{"id":"https://openalex.org/W4389518760","doi":"https://doi.org/10.18653/v1/2023.emnlp-main.298","title":"GQA: Training Generalized Multi-Query Transformer Models from Multi-Head Checkpoints","display_name":"GQA: Training Generalized Multi-Query Transformer Models from Multi-Head Checkpoints","publication_year":2023,"publication_date":"2023-01-01","ids":{"openalex":"https://openalex.org/W4389518760","doi":"https://doi.org/10.18653/v1/2023.emnlp-main.298"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2023.emnlp-main.298","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2023.emnlp-main.298","pdf_url":"https://aclanthology.org/2023.emnlp-main.298.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2023.emnlp-main.298.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5072605113","display_name":"Joshua Ainslie","orcid":null},"institutions":[{"id":"https://openalex.org/I1291425158","display_name":"Google (United States)","ror":"https://ror.org/00njsd438","country_code":"US","type":"company","lineage":["https://openalex.org/I1291425158","https://openalex.org/I4210128969"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Joshua Ainslie","raw_affiliation_strings":["Google Research"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Google Research","institution_ids":["https://openalex.org/I1291425158"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5018339854","display_name":"James Lee-Thorp","orcid":"https://orcid.org/0000-0001-6445-7155"},"institutions":[{"id":"https://openalex.org/I1291425158","display_name":"Google (United States)","ror":"https://ror.org/00njsd438","country_code":"US","type":"company","lineage":["https://openalex.org/I1291425158","https://openalex.org/I4210128969"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"James Lee-Thorp","raw_affiliation_strings":["Google Research"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Google Research","institution_ids":["https://openalex.org/I1291425158"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5021943393","display_name":"Michiel de Jong","orcid":"https://orcid.org/0000-0002-2153-3362"},"institutions":[{"id":"https://openalex.org/I1291425158","display_name":"Google (United States)","ror":"https://ror.org/00njsd438","country_code":"US","type":"company","lineage":["https://openalex.org/I1291425158","https://openalex.org/I4210128969"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Michiel de Jong","raw_affiliation_strings":["University of Southern California. Work done at Google Research"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Southern California. Work done at Google Research","institution_ids":["https://openalex.org/I1291425158"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5036319063","display_name":"Yury Zemlyanskiy","orcid":null},"institutions":[{"id":"https://openalex.org/I1291425158","display_name":"Google (United States)","ror":"https://ror.org/00njsd438","country_code":"US","type":"company","lineage":["https://openalex.org/I1291425158","https://openalex.org/I4210128969"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Yury Zemlyanskiy","raw_affiliation_strings":["University of Southern California. Work done at Google Research"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Southern California. Work done at Google Research","institution_ids":["https://openalex.org/I1291425158"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5092005408","display_name":"Federico Lebr\u00f3n","orcid":null},"institutions":[{"id":"https://openalex.org/I1291425158","display_name":"Google (United States)","ror":"https://ror.org/00njsd438","country_code":"US","type":"company","lineage":["https://openalex.org/I1291425158","https://openalex.org/I4210128969"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Federico Lebron","raw_affiliation_strings":["Google Research"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Google Research","institution_ids":["https://openalex.org/I1291425158"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5012884492","display_name":"Sumit Sanghai","orcid":null},"institutions":[{"id":"https://openalex.org/I1291425158","display_name":"Google (United States)","ror":"https://ror.org/00njsd438","country_code":"US","type":"company","lineage":["https://openalex.org/I1291425158","https://openalex.org/I4210128969"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Sumit Sanghai","raw_affiliation_strings":["Google Research"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Google Research","institution_ids":["https://openalex.org/I1291425158"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I1291425158"],"apc_list":null,"apc_paid":null,"fwci":77.4203,"has_fulltext":true,"cited_by_count":371,"citation_normalized_percentile":{"value":0.99966525,"is_in_top_1_percent":true,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":99,"max":100},"biblio":{"volume":null,"issue":null,"first_page":"4895","last_page":"4901"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.9984999895095825,"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":0.9984999895095825,"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":0.9962000250816345,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.9876000285148621,"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":0.8215606212615967},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.7266016006469727},{"id":"https://openalex.org/keywords/generalization","display_name":"Generalization","score":0.4506067931652069},{"id":"https://openalex.org/keywords/query-optimization","display_name":"Query optimization","score":0.4322908818721771},{"id":"https://openalex.org/keywords/query-expansion","display_name":"Query expansion","score":0.42559635639190674},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.3847692608833313},{"id":"https://openalex.org/keywords/data-mining","display_name":"Data mining","score":0.321644127368927},{"id":"https://openalex.org/keywords/information-retrieval","display_name":"Information retrieval","score":0.13659173250198364},{"id":"https://openalex.org/keywords/mathematics","display_name":"Mathematics","score":0.08126333355903625}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8215606212615967},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.7266016006469727},{"id":"https://openalex.org/C177148314","wikidata":"https://www.wikidata.org/wiki/Q170084","display_name":"Generalization","level":2,"score":0.4506067931652069},{"id":"https://openalex.org/C157692150","wikidata":"https://www.wikidata.org/wiki/Q2919848","display_name":"Query optimization","level":2,"score":0.4322908818721771},{"id":"https://openalex.org/C99016210","wikidata":"https://www.wikidata.org/wiki/Q5488129","display_name":"Query expansion","level":2,"score":0.42559635639190674},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3847692608833313},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.321644127368927},{"id":"https://openalex.org/C23123220","wikidata":"https://www.wikidata.org/wiki/Q816826","display_name":"Information retrieval","level":1,"score":0.13659173250198364},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.08126333355903625},{"id":"https://openalex.org/C134306372","wikidata":"https://www.wikidata.org/wiki/Q7754","display_name":"Mathematical analysis","level":1,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2023.emnlp-main.298","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2023.emnlp-main.298","pdf_url":"https://aclanthology.org/2023.emnlp-main.298.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2023.emnlp-main.298","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2023.emnlp-main.298","pdf_url":"https://aclanthology.org/2023.emnlp-main.298.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"},"sustainable_development_goals":[{"id":"https://metadata.un.org/sdg/4","display_name":"Quality Education","score":0.41999998688697815}],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4389518760.pdf","grobid_xml":"https://content.openalex.org/works/W4389518760.grobid-xml"},"referenced_works_count":21,"referenced_works":["https://openalex.org/W1821462560","https://openalex.org/W2002555321","https://openalex.org/W2923014074","https://openalex.org/W2963204221","https://openalex.org/W2963339397","https://openalex.org/W2963926728","https://openalex.org/W2963929190","https://openalex.org/W3138154797","https://openalex.org/W3169942382","https://openalex.org/W4224308101","https://openalex.org/W4281758439","https://openalex.org/W4288028629","https://openalex.org/W4288089799","https://openalex.org/W4292119927","https://openalex.org/W4307934016","https://openalex.org/W4308760184","https://openalex.org/W4310561894","https://openalex.org/W4311252752","https://openalex.org/W4319166707","https://openalex.org/W4322718191","https://openalex.org/W4385570209"],"related_works":["https://openalex.org/W2096359267","https://openalex.org/W2006459955","https://openalex.org/W2146885082","https://openalex.org/W2572349046","https://openalex.org/W4386051213","https://openalex.org/W185198413","https://openalex.org/W3125756434","https://openalex.org/W3049728138","https://openalex.org/W2348136644","https://openalex.org/W1936839791"],"abstract_inverted_index":{"Multi-query":[0],"attention":[1,64,70,98],"(MQA),":[2],"which":[3,71],"only":[4],"uses":[5,72],"a":[6,31,41,66],"single":[7],"key-value":[8,86],"head,":[9],"drastically":[10],"speeds":[11],"up":[12],"decoder":[13],"inference.":[14,37],"However,":[15],"MQA":[16,53],"can":[17],"lead":[18],"to":[19,29,96,102],"quality":[20,94],"degradation,":[21],"and":[22,60],"moreover":[23],"it":[24],"may":[25],"not":[26],"be":[27],"desirable":[28],"train":[30],"separate":[32],"model":[33,48],"just":[34],"for":[35,43],"faster":[36],"We":[38,88],"(1)":[39],"propose":[40],"recipe":[42],"uptraining":[44],"existing":[45],"multi-head":[46,97],"language":[47],"checkpoints":[49],"into":[50],"models":[51],"with":[52,99],"using":[54],"5%":[55],"of":[56,68,81,85],"original":[57],"pre-training":[58],"compute,":[59],"(2)":[61],"introduce":[62],"grouped-query":[63],"(GQA),":[65],"generalization":[67],"multi-query":[69],"an":[73],"intermediate":[74],"(more":[75],"than":[76,79],"one,":[77],"less":[78],"number":[80,84],"query":[82],"heads)":[83],"heads.":[87],"show":[89],"that":[90],"uptrained":[91],"GQA":[92],"achieves":[93],"close":[95],"comparable":[100],"speed":[101],"MQA.":[103]},"counts_by_year":[{"year":2026,"cited_by_count":109},{"year":2025,"cited_by_count":187},{"year":2024,"cited_by_count":67},{"year":2023,"cited_by_count":8}],"updated_date":"2026-08-24T07:32:12.397491","created_date":"2025-10-10T00:00:00"}