{"id":"https://openalex.org/W4404782006","doi":"https://doi.org/10.18653/v1/2024.findings-emnlp.915","title":"Merge to Learn: Efficiently Adding Skills to Language Models with Model Merging","display_name":"Merge to Learn: Efficiently Adding Skills to Language Models with Model Merging","publication_year":2024,"publication_date":"2024-01-01","ids":{"openalex":"https://openalex.org/W4404782006","doi":"https://doi.org/10.18653/v1/2024.findings-emnlp.915"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2024.findings-emnlp.915","is_oa":true,"landing_page_url":"http://dx.doi.org/10.18653/v1/2024.findings-emnlp.915","pdf_url":"https://aclanthology.org/2024.findings-emnlp.915.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":"Findings of the Association for Computational Linguistics: EMNLP 2024","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2024.findings-emnlp.915.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5088629436","display_name":"Jacob Morrison","orcid":"https://orcid.org/0000-0001-8592-4744"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jacob Morrison","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5088517824","display_name":"Noah A. Smith","orcid":"https://orcid.org/0000-0002-2310-6380"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Noah A. Smith","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5082305994","display_name":"Hannaneh Hajishirzi","orcid":"https://orcid.org/0000-0002-1055-6657"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hannaneh Hajishirzi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5079904764","display_name":"Pang Wei Koh","orcid":"https://orcid.org/0000-0003-4330-6969"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Pang Wei Koh","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5008013895","display_name":"Jesse Dodge","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jesse Dodge","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5029074038","display_name":"Pradeep Dasigi","orcid":"https://orcid.org/0000-0001-7127-1316"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Pradeep Dasigi","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":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"15604","last_page":"15621"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.9731000065803528,"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":0.9731000065803528,"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":0.9531000256538391,"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/T10215","display_name":"Semantic Web and Ontologies","score":0.9023000001907349,"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/merge","display_name":"Merge (version control)","score":0.8275878429412842},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.8069452047348022},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.44828781485557556},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.43516722321510315},{"id":"https://openalex.org/keywords/programming-language","display_name":"Programming language","score":0.42136090993881226},{"id":"https://openalex.org/keywords/natural-language-processing","display_name":"Natural language processing","score":0.3910466134548187},{"id":"https://openalex.org/keywords/parallel-computing","display_name":"Parallel computing","score":0.1796855628490448}],"concepts":[{"id":"https://openalex.org/C197129107","wikidata":"https://www.wikidata.org/wiki/Q1921621","display_name":"Merge (version control)","level":2,"score":0.8275878429412842},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8069452047348022},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.44828781485557556},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.43516722321510315},{"id":"https://openalex.org/C199360897","wikidata":"https://www.wikidata.org/wiki/Q9143","display_name":"Programming language","level":1,"score":0.42136090993881226},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.3910466134548187},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.1796855628490448}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2024.findings-emnlp.915","is_oa":true,"landing_page_url":"http://dx.doi.org/10.18653/v1/2024.findings-emnlp.915","pdf_url":"https://aclanthology.org/2024.findings-emnlp.915.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":"Findings of the Association for Computational Linguistics: EMNLP 2024","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2024.findings-emnlp.915","is_oa":true,"landing_page_url":"http://dx.doi.org/10.18653/v1/2024.findings-emnlp.915","pdf_url":"https://aclanthology.org/2024.findings-emnlp.915.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":"Findings of the Association for Computational Linguistics: EMNLP 2024","raw_type":"proceedings-article"},"sustainable_development_goals":[{"display_name":"Quality Education","score":0.6299999952316284,"id":"https://metadata.un.org/sdg/4"}],"awards":[{"id":"https://openalex.org/G8876996369","display_name":null,"funder_award_id":"N00014","funder_id":"https://openalex.org/F4320337345","funder_display_name":"Office of Naval Research"}],"funders":[{"id":"https://openalex.org/F4320320671","display_name":"National Research Foundation","ror":"https://ror.org/05s0g1g46"},{"id":"https://openalex.org/F4320320709","display_name":"National Research Foundation Singapore","ror":"https://ror.org/03cpyc314"},{"id":"https://openalex.org/F4320337345","display_name":"Office of Naval Research","ror":"https://ror.org/00rk2pe57"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4404782006.pdf","grobid_xml":"https://content.openalex.org/works/W4404782006.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4234886518","https://openalex.org/W2389591058","https://openalex.org/W2382112581","https://openalex.org/W3124036233","https://openalex.org/W4229787472","https://openalex.org/W2486541857","https://openalex.org/W2108840191","https://openalex.org/W2759366996","https://openalex.org/W2110679372","https://openalex.org/W3204019825"],"abstract_inverted_index":{"Adapting":[0],"general-purpose":[1],"language":[2],"models":[3,29,46,88],"to":[4,30,44,113,131],"new":[5,17,21,42,51],"skills":[6,22,43,52],"is":[7,82,93,103],"currently":[8],"an":[9],"expensive":[10],"process":[11],"that":[12,77,100],"must":[13],"be":[14],"repeated":[15],"as":[16,118],"instruction":[18],"datasets":[19],"targeting":[20],"are":[23],"created,":[24],"or":[25,134],"can":[26],"cause":[27],"the":[28,38,50,59,78,87],"forget":[31],"older":[32],"skills.In":[33],"this":[34],"work,":[35],"we":[36,75],"investigate":[37],"effectiveness":[39],"of":[40],"adding":[41],"preexisting":[45],"by":[47],"training":[48,102],"on":[49,68,89],"in":[53,110],"isolation":[54],"and":[55,73,116],"later":[56],"merging":[57],"with":[58,124],"general":[60],"model":[61,122],"(e.g.":[62],"using":[63],"task":[64],"vectors).In":[65],"experiments":[66,97],"focusing":[67],"scientific":[69],"literature":[70],"understanding,":[71],"safety,":[72],"coding,":[74],"find":[76],"parallel-train-then-merge":[79],"procedure,":[80],"which":[81],"significantly":[83],"cheaper":[84],"than":[85],"retraining":[86],"updated":[90],"data":[91],"mixtures,":[92],"often":[94],"comparably":[95],"effective.Our":[96],"also":[98],"show":[99],"parallel":[101],"especially":[104],"well-suited":[105],"for":[106],"enabling":[107],"safety":[108],"features":[109],"LMs":[111],"relative":[112],"continued":[114],"finetuning":[115],"retraining,":[117],"it":[119],"dramatically":[120],"improves":[121],"compliance":[123],"safe":[125],"prompts":[126],"while":[127],"preserving":[128],"its":[129],"ability":[130],"refuse":[132],"dangerous":[133],"harmful":[135],"prompts.":[136]},"counts_by_year":[],"updated_date":"2026-07-14T23:27:15.235271","created_date":"2025-10-10T00:00:00"}