{"id":"https://openalex.org/W4390489725","doi":"https://doi.org/10.48550/arxiv.2312.17482","title":"MosaicBERT: A Bidirectional Encoder Optimized for Fast Pretraining","display_name":"MosaicBERT: A Bidirectional Encoder Optimized for Fast Pretraining","publication_year":2023,"publication_date":"2023-12-29","ids":{"openalex":"https://openalex.org/W4390489725","doi":"https://doi.org/10.48550/arxiv.2312.17482"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2312.17482","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2312.17482","pdf_url":"https://arxiv.org/pdf/2312.17482","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"","raw_type":"text"},"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2312.17482","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5051310825","display_name":"Jacob Portes","orcid":"https://orcid.org/0000-0003-3102-012X"},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Portes, Jacob","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5046071180","display_name":"Alex Trott","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Trott, Alex","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5109658192","display_name":"Sam Havens","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Havens, Sam","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5067437521","display_name":"Daniel King","orcid":"https://orcid.org/0000-0002-6678-0586"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"King, Daniel","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5045587960","display_name":"Abhinav Venigalla","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Venigalla, Abhinav","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5000852572","display_name":"Moin Nadeem","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Nadeem, Moin","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5048231039","display_name":"Nikhil Sardana","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sardana, Nikhil","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5060661843","display_name":"Daya Shanker Khudia","orcid":"https://orcid.org/0000-0002-6706-5074"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Khudia, Daya","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5078716102","display_name":"Jonathan Frankle","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Frankle, Jonathan","raw_affiliation_strings":[],"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":9,"corresponding_author_ids":["https://openalex.org/A5051310825"],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":4,"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/T10028","display_name":"Topic Modeling","score":0.9977999925613403,"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/T10028","display_name":"Topic Modeling","score":0.9977999925613403,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.9976999759674072,"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/T12535","display_name":"Machine Learning and Data Classification","score":0.97079998254776,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.8288034200668335},{"id":"https://openalex.org/keywords/encoder","display_name":"Encoder","score":0.703739583492279},{"id":"https://openalex.org/keywords/transformer","display_name":"Transformer","score":0.6248695850372314},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.5717306137084961},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.4336514472961426},{"id":"https://openalex.org/keywords/vocabulary","display_name":"Vocabulary","score":0.4252700209617615},{"id":"https://openalex.org/keywords/resolver","display_name":"Resolver","score":0.4158935546875},{"id":"https://openalex.org/keywords/speech-recognition","display_name":"Speech recognition","score":0.4076923727989197},{"id":"https://openalex.org/keywords/natural-language-processing","display_name":"Natural language processing","score":0.36762386560440063},{"id":"https://openalex.org/keywords/computer-hardware","display_name":"Computer hardware","score":0.3300965428352356}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8288034200668335},{"id":"https://openalex.org/C118505674","wikidata":"https://www.wikidata.org/wiki/Q42586063","display_name":"Encoder","level":2,"score":0.703739583492279},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.6248695850372314},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.5717306137084961},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4336514472961426},{"id":"https://openalex.org/C2777601683","wikidata":"https://www.wikidata.org/wiki/Q6499736","display_name":"Vocabulary","level":2,"score":0.4252700209617615},{"id":"https://openalex.org/C80156102","wikidata":"https://www.wikidata.org/wiki/Q788036","display_name":"Resolver","level":3,"score":0.4158935546875},{"id":"https://openalex.org/C28490314","wikidata":"https://www.wikidata.org/wiki/Q189436","display_name":"Speech recognition","level":1,"score":0.4076923727989197},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.36762386560440063},{"id":"https://openalex.org/C9390403","wikidata":"https://www.wikidata.org/wiki/Q3966","display_name":"Computer hardware","level":1,"score":0.3300965428352356},{"id":"https://openalex.org/C41895202","wikidata":"https://www.wikidata.org/wiki/Q8162","display_name":"Linguistics","level":1,"score":0.0},{"id":"https://openalex.org/C76155785","wikidata":"https://www.wikidata.org/wiki/Q418","display_name":"Telecommunications","level":1,"score":0.0},{"id":"https://openalex.org/C138885662","wikidata":"https://www.wikidata.org/wiki/Q5891","display_name":"Philosophy","level":0,"score":0.0},{"id":"https://openalex.org/C62520636","wikidata":"https://www.wikidata.org/wiki/Q944","display_name":"Quantum mechanics","level":1,"score":0.0},{"id":"https://openalex.org/C111919701","wikidata":"https://www.wikidata.org/wiki/Q9135","display_name":"Operating system","level":1,"score":0.0},{"id":"https://openalex.org/C165801399","wikidata":"https://www.wikidata.org/wiki/Q25428","display_name":"Voltage","level":2,"score":0.0},{"id":"https://openalex.org/C165005293","wikidata":"https://www.wikidata.org/wiki/Q1074500","display_name":"Chip","level":2,"score":0.0},{"id":"https://openalex.org/C121332964","wikidata":"https://www.wikidata.org/wiki/Q413","display_name":"Physics","level":0,"score":0.0}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2312.17482","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2312.17482","pdf_url":"https://arxiv.org/pdf/2312.17482","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"","raw_type":"text"},{"id":"doi:10.48550/arxiv.2312.17482","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2312.17482","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":"article-journal"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2312.17482","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2312.17482","pdf_url":"https://arxiv.org/pdf/2312.17482","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"","raw_type":"text"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":false},"content_urls":{"pdf":"https://content.openalex.org/works/W4390489725.pdf"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W2964964914","https://openalex.org/W4288365749","https://openalex.org/W2936497627","https://openalex.org/W3013624417","https://openalex.org/W4287826556","https://openalex.org/W4287598411","https://openalex.org/W3098382480","https://openalex.org/W3094871513","https://openalex.org/W3100913109","https://openalex.org/W3198458223"],"abstract_inverted_index":{"Although":[0],"BERT-style":[1,63,218],"encoder":[2,64,105,139],"models":[3,219],"are":[4,192],"heavily":[5],"used":[6],"in":[7,131,161,209],"NLP":[8],"research,":[9],"many":[10,37],"researchers":[11,212],"do":[12],"not":[13],"pretrain":[14,216],"their":[15],"own":[16],"BERTs":[17],"from":[18,135,143],"scratch":[19,144],"due":[20],"to":[21,35,52,92,133,198,215],"the":[22,28,102,116,146],"high":[23],"cost":[24,172,222],"of":[25,159,173,224],"training.":[26],"In":[27],"past":[29],"half-decade":[30],"since":[31],"BERT":[32,201],"first":[33],"rose":[34],"prominence,":[36],"advances":[38],"have":[39,50],"been":[40],"made":[41],"with":[42,82],"other":[43,138],"transformer":[44,104],"architectures":[45],"and":[46,66,97,124,137,185,190,203,213,236],"training":[47,67,108],"configurations":[48],"that":[49,69,187],"yet":[51],"be":[53],"systematically":[54],"incorporated":[55],"into":[56,101],"BERT.":[57],"Here,":[58],"we":[59],"introduce":[60],"MosaicBERT,":[61],"a":[62,90,111,153,171,199],"architecture":[65,78],"recipe":[68,109],"is":[70],"empirically":[71],"optimized":[72,127],"for":[73,115,128],"fast":[74],"pretraining.":[75],"This":[76,205],"efficient":[77],"incorporates":[79],"FlashAttention,":[80],"Attention":[81],"Linear":[83,87],"Biases":[84],"(ALiBi),":[85],"Gated":[86],"Units":[88],"(GLU),":[89],"module":[91],"dynamically":[93],"remove":[94],"padded":[95],"tokens,":[96],"low":[98,221],"precision":[99],"LayerNorm":[100],"classic":[103],"block.":[106],"The":[107],"includes":[110],"30%":[112],"masking":[113],"ratio":[114],"Masked":[117],"Language":[118],"Modeling":[119],"(MLM)":[120],"objective,":[121],"bfloat16":[122],"precision,":[123],"vocabulary":[125],"size":[126],"GPU":[129],"throughput,":[130],"addition":[132],"best-practices":[134],"RoBERTa":[136],"models.":[140,229],"When":[141],"pretrained":[142],"on":[145,164,226],"C4":[147],"dataset,":[148],"this":[149],"base":[150,189,202],"model":[151,234],"achieves":[152],"downstream":[154],"average":[155],"GLUE":[156],"(dev)":[157],"score":[158],"79.6":[160],"1.13":[162],"hours":[163],"8":[165],"A100":[166],"80":[167],"GB":[168],"GPUs":[169],"at":[170,220],"roughly":[174],"$20.":[175],"We":[176,230],"plot":[177],"extensive":[178],"accuracy":[179],"vs.":[180],"pretraining":[181,210],"speed":[182,207],"Pareto":[183,194],"curves":[184],"show":[186],"MosaicBERT":[188],"large":[191],"consistently":[193],"optimal":[195],"when":[196],"compared":[197],"competitive":[200],"large.":[204],"empirical":[206],"up":[208],"enables":[211],"engineers":[214],"custom":[217],"instead":[223],"finetune":[225],"existing":[227],"generic":[228],"open":[231],"source":[232],"our":[233],"weights":[235],"code.":[237]},"counts_by_year":[{"year":2025,"cited_by_count":1},{"year":2024,"cited_by_count":2},{"year":2023,"cited_by_count":1}],"updated_date":"2026-03-10T16:38:18.471706","created_date":"2025-10-10T00:00:00"}