115],"user":[116],"experiences":[117],"high":[119],"training":[120],"efficiency.":[121],"Additionally,":[122],"natively":[124],"incorporates":[125],"techniques":[129],"settings":[131],"optimize":[133],"resource":[134],"utilization":[135],"variety":[138],"hardware":[140],"configurations.":[141],"The":[142],"experimental":[143],"results":[144],"demonstrate":[145],"capable":[149],"achieving":[151],"comparable":[152],"Distributed":[155],"while":[158],"providing":[159],"support":[160],"significantly":[162],"larger":[163],"near-linear":[166],"scalability":[167],"terms":[169],"TFLOPS.":[171]},"counts_by_year":[{"year":2026,"cited_by_count":13},{"year":2025,"cited_by_count":83},{"year":2024,"cited_by_count":37},{"year":2023,"cited_by_count":12}],"updated_date":"2026-04-11T08:14:18.477133","created_date":"2025-10-10T00:00:00"}