{"id":"https://openalex.org/W4387560076","doi":"https://doi.org/10.48550/arxiv.2310.05492","title":"How Abilities in Large Language Models are Affected by Supervised Fine-tuning Data Composition","display_name":"How Abilities in Large Language Models are Affected by Supervised Fine-tuning Data Composition","publication_year":2023,"publication_date":"2023-01-01","ids":{"openalex":"https://openalex.org/W4387560076","doi":"https://doi.org/10.48550/arxiv.2310.05492"},"language":"en","primary_location":{"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2310.05492","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_indexed_in_scopus":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":["Cornell University"],"type":"repository"},"license":"other-oa","license_id":"https://openalex.org/licenses/other-oa","version":"submittedVersion","is_accepted":false,"is_published":false},"type":"preprint","type_crossref":"posted-content","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/abs/2310.05492","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5008320336","display_name":"Guanting Dong","orcid":"https://orcid.org/0000-0002-2318-0281"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Dong, Guanting","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5038332216","display_name":"Hongyi Yuan","orcid":"https://orcid.org/0000-0003-2597-1973"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yuan, Hongyi","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5110849478","display_name":"Keming Lu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lu, Keming","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5041916202","display_name":"Cheng\u2010Peng Li","orcid":"https://orcid.org/0000-0002-6798-9541"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Chengpeng","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5085563975","display_name":"Mingfeng Xue","orcid":"https://orcid.org/0000-0002-4801-3754"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xue, Mingfeng","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5062188134","display_name":"Dayiheng Liu","orcid":"https://orcid.org/0000-0002-8755-8941"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Dayiheng","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100391896","display_name":"Wei Wang","orcid":"https://orcid.org/0000-0002-1717-5785"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Wei","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100591132","display_name":"Zheng Yuan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yuan, Zheng","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5102025315","display_name":"Chang Zhou","orcid":"https://orcid.org/0000-0001-9241-702X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhou, Chang","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5057864403","display_name":"Jingren Zhou","orcid":"https://orcid.org/0000-0002-4220-2634"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhou, Jingren","raw_affiliation_strings":[],"affiliations":[]}],"institution_assertions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"cited_by_count":3,"citation_normalized_percentile":{"value":0.917444,"is_in_top_1_percent":false,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":82,"max":85},"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.9958,"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":0.9958,"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":0.9728,"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/T11948","display_name":"Machine Learning in Materials Science","score":0.9167,"subfield":{"id":"https://openalex.org/subfields/2505","display_name":"Materials Chemistry"},"field":{"id":"https://openalex.org/fields/25","display_name":"Materials Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/code","display_name":"Code (set theory)","score":0.4976602}],"concepts":[{"id":"https://openalex.org/C7149132","wikidata":"https://www.wikidata.org/wiki/Q1377840","display_name":"Forgetting","level":2,"score":0.6906078},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6525472},{"id":"https://openalex.org/C40231798","wikidata":"https://www.wikidata.org/wiki/Q1333743","display_name":"Composition (language)","level":2,"score":0.5929573},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.4976602},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.419347},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.38854486},{"id":"https://openalex.org/C180747234","wikidata":"https://www.wikidata.org/wiki/Q23373","display_name":"Cognitive psychology","level":1,"score":0.37698102},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.19825041},{"id":"https://openalex.org/C41895202","wikidata":"https://www.wikidata.org/wiki/Q8162","display_name":"Linguistics","level":1,"score":0.11255416},{"id":"https://openalex.org/C199360897","wikidata":"https://www.wikidata.org/wiki/Q9143","display_name":"Programming language","level":1,"score":0.09088555},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.0},{"id":"https://openalex.org/C138885662","wikidata":"https://www.wikidata.org/wiki/Q5891","display_name":"Philosophy","level":0,"score":0.0}],"mesh":[],"locations_count":2,"locations":[{"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2310.05492","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_indexed_in_scopus":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":["Cornell University"],"type":"repository"},"license":"other-oa","license_id":"https://openalex.org/licenses/other-oa","version":"submittedVersion","is_accepted":false,"is_published":false},{"is_oa":false,"landing_page_url":"https://api.datacite.org/dois/10.48550/arxiv.2310.05492","pdf_url":null,"source":{"id":"https://openalex.org/S4393179698","display_name":"DataCite API","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_indexed_in_scopus":false,"is_core":false,"host_organization":"https://openalex.org/I4210145204","host_organization_name":"DataCite","host_organization_lineage":["https://openalex.org/I4210145204"],"host_organization_lineage_names":["DataCite"],"type":"metadata"},"license":null,"license_id":null,"version":null}],"best_oa_location":{"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2310.05492","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_indexed_in_scopus":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":["Cornell University"],"type":"repository"},"license":"other-oa","license_id":"https://openalex.org/licenses/other-oa","version":"submittedVersion","is_accepted":false,"is_published":false},"sustainable_development_goals":[{"score":0.79,"display_name":"Quality education","id":"https://metadata.un.org/sdg/4"}],"grants":[],"datasets":[],"versions":[],"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4310285384","https://openalex.org/W4289718052","https://openalex.org/W4287549553","https://openalex.org/W4248896073","https://openalex.org/W3183027292","https://openalex.org/W3116498279","https://openalex.org/W2974871044","https://openalex.org/W2905319430","https://openalex.org/W2164121020","https://openalex.org/W2145559838"],"abstract_inverted_index":{"Large":[0],"language":[1],"models":[2,119],"(LLMs)":[3],"with":[4,124,136,219],"enormous":[5],"pre-training":[6],"tokens":[7],"and":[8,18,77,96,106,117,131],"parameters":[9],"emerge":[10],"diverse":[11],"abilities,":[12],"including":[13,99],"math":[14],"reasoning,":[15,74],"code":[16,75,132],"generation,":[17,76],"instruction":[19],"following.":[20],"These":[21],"abilities":[22,55,80,142,158,218],"are":[23],"further":[24],"enhanced":[25],"by":[26],"supervised":[27],"fine-tuning":[28],"(SFT).":[29],"While":[30],"the":[31,51,67,91,177,186],"open-source":[32],"community":[33],"has":[34],"explored":[35],"ad-hoc":[36],"SFT":[37,57,107,192],"for":[38],"enhancing":[39],"individual":[40],"capabilities,":[41],"proprietary":[42],"LLMs":[43],"exhibit":[44],"versatility":[45],"across":[46],"various":[47,97,157],"skills.":[48],"Therefore,":[49],"understanding":[50],"facilitation":[52],"of":[53,69,127,179,191],"multiple":[54,199,217],"via":[56],"is":[58,171],"paramount.":[59],"In":[60,189],"this":[61],"study,":[62],"we":[63,150,194],"specifically":[64],"focuses":[65],"on":[66],"interplay":[68],"data":[70,100,138,152,161,170,181],"composition":[71,102,153,180,187],"between":[72,93],"mathematical":[73],"general":[78,141],"human-aligning":[79],"during":[81],"SFT.":[82],"We":[83],"propose":[84],"four":[85],"intriguing":[86],"research":[87],"questions":[88],"to":[89,155,166,215],"explore":[90],"association":[92],"model":[94,104],"performance":[95,123,167,183],"factors":[98],"amount,":[101,139],"ratio,":[103],"size":[105],"strategies.":[108],"Our":[109,173,204],"experiments":[110],"reveal":[111],"that":[112,196],"distinct":[113],"capabilities":[114],"scale":[115],"differently":[116],"larger":[118],"generally":[120],"show":[121],"superior":[122],"same":[125],"amount":[126,178],"data.":[128],"Mathematical":[129],"reasoning":[130],"generation":[133],"consistently":[134],"improve":[135],"increasing":[137],"whereas":[140],"plateau":[143],"after":[144],"roughly":[145],"a":[146,212],"thousand":[147],"samples.":[148],"Moreover,":[149],"observe":[151],"appears":[154],"enhance":[156],"under":[159],"limited":[160],"conditions,":[162],"yet":[163],"can":[164],"lead":[165],"conflicts":[168],"when":[169],"plentiful.":[172],"findings":[174],"also":[175],"suggest":[176],"influences":[182],"more":[184],"than":[185],"ratio.":[188],"analysis":[190],"strategies,":[193],"find":[195],"sequentially":[197],"learning":[198],"skills":[200],"risks":[201],"catastrophic":[202],"forgetting.":[203],"proposed":[205],"Dual-stage":[206],"Mixed":[207],"Fine-tuning":[208],"(DMT)":[209],"strategy":[210],"offers":[211],"promising":[213],"solution":[214],"learn":[216],"different":[220],"scaling":[221],"patterns.":[222]},"abstract_inverted_index_v3":null,"cited_by_api_url":"https://api.openalex.org/works?filter=cites:W4387560076","counts_by_year":[{"year":2025,"cited_by_count":2},{"year":2024,"cited_by_count":1}],"updated_date":"2025-05-03T03:41:52.059719","created_date":"2023-10-12"}