{"id":"https://openalex.org/W4392271198","doi":"https://doi.org/10.48550/arxiv.2402.17400","title":"Investigating Continual Pretraining in Large Language Models: Insights and Implications","display_name":"Investigating Continual Pretraining in Large Language Models: Insights and Implications","publication_year":2024,"publication_date":"2024-02-27","ids":{"openalex":"https://openalex.org/W4392271198","doi":"https://doi.org/10.48550/arxiv.2402.17400"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2402.17400","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2402.17400","pdf_url":"https://arxiv.org/pdf/2402.17400","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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2402.17400","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":null,"display_name":"Y\u0131ld\u0131z, \u00c7a\u011fatay","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Y\u0131ld\u0131z, \u00c7a\u011fatay","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5108904670","display_name":"Nishaanth Kanna Ravichandran","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ravichandran, Nishaanth Kanna","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Sharma, Nitin","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I8087733","display_name":"University of T\u00fcbingen","ror":"https://ror.org/03a1kwz48","country_code":"DE","type":"education","lineage":["https://openalex.org/I8087733"]}],"countries":["DE"],"is_corresponding":false,"raw_author_name":"Sharma, Nitin","raw_affiliation_strings":["University of T\u00fcbingen"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of T\u00fcbingen","institution_ids":["https://openalex.org/I8087733"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5061457780","display_name":"Matthias Bethge","orcid":"https://orcid.org/0000-0002-6417-7812","observed_orcids":["https://orcid.org/0000-0002-6417-7812","https://orcid.org/0009-0008-5601-6120"]},"institutions":[{"id":"https://openalex.org/I8087733","display_name":"University of T\u00fcbingen","ror":"https://ror.org/03a1kwz48","country_code":"DE","type":"education","lineage":["https://openalex.org/I8087733"]}],"countries":["DE"],"is_corresponding":false,"raw_author_name":"Bethge, Matthias","raw_affiliation_strings":["University of T\u00fcbingen"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of T\u00fcbingen","institution_ids":["https://openalex.org/I8087733"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5057377890","display_name":"Beyza Ermi\u015f","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I4401726832","display_name":"T\u00fcbingen AI Center","ror":"https://ror.org/0107nyd78","country_code":"DE","type":"education","lineage":["https://openalex.org/I149899117","https://openalex.org/I4210135521","https://openalex.org/I4401726832","https://openalex.org/I8087733"]},{"id":"https://openalex.org/I8087733","display_name":"University of T\u00fcbingen","ror":"https://ror.org/03a1kwz48","country_code":"DE","type":"education","lineage":["https://openalex.org/I8087733"]}],"countries":["DE"],"is_corresponding":false,"raw_author_name":"Ermis, Beyza","raw_affiliation_strings":["T\u00fcbingen AI Center , University of T\u00fcbingen"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"T\u00fcbingen AI Center , University of T\u00fcbingen","institution_ids":["https://openalex.org/I4401726832","https://openalex.org/I8087733"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":83,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.9922000169754028,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.9922000169754028,"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/T10028","display_name":"Topic Modeling","score":0.007000000216066837,"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":0.00019999999494757503,"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/continual-learning","display_name":"continual learning","score":0.8650000095367432},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.9350000023841858},{"id":"https://openalex.org/keywords/continual-pre-training","display_name":"continual pre-training","score":0.9750000238418579},{"id":"https://openalex.org/keywords/domain-adaptive-pretraining","display_name":"domain-adaptive pretraining","score":0.8090000152587891},{"id":"https://openalex.org/keywords/benchmark","display_name":"benchmark","score":0.36899998784065247},{"id":"https://openalex.org/keywords/knowledge-retention","display_name":"knowledge retention","score":0.6669999957084656},{"id":"https://openalex.org/keywords/forgetting","display_name":"forgetting","score":0.6980000138282776},{"id":"https://openalex.org/keywords/model-size-effects","display_name":"model size effects","score":0.7350000143051147},{"id":"https://openalex.org/keywords/knowledge-transfer","display_name":"knowledge transfer","score":0.7160000205039978},{"id":"https://openalex.org/keywords/domain-similarity","display_name":"domain similarity","score":0.6169999837875366},{"id":"https://openalex.org/keywords/perplexity","display_name":"perplexity","score":0.7049999833106995},{"id":"https://openalex.org/keywords/gpt-2","display_name":"GPT-2","score":0.9359999895095825}],"concepts":[{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.441225528717041},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.4012817442417145},{"id":"https://openalex.org/C180747234","wikidata":"https://www.wikidata.org/wiki/Q23373","display_name":"Cognitive psychology","level":1,"score":0.38418176770210266},{"id":"https://openalex.org/C41895202","wikidata":"https://www.wikidata.org/wiki/Q8162","display_name":"Linguistics","level":1,"score":0.38329026103019714},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.36337706446647644},{"id":"https://openalex.org/C188147891","wikidata":"https://www.wikidata.org/wiki/Q147638","display_name":"Cognitive science","level":1,"score":0.34714365005493164},{"id":"https://openalex.org/C138885662","wikidata":"https://www.wikidata.org/wiki/Q5891","display_name":"Philosophy","level":0,"score":0.15087366104125977}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2402.17400","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2402.17400","pdf_url":"https://arxiv.org/pdf/2402.17400","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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2402.17400","doi":"https://doi.org/10.48550/arxiv.2402.17400","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2402.17400","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2402.17400","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2402.17400","pdf_url":"https://arxiv.org/pdf/2402.17400","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":null,"license_id":null,"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/W4392271198.pdf","grobid_xml":"https://content.openalex.org/works/W4392271198.grobid-xml"},"referenced_works_count":28,"referenced_works":["https://openalex.org/W2896457183","https://openalex.org/W2952357537","https://openalex.org/W2963748441","https://openalex.org/W2995409942","https://openalex.org/W3001279689","https://openalex.org/W3017961061","https://openalex.org/W3034793897","https://openalex.org/W3112170794","https://openalex.org/W3133450183","https://openalex.org/W4221144388","https://openalex.org/W4221147360","https://openalex.org/W4286379669","https://openalex.org/W4287896269","https://openalex.org/W4318751794","https://openalex.org/W4376167151","https://openalex.org/W4384263489","https://openalex.org/W4385572749","https://openalex.org/W4385573474","https://openalex.org/W4385714663","https://openalex.org/W4389520299","https://openalex.org/W4389520380","https://openalex.org/W4391555991","https://openalex.org/W4404391902","https://openalex.org/W4404781090","https://openalex.org/W4405425667","https://openalex.org/W4405626946","https://openalex.org/W4414037010","https://openalex.org/W4416017454"],"related_works":["https://openalex.org/W4410371768","https://openalex.org/W4399990130","https://openalex.org/W7136463706","https://openalex.org/W4319653544","https://openalex.org/W4414464247","https://openalex.org/W4412889571","https://openalex.org/W7077069873","https://openalex.org/W7117534842","https://openalex.org/W3015881266","https://openalex.org/W7090701857"],"abstract_inverted_index":{"Continual":[0],"learning":[1,106,185],"(CL)":[2],"in":[3,30,142,237],"large":[4],"language":[5],"models":[6,23,140,154,171],"(LLMs)":[7],"is":[8,36,146],"an":[9],"evolving":[10],"domain":[11,150],"that":[12,228],"focuses":[13],"on":[14,37,67,105,165],"developing":[15],"efficient":[16],"and":[17,27,108,116,145,186,222,247],"sustainable":[18],"training":[19,78,216],"strategies":[20],"to":[21,24,44,50,86,92,149,175,203,219],"adapt":[22],"emerging":[25,119],"knowledge":[26,123,245],"achieve":[28,156],"robustness":[29],"dynamic":[31],"environments.":[32],"Our":[33,128],"primary":[34],"emphasis":[35],"continual":[38,68,135,176,189,199],"domain-adaptive":[39],"pretraining,":[40,177],"a":[41,71,82,232,240],"process":[42],"designed":[43,85],"equip":[45],"LLMs":[46,91,202],"with":[47],"the":[48,88,100,114,122,166,179,207],"ability":[49],"integrate":[51],"new":[52,83,233],"information":[53],"from":[54],"various":[55],"domains":[56,120,210,217],"while":[57,214],"retaining":[58],"previously":[59],"learned":[60],"knowledge.":[61],"Since":[62],"existing":[63],"works":[64],"concentrate":[65],"mostly":[66],"fine-tuning":[69],"for":[70,235],"limited":[72],"selection":[73],"of":[74,90,102,118,183,195,209,244],"downstream":[75,192],"tasks":[76],"or":[77],"domains,":[79],"we":[80],"introduce":[81],"benchmark":[84,234],"measure":[87],"adaptability":[89],"changing":[93],"pretraining":[94,136,190,200],"data":[95],"landscapes.":[96],"We":[97,226],"further":[98],"examine":[99],"impact":[101],"model":[103],"size":[104],"efficacy":[107],"forgetting,":[109,187],"as":[110,112],"well":[111],"how":[113],"progression":[115],"similarity":[117,213],"affect":[121],"transfer":[124,221,248],"within":[125],"these":[126],"models.":[127],"findings":[129],"uncover":[130],"several":[131],"key":[132],"insights:":[133],"(i)":[134],"consistently":[137],"improves":[138],"<1.5B":[139],"studied":[141],"this":[143],"work":[144],"also":[147],"superior":[148],"adaptation,":[151],"(ii)":[152],"larger":[153],"always":[155],"better":[157,205,220],"perplexity":[158],"than":[159],"smaller":[160,170],"ones":[161],"when":[162,206],"continually":[163],"pretrained":[164],"same":[167],"corpus,":[168],"(iii)":[169],"are":[172],"particularly":[173],"sensitive":[174],"showing":[178],"most":[180],"significant":[181],"rates":[182],"both":[184],"(iv)":[188],"boosts":[191],"task":[193],"performance":[194,224],"GPT-2":[196],"family,":[197],"(v)":[198],"enables":[201],"specialize":[204],"sequence":[208],"shows":[211],"semantic":[212],"randomizing":[215],"leads":[218],"final":[223],"otherwise.":[225],"posit":[227],"our":[229],"research":[230],"establishes":[231],"CL":[236],"LLMs,":[238],"providing":[239],"more":[241],"realistic":[242],"evaluation":[243],"retention":[246],"across":[249],"diverse":[250],"domains.":[251]},"counts_by_year":[{"year":2026,"cited_by_count":28},{"year":2025,"cited_by_count":36},{"year":2024,"cited_by_count":19}],"updated_date":"2026-11-10T11:39:43.021295","created_date":"2024-03-05T00:00:00"}