{"id":"https://openalex.org/W7166831519","doi":"https://doi.org/10.18653/v1/2026.acl-long.9","title":"Learning Diverse Responses with Prefix-Conditioned Supervised Fine-Tuning","display_name":"Learning Diverse Responses with Prefix-Conditioned Supervised Fine-Tuning","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W7166831519","doi":"https://doi.org/10.18653/v1/2026.acl-long.9"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2026.acl-long.9","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.acl-long.9","pdf_url":"https://aclanthology.org/2026.acl-long.9.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":"Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2026.acl-long.9.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5139767675","display_name":"Zhiyuan Fan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhiyuan Fan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139787104","display_name":"Guanqiao Chen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Guanqiao Chen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5079605407","display_name":"Yanyi Huang","orcid":"https://orcid.org/0000-0002-7297-1266"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yanyi Huang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139782427","display_name":"Mingkuan Zhao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Mingkuan Zhao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139791718","display_name":"Dadi Guo","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Dadi Guo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5139812376","display_name":"Yi R. Fung","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yi R. Fung","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":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.82128418,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"247","last_page":"276"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10201","display_name":"Speech Recognition and Synthesis","score":0.5870000123977661,"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/T10201","display_name":"Speech Recognition and Synthesis","score":0.5870000123977661,"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/T10860","display_name":"Speech and Audio Processing","score":0.11050000041723251,"subfield":{"id":"https://openalex.org/subfields/1711","display_name":"Signal Processing"},"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/T10057","display_name":"Face and Expression Recognition","score":0.03180000185966492,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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/set","display_name":"Set (abstract data type)","score":0.30090001225471497},{"id":"https://openalex.org/keywords/feature","display_name":"Feature (linguistics)","score":0.2793999910354614},{"id":"https://openalex.org/keywords/matching","display_name":"Matching (statistics)","score":0.27489998936653137},{"id":"https://openalex.org/keywords/supervised-learning","display_name":"Supervised learning","score":0.2635999917984009},{"id":"https://openalex.org/keywords/identification","display_name":"Identification (biology)","score":0.2492000013589859}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.48980000615119934},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.44699999690055847},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.30090001225471497},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.2793999910354614},{"id":"https://openalex.org/C165064840","wikidata":"https://www.wikidata.org/wiki/Q1321061","display_name":"Matching (statistics)","level":2,"score":0.27489998936653137},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.27480000257492065},{"id":"https://openalex.org/C136389625","wikidata":"https://www.wikidata.org/wiki/Q334384","display_name":"Supervised learning","level":3,"score":0.2635999917984009},{"id":"https://openalex.org/C116834253","wikidata":"https://www.wikidata.org/wiki/Q2039217","display_name":"Identification (biology)","level":2,"score":0.2492000013589859},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.24629999697208405},{"id":"https://openalex.org/C192209626","wikidata":"https://www.wikidata.org/wiki/Q190909","display_name":"Focus (optics)","level":2,"score":0.24619999527931213}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2026.acl-long.9","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.acl-long.9","pdf_url":"https://aclanthology.org/2026.acl-long.9.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":"Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2026.acl-long.9","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.acl-long.9","pdf_url":"https://aclanthology.org/2026.acl-long.9.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":"Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"},"sustainable_development_goals":[{"score":0.536120593547821,"display_name":"Quality Education","id":"https://metadata.un.org/sdg/4"}],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7166831519.pdf","grobid_xml":"https://content.openalex.org/works/W7166831519.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Large":[0],"language":[1,137],"models":[2,138],"(LLMs)":[3],"have":[4],"shown":[5],"strong":[6],"performance":[7,145,192],"on":[8,134,148,157],"hard":[9],"reasoning":[10,136,149,191],"and":[11,40,86,129,151,164,182,190,196],"general":[12],"instruction-following":[13],"tasks.However,":[14],"when":[15,59],"sampling":[16],"multiple":[17,66],"outputs":[18,76],"for":[19,193],"the":[20,35,42,69,78,118,131],"same":[21,79],"prompt,":[22],"they":[23],"often":[24],"produce":[25],"highly":[26],"homogeneous,":[27],"repetitive":[28],"responses,":[29,68,115],"resulting":[30],"in":[31,53,187],"inefficient":[32],"exploration.This":[33],"limits":[34],"gains":[36,186],"from":[37],"test-time":[38,165],"scaling":[39],"constrains":[41],"upper":[43],"bound":[44],"of":[45],"reinforcement":[46],"learning":[47],"(RL)":[48],"training.We":[49],"attribute":[50],"this":[51],"issue":[52],"part":[54],"to":[55,73,89,113,124],"supervised":[56],"fine-tuning":[57],"(SFT):":[58],"a":[60,99],"single":[61],"prompt":[62],"is":[63,71],"paired":[64],"with":[65],"reference":[67],"model":[70],"trained":[72],"generate":[74],"diverse":[75,126],"under":[77],"prior":[80,111,127],"condition,":[81],"which":[82],"induces":[83],"optimization":[84],"interference":[85],"can":[87,175],"lead":[88],"diversity":[90,154,163,189],"collapse.To":[91],"address":[92],"this,":[93],"we":[94],"propose":[95],"Prefix-Conditioned":[96],"SFT":[97],"(P-SFT),":[98],"simple":[100],"yet":[101,108],"effective":[102],"method":[103],"that":[104,140],"constructs":[105],"semantically":[106],"consistent":[107],"distributionally":[109],"distinct":[110,121],"contents":[112],"different":[114],"thereby":[116],"projecting":[117],"instruction":[119],"into":[120],"latent":[122],"regions":[123],"establish":[125],"distributions":[128],"decouple":[130],"one-to-many":[132],"mapping.Experiments":[133],"large":[135],"show":[139],"our":[141,172],"approach":[142],"improves":[143],"absolute":[144],"by":[146,155],"5.3%":[147],"benchmarks":[150],"increases":[152],"generation":[153],"198.3%":[156],"average,":[158],"while":[159],"substantially":[160],"enhancing":[161],"output":[162],"scaling.Notably,":[166],"even":[167],"without":[168],"any":[169],"additional":[170],"training,":[171],"prefixing":[173],"strategy":[174],"be":[176],"applied":[177],"at":[178],"inference":[179],"time":[180],"alone":[181],"still":[183],"yields":[184],"significant":[185],"both":[188],"instruction-tuned":[194],"LLMs":[195],"reasoning-enhanced":[197],"models.":[198]},"counts_by_year":[],"updated_date":"2026-07-29T14:22:42.915294","created_date":"2026-07-02T00:00:00"}