{"id":"https://openalex.org/W4412888885","doi":"https://doi.org/10.18653/v1/2025.findings-acl.66","title":"Separate the Wheat from the Chaff: A Post-Hoc Approach to Safety Re-Alignment for Fine-Tuned Language Models","display_name":"Separate the Wheat from the Chaff: A Post-Hoc Approach to Safety Re-Alignment for Fine-Tuned Language Models","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4412888885","doi":"https://doi.org/10.18653/v1/2025.findings-acl.66"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2025.findings-acl.66","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.66","pdf_url":"https://aclanthology.org/2025.findings-acl.66.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: ACL 2025","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2025.findings-acl.66.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5087166824","display_name":"Di Wu","orcid":"https://orcid.org/0000-0002-6818-1886"},"institutions":[{"id":"https://openalex.org/I204983213","display_name":"Harbin Institute of Technology","ror":"https://ror.org/01yqg2h08","country_code":"CN","type":"education","lineage":["https://openalex.org/I204983213"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Di Wu","raw_affiliation_strings":["Research Center for Social Computing and Information Retrieval Harbin Institute of Technology , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Research Center for Social Computing and Information Retrieval Harbin Institute of Technology , China","institution_ids":["https://openalex.org/I204983213"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5091489017","display_name":"Xinzheng Lu","orcid":"https://orcid.org/0000-0002-3313-7420"},"institutions":[{"id":"https://openalex.org/I204983213","display_name":"Harbin Institute of Technology","ror":"https://ror.org/01yqg2h08","country_code":"CN","type":"education","lineage":["https://openalex.org/I204983213"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Xin Lu","raw_affiliation_strings":["Research Center for Social Computing and Information Retrieval Harbin Institute of Technology , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Research Center for Social Computing and Information Retrieval Harbin Institute of Technology , China","institution_ids":["https://openalex.org/I204983213"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5040196605","display_name":"Yanyan Zhao","orcid":"https://orcid.org/0000-0002-1166-1459"},"institutions":[{"id":"https://openalex.org/I204983213","display_name":"Harbin Institute of Technology","ror":"https://ror.org/01yqg2h08","country_code":"CN","type":"education","lineage":["https://openalex.org/I204983213"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yanyan Zhao","raw_affiliation_strings":["Research Center for Social Computing and Information Retrieval Harbin Institute of Technology , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Research Center for Social Computing and Information Retrieval Harbin Institute of Technology , China","institution_ids":["https://openalex.org/I204983213"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5039365023","display_name":"Bing Qin","orcid":"https://orcid.org/0000-0002-1305-7183"},"institutions":[{"id":"https://openalex.org/I204983213","display_name":"Harbin Institute of Technology","ror":"https://ror.org/01yqg2h08","country_code":"CN","type":"education","lineage":["https://openalex.org/I204983213"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Bing Qin","raw_affiliation_strings":["Research Center for Social Computing and Information Retrieval Harbin Institute of Technology , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Research Center for Social Computing and Information Retrieval Harbin Institute of Technology , China","institution_ids":["https://openalex.org/I204983213"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I204983213"],"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":"1210","last_page":"1225"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.9894999861717224,"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":0.9894999861717224,"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":0.9491999745368958,"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/chaff","display_name":"Chaff","score":0.9142605066299438},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.669570803642273},{"id":"https://openalex.org/keywords/natural-language-processing","display_name":"Natural language processing","score":0.373283326625824},{"id":"https://openalex.org/keywords/geography","display_name":"Geography","score":0.0679379403591156}],"concepts":[{"id":"https://openalex.org/C78573896","wikidata":"https://www.wikidata.org/wiki/Q30470787","display_name":"Chaff","level":2,"score":0.9142605066299438},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.669570803642273},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.373283326625824},{"id":"https://openalex.org/C205649164","wikidata":"https://www.wikidata.org/wiki/Q1071","display_name":"Geography","level":0,"score":0.0679379403591156},{"id":"https://openalex.org/C166957645","wikidata":"https://www.wikidata.org/wiki/Q23498","display_name":"Archaeology","level":1,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2025.findings-acl.66","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.66","pdf_url":"https://aclanthology.org/2025.findings-acl.66.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: ACL 2025","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2025.findings-acl.66","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.66","pdf_url":"https://aclanthology.org/2025.findings-acl.66.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: ACL 2025","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[{"id":"https://openalex.org/G2656429412","display_name":null,"funder_award_id":"62176078","funder_id":"https://openalex.org/F4320321001","funder_display_name":"National Natural Science Foundation of China"}],"funders":[{"id":"https://openalex.org/F4320321001","display_name":"National Natural Science Foundation of China","ror":"https://ror.org/01h0zpd94"}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4412888885.pdf","grobid_xml":"https://content.openalex.org/works/W4412888885.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2899084033","https://openalex.org/W2748952813","https://openalex.org/W2365762530","https://openalex.org/W2157107224","https://openalex.org/W2768124060","https://openalex.org/W4408056645","https://openalex.org/W1511072557","https://openalex.org/W2380447882","https://openalex.org/W2007443611"],"abstract_inverted_index":{"Although":[0],"large":[1],"language":[2],"models":[3,100],"(LLMs)":[4],"achieve":[5],"effective":[6],"safety":[7,18,28,50,96,102],"alignment":[8,29],"at":[9],"the":[10,27,66,71,75,95],"time":[11],"of":[12,30,55,77,98],"release,":[13],"they":[14],"still":[15],"face":[16],"various":[17,80],"challenges.A":[19],"key":[20],"issue":[21],"is":[22,57,120],"that":[23,48,91],"fine-tuning":[24,85],"often":[25],"compromises":[26],"LLMs.To":[31],"address":[32],"this":[33],"issue,":[34],"we":[35],"propose":[36],"a":[37],"method":[38],"named":[39],"IRR":[40,56,78,92],"(Identify,":[41],"Remove,":[42],"and":[43,60,86,108],"Recalibrate":[44],"for":[45,52],"Safety":[46],"Realignment)":[47],"performs":[49],"realignment":[51],"LLMs.The":[53],"core":[54],"to":[58],"identify":[59],"remove":[61],"unsafe":[62],"delta":[63],"parameters":[64],"from":[65],"fine-tuned":[67],"models,":[68],"while":[69,111],"recalibrating":[70],"retained":[72],"parameters.We":[73],"evaluate":[74],"effectiveness":[76],"across":[79],"datasets,":[81],"including":[82],"both":[83],"full":[84],"LoRA":[87],"methods.Our":[88],"results":[89],"demonstrate":[90],"significantly":[93],"enhances":[94],"performance":[97,114],"finetuned":[99],"on":[101,115],"benchmarks,":[103],"such":[104],"as":[105],"harmful":[106],"queries":[107],"jailbreak":[109],"attacks,":[110],"maintaining":[112],"their":[113],"downstream":[116],"tasks.The":[117],"source":[118],"code":[119],"available":[121],"at:":[122]},"counts_by_year":[],"updated_date":"2026-07-14T23:27:15.235271","created_date":"2025-10-10T00:00:00"}