{"id":"https://openalex.org/W4391912901","doi":"https://doi.org/10.48550/arxiv.2402.09668","title":"How to Train Data-Efficient LLMs","display_name":"How to Train Data-Efficient LLMs","publication_year":2024,"publication_date":"2024-02-15","ids":{"openalex":"https://openalex.org/W4391912901","doi":"https://doi.org/10.48550/arxiv.2402.09668"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2402.09668","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2402.09668","pdf_url":"https://arxiv.org/pdf/2402.09668","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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"},"volume":null,"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2402.09668","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5005200778","display_name":"Noveen Sachdeva","orcid":"https://orcid.org/0000-0003-1228-746X","observed_orcids":["https://orcid.org/0000-0003-1228-746X"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sachdeva, Noveen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5103253242","display_name":"Benjamin Coleman","orcid":"https://orcid.org/0000-0002-2010-796X","observed_orcids":["https://orcid.org/0000-0002-2010-796X"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Coleman, Benjamin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5090597599","display_name":"Wang-Cheng Kang","orcid":"https://orcid.org/0009-0006-8795-3665","observed_orcids":["https://orcid.org/0009-0006-8795-3665"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kang, Wang-Cheng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5077817759","display_name":"Jianmo Ni","orcid":"https://orcid.org/0000-0002-6863-8073","observed_orcids":["https://orcid.org/0000-0002-6863-8073"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ni, Jianmo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5079085366","display_name":"Lichan Hong","orcid":"https://orcid.org/0009-0004-9563-554X","observed_orcids":["https://orcid.org/0009-0004-9563-554X"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hong, Lichan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5028125399","display_name":"Ed H.","orcid":"https://orcid.org/0000-0003-3230-5338","observed_orcids":["https://orcid.org/0000-0003-3230-5338"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chi, Ed H.","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5048489384","display_name":"James Caverlee","orcid":"https://orcid.org/0000-0001-8350-8528","observed_orcids":["https://orcid.org/0000-0001-8350-8528"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Caverlee, James","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5021827617","display_name":"Julian McAuley","orcid":"https://orcid.org/0000-0003-0955-7588","observed_orcids":["https://orcid.org/0000-0003-0955-7588"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"McAuley, Julian","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5055484401","display_name":"Derek Zhiyuan Cheng","orcid":"https://orcid.org/0009-0000-7943-8328","observed_orcids":["https://orcid.org/0009-0000-7943-8328"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Cheng, Derek Zhiyuan","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":true,"cited_by_count":8,"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":0.9394999742507935,"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":0.9394999742507935,"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":0.01640000008046627,"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":0.015200000256299973,"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/data-efficient-training","display_name":"data-efficient training","score":0.7870000004768372},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.8410000205039978},{"id":"https://openalex.org/keywords/pretraining","display_name":"pre-training","score":0.753000020980835},{"id":"https://openalex.org/keywords/training-sample-selection","display_name":"training data selection","score":0.8320000171661377},{"id":"https://openalex.org/keywords/data-quality-estimation","display_name":"data quality estimation","score":0.7009999752044678},{"id":"https://openalex.org/keywords/zero-shot-reasoning","display_name":"zero-shot reasoning","score":0.7739999890327454},{"id":"https://openalex.org/keywords/diversity-sampling","display_name":"diversity sampling","score":0.48500001430511475},{"id":"https://openalex.org/keywords/pareto-front","display_name":"Pareto front","score":0.6439999938011169},{"id":"https://openalex.org/keywords/training-cost-reduction","display_name":"training cost reduction","score":0.6269999742507935}],"concepts":[{"id":"https://openalex.org/C17744445","wikidata":"https://www.wikidata.org/wiki/Q36442","display_name":"Political science","level":0,"score":0.3160645067691803}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2402.09668","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2402.09668","pdf_url":"https://arxiv.org/pdf/2402.09668","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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.2402.09668","doi":"https://doi.org/10.48550/arxiv.2402.09668","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2402.09668","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2402.09668","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2402.09668","pdf_url":"https://arxiv.org/pdf/2402.09668","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4391912901.pdf","grobid_xml":"https://content.openalex.org/works/W4391912901.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7169877707","https://openalex.org/W4412889002","https://openalex.org/W4414981092","https://openalex.org/W7133241490","https://openalex.org/W4406733243","https://openalex.org/W4404307413","https://openalex.org/W7171312960","https://openalex.org/W4415068397","https://openalex.org/W4416037021","https://openalex.org/W4386644265"],"abstract_inverted_index":{"The":[0],"training":[1,32,83,149],"of":[2,28,55,73,81,105,110,136,156],"large":[3],"language":[4],"models":[5,93,141],"(LLMs)":[6],"is":[7],"expensive.":[8],"In":[9,102],"this":[10],"paper,":[11],"we":[12,88,116,153],"study":[13],"data-efficient":[14],"approaches":[15],"for":[16],"pre-training":[17,114],"LLMs,":[18],"i.e.,":[19],"techniques":[20],"that":[21,118],"aim":[22],"to":[23,37,76,97,163],"optimize":[24],"the":[25,39,61,69,79,94,123,134,137,157],"Pareto":[26],"frontier":[27],"model":[29],"quality":[30,80],"and":[31,52,57,113,120],"resource/data":[33],"consumption.":[34],"We":[35],"seek":[36],"understand":[38],"tradeoffs":[40],"associated":[41],"with":[42],"data":[43,95,145],"selection":[44],"routines":[45],"based":[46],"on":[47,143],"(i)":[48],"expensive-to-compute":[49],"data-quality":[50],"estimates,":[51],"(ii)":[53],"maximization":[54],"coverage":[56],"diversity-based":[58],"measures":[59],"in":[60,126],"feature":[62],"space.":[63],"Our":[64],"first":[65],"technique,":[66],"Ask-LLM,":[67],"leverages":[68],"zero-shot":[70],"reasoning":[71],"capabilities":[72],"instruction-tuned":[74],"LLMs":[75],"directly":[77],"assess":[78],"a":[82,99],"example.":[84],"To":[85],"target":[86],"coverage,":[87],"propose":[89],"Density":[90,121],"sampling,":[91],"which":[92],"distribution":[96],"select":[98],"diverse":[100],"sample.":[101],"our":[103],"comparison":[104],"19":[106],"samplers,":[107],"involving":[108],"hundreds":[109],"evaluation":[111],"tasks":[112],"runs,":[115],"find":[117],"Ask-LLM":[119,144],"are":[122],"best":[124],"methods":[125],"their":[127],"respective":[128],"categories.":[129],"Coverage":[130],"sampling":[131],"can":[132],"recover":[133],"performance":[135],"full":[138],"data,":[139],"while":[140,160],"trained":[142],"consistently":[146],"outperform":[147],"full-data":[148],"--":[150],"even":[151],"when":[152],"reject":[154],"90%":[155],"original":[158],"dataset,":[159],"converging":[161],"up":[162],"70%":[164],"faster.":[165]},"counts_by_year":[{"year":2026,"cited_by_count":2},{"year":2025,"cited_by_count":5},{"year":2024,"cited_by_count":1}],"updated_date":"2026-10-17T09:40:24.721904","created_date":"2024-02-18T00:00:00"}