{"id":"https://openalex.org/W4399115160","doi":"https://doi.org/10.48550/arxiv.2405.17428","title":"NV-Embed: Improved Techniques for Training LLMs as Generalist Embedding Models","display_name":"NV-Embed: Improved Techniques for Training LLMs as Generalist Embedding Models","publication_year":2024,"publication_date":"2024-05-27","ids":{"openalex":"https://openalex.org/W4399115160","doi":"https://doi.org/10.48550/arxiv.2405.17428"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2405.17428","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2405.17428","pdf_url":"https://arxiv.org/pdf/2405.17428","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":null,"raw_type":"text"},"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2405.17428","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5089891838","display_name":"Chankyu Lee","orcid":"https://orcid.org/0000-0002-6933-7459"},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Lee, Chankyu","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5054331314","display_name":"Rajarshi Roy","orcid":"https://orcid.org/0000-0001-9474-0031"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Roy, Rajarshi","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5112528146","display_name":"Mengyao Xu","orcid":"https://orcid.org/0009-0000-5772-1739"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xu, Mengyao","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5058844317","display_name":"Jonathan Raiman","orcid":"https://orcid.org/0009-0000-5229-7318"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Raiman, Jonathan","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5072436307","display_name":"Mohammad Shoeybi","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shoeybi, Mohammad","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5066242985","display_name":"Bryan Catanzaro","orcid":"https://orcid.org/0000-0003-0034-7728"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Catanzaro, Bryan","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5101947241","display_name":"Ping Wei","orcid":"https://orcid.org/0000-0002-8535-9527"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ping, Wei","raw_affiliation_strings":[],"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":7,"corresponding_author_ids":["https://openalex.org/A5089891838"],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":10,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.8385000228881836,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.8385000228881836,"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/T13643","display_name":"Artificial Intelligence in Law","score":0.7390000224113464,"subfield":{"id":"https://openalex.org/subfields/3320","display_name":"Political Science and International Relations"},"field":{"id":"https://openalex.org/fields/33","display_name":"Social Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},{"id":"https://openalex.org/T11652","display_name":"Imbalanced Data Classification Techniques","score":0.6995999813079834,"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/generalist-and-specialist-species","display_name":"Generalist and specialist species","score":0.7702559232711792},{"id":"https://openalex.org/keywords/training","display_name":"Training (meteorology)","score":0.7214134931564331},{"id":"https://openalex.org/keywords/embedding","display_name":"Embedding","score":0.7023699879646301},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.4985501766204834},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.3372907042503357},{"id":"https://openalex.org/keywords/geography","display_name":"Geography","score":0.13996249437332153},{"id":"https://openalex.org/keywords/ecology","display_name":"Ecology","score":0.09870076179504395},{"id":"https://openalex.org/keywords/biology","display_name":"Biology","score":0.0934033989906311}],"concepts":[{"id":"https://openalex.org/C45371612","wikidata":"https://www.wikidata.org/wiki/Q3058587","display_name":"Generalist and specialist species","level":3,"score":0.7702559232711792},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.7214134931564331},{"id":"https://openalex.org/C41608201","wikidata":"https://www.wikidata.org/wiki/Q980509","display_name":"Embedding","level":2,"score":0.7023699879646301},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.4985501766204834},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3372907042503357},{"id":"https://openalex.org/C205649164","wikidata":"https://www.wikidata.org/wiki/Q1071","display_name":"Geography","level":0,"score":0.13996249437332153},{"id":"https://openalex.org/C18903297","wikidata":"https://www.wikidata.org/wiki/Q7150","display_name":"Ecology","level":1,"score":0.09870076179504395},{"id":"https://openalex.org/C86803240","wikidata":"https://www.wikidata.org/wiki/Q420","display_name":"Biology","level":0,"score":0.0934033989906311},{"id":"https://openalex.org/C185933670","wikidata":"https://www.wikidata.org/wiki/Q52105","display_name":"Habitat","level":2,"score":0.0},{"id":"https://openalex.org/C153294291","wikidata":"https://www.wikidata.org/wiki/Q25261","display_name":"Meteorology","level":1,"score":0.0}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2405.17428","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2405.17428","pdf_url":"https://arxiv.org/pdf/2405.17428","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":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2405.17428","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2405.17428","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"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2405.17428","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2405.17428","pdf_url":"https://arxiv.org/pdf/2405.17428","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":null,"raw_type":"text"},"sustainable_development_goals":[],"awards":[],"funders":[{"id":"https://openalex.org/F4320309480","display_name":"Nvidia","ror":"https://ror.org/03jdj4y14"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4399115160.pdf","grobid_xml":"https://content.openalex.org/works/W4399115160.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2748952813","https://openalex.org/W2306541987","https://openalex.org/W3135453479","https://openalex.org/W2947952289","https://openalex.org/W4322625948","https://openalex.org/W1986194836","https://openalex.org/W2896365120","https://openalex.org/W1986504168","https://openalex.org/W2095481932"],"abstract_inverted_index":{"Decoder-only":[0],"LLM-based":[1],"embedding":[2,11,16,46,84,172,260],"models":[3,12,182],"are":[4],"beginning":[5],"to":[6,36,63,76,167],"outperform":[7],"BERT":[8],"or":[9,79],"T5-based":[10],"in":[13,219,228,247],"general-purpose":[14],"text":[15],"tasks,":[17,202],"including":[18],"dense":[19],"vector-based":[20],"retrieval.":[21],"In":[22],"this":[23],"work,":[24],"we":[25,57,91,105,154],"introduce":[26,106],"NV-Embed,":[27],"incorporating":[28],"architectural":[29],"designs,":[30],"training":[31,103,116,152],"procedures,":[32],"and":[33,52,71,125,162,180,195,224],"curated":[34,126],"datasets":[35,166],"significantly":[37],"enhance":[38,88],"the":[39,81,93,156,169,184,188,204,208,216,220,225,229,233,252],"performance":[40,170],"of":[41,97,171,192,207,232,240,254],"LLM":[42],"as":[43],"a":[44,59,107,238],"versatile":[45],"model,":[47],"while":[48],"maintaining":[49],"its":[50],"simplicity":[51],"reproducibility.":[53],"For":[54,102,151],"model":[55,255],"architecture,":[56],"propose":[58],"latent":[60],"attention":[61,95],"layer":[62],"obtain":[64],"pooled":[65],"embeddings,":[66],"which":[67,139,236],"consistently":[68],"improves":[69,148],"retrieval":[70,120,149,243],"downstream":[72],"task":[73,144],"accuracy":[74,145],"compared":[75],"mean":[77],"pooling":[78],"using":[80],"last":[82],"token":[83],"from":[85],"LLMs.":[86],"To":[87],"representation":[89],"learning,":[90],"remove":[92],"causal":[94],"mask":[96],"LLMs":[98],"during":[99],"contrastive":[100,109,115],"training.":[101],"algorithm,":[104],"two-stage":[108],"instruction-tuning":[110],"method.":[111],"It":[112,213],"first":[113],"applies":[114],"with":[117],"instructions":[118],"on":[119,187],"datasets,":[121],"utilizing":[122],"in-batch":[123],"negatives":[124],"hard":[127],"negative":[128],"examples.":[129],"At":[130],"stage-2,":[131],"it":[132],"blends":[133],"various":[134],"non-retrieval":[135,143],"into":[136],"instruction":[137],"tuning,":[138],"not":[140],"only":[141],"enhances":[142],"but":[146],"also":[147,214],"performance.":[150],"data,":[153],"utilize":[155],"hard-negative":[157],"mining,":[158],"synthetic":[159],"data":[160],"generation":[161],"existing":[163],"public":[164],"available":[165],"boost":[168],"model.":[173],"By":[174],"combining":[175],"these":[176],"techniques,":[177],"our":[178],"NV-Embed-v1":[179],"NV-Embed-v2":[181],"obtained":[183],"No.1":[185],"position":[186],"MTEB":[189],"leaderboard":[190],"(as":[191],"May":[193],"24":[194],"August":[196],"30,":[197],"2024,":[198],"respectively)":[199],"across":[200],"56":[201],"demonstrating":[203],"sustained":[205],"effectiveness":[206],"proposed":[209],"methods":[210],"over":[211],"time.":[212],"achieved":[215],"highest":[217],"scores":[218,227],"Long":[221],"Doc":[222],"section":[223,231],"second-highest":[226],"QA":[230],"AIR":[234],"Benchmark,":[235],"covers":[237],"range":[239],"out-of-domain":[241],"information":[242],"topics":[244],"beyond":[245],"those":[246],"MTEB.":[248],"We":[249],"further":[250],"provide":[251],"analysis":[253],"compression":[256],"techniques":[257],"for":[258],"generalist":[259],"models.":[261]},"counts_by_year":[{"year":2025,"cited_by_count":9},{"year":2024,"cited_by_count":1}],"updated_date":"2026-03-20T23:20:44.827607","created_date":"2025-10-10T00:00:00"}