{"id":"https://openalex.org/W4412888924","doi":"https://doi.org/10.18653/v1/2025.findings-acl.7","title":"BayesKD: Bayesian Knowledge Distillation for Compact LLMs in Constrained Fine-tuning Scenarios","display_name":"BayesKD: Bayesian Knowledge Distillation for Compact LLMs in Constrained Fine-tuning Scenarios","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4412888924","doi":"https://doi.org/10.18653/v1/2025.findings-acl.7"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2025.findings-acl.7","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.7","pdf_url":"https://aclanthology.org/2025.findings-acl.7.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.7.pdf","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5100318079","display_name":"Wei Li","orcid":"https://orcid.org/0000-0001-9822-6021"},"institutions":[{"id":"https://openalex.org/I79619799","display_name":"University of Birmingham","ror":"https://ror.org/03angcq70","country_code":"GB","type":"education","lineage":["https://openalex.org/I79619799"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Wei Li","raw_affiliation_strings":["University of Birmingham"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Birmingham","institution_ids":["https://openalex.org/I79619799"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5085482371","display_name":"Lujun Li","orcid":"https://orcid.org/0000-0002-4329-2707"},"institutions":[{"id":"https://openalex.org/I200769079","display_name":"Hong Kong University of Science and Technology","ror":"https://ror.org/00q4vv597","country_code":"HK","type":"education","lineage":["https://openalex.org/I200769079"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Lujun Li","raw_affiliation_strings":["HKUST"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"HKUST","institution_ids":["https://openalex.org/I200769079"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100620082","display_name":"Mark Lee","orcid":"https://orcid.org/0000-0003-1262-2045"},"institutions":[{"id":"https://openalex.org/I79619799","display_name":"University of Birmingham","ror":"https://ror.org/03angcq70","country_code":"GB","type":"education","lineage":["https://openalex.org/I79619799"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Mark G. Lee","raw_affiliation_strings":["University of Birmingham"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Birmingham","institution_ids":["https://openalex.org/I79619799"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101427879","display_name":"Shengjie Sun","orcid":"https://orcid.org/0000-0002-8580-4280"},"institutions":[{"id":"https://openalex.org/I136848882","display_name":"Optech (Canada)","ror":"https://ror.org/03njz2k95","country_code":"CA","type":"company","lineage":["https://openalex.org/I136848882"]}],"countries":["CA"],"is_corresponding":false,"raw_author_name":"Shengjie Sun","raw_affiliation_strings":["AISpeech Co. , Ltd"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"AISpeech Co. , Ltd","institution_ids":["https://openalex.org/I136848882"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5023526748","display_name":"Lei Zhang","orcid":"https://orcid.org/0000-0003-2002-8345"},"institutions":[{"id":"https://openalex.org/I23923803","display_name":"University of Exeter","ror":"https://ror.org/03yghzc09","country_code":"GB","type":"education","lineage":["https://openalex.org/I23923803"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Lei Zhang","raw_affiliation_strings":["University of Exeter"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Exeter","institution_ids":["https://openalex.org/I23923803"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100728740","display_name":"Wei Xue","orcid":"https://orcid.org/0000-0002-3797-3928"},"institutions":[{"id":"https://openalex.org/I200769079","display_name":"Hong Kong University of Science and Technology","ror":"https://ror.org/00q4vv597","country_code":"HK","type":"education","lineage":["https://openalex.org/I200769079"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Wei Xue","raw_affiliation_strings":["HKUST"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"HKUST","institution_ids":["https://openalex.org/I200769079"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5051748619","display_name":"Yike Guo","orcid":null},"institutions":[{"id":"https://openalex.org/I200769079","display_name":"Hong Kong University of Science and Technology","ror":"https://ror.org/00q4vv597","country_code":"HK","type":"education","lineage":["https://openalex.org/I200769079"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Yike Guo","raw_affiliation_strings":["HKUST"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"HKUST","institution_ids":["https://openalex.org/I200769079"]}]}],"institutions":[],"countries_distinct_count":3,"institutions_distinct_count":4,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":4.2429,"has_fulltext":true,"cited_by_count":3,"citation_normalized_percentile":{"value":0.94406184,"is_in_top_1_percent":false,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":97,"max":98},"biblio":{"volume":null,"issue":null,"first_page":"138","last_page":"152"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.9567000269889832,"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":0.9567000269889832,"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/T12535","display_name":"Machine Learning and Data Classification","score":0.9451000094413757,"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/T12072","display_name":"Machine Learning and Algorithms","score":0.9277999997138977,"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/distillation","display_name":"Distillation","score":0.6709166169166565},{"id":"https://openalex.org/keywords/bayesian-probability","display_name":"Bayesian probability","score":0.5887528657913208},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.5746316313743591},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.3239281177520752},{"id":"https://openalex.org/keywords/chemistry","display_name":"Chemistry","score":0.10125967860221863},{"id":"https://openalex.org/keywords/chromatography","display_name":"Chromatography","score":0.0767689049243927}],"concepts":[{"id":"https://openalex.org/C204030448","wikidata":"https://www.wikidata.org/wiki/Q101017","display_name":"Distillation","level":2,"score":0.6709166169166565},{"id":"https://openalex.org/C107673813","wikidata":"https://www.wikidata.org/wiki/Q812534","display_name":"Bayesian probability","level":2,"score":0.5887528657913208},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5746316313743591},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3239281177520752},{"id":"https://openalex.org/C185592680","wikidata":"https://www.wikidata.org/wiki/Q2329","display_name":"Chemistry","level":0,"score":0.10125967860221863},{"id":"https://openalex.org/C43617362","wikidata":"https://www.wikidata.org/wiki/Q170050","display_name":"Chromatography","level":1,"score":0.0767689049243927}],"mesh":[],"locations_count":4,"locations":[{"id":"doi:10.18653/v1/2025.findings-acl.7","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.7","pdf_url":"https://aclanthology.org/2025.findings-acl.7.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"},{"id":"pmh:oai:repository.hkust.edu.hk:1783.1-168686","is_oa":false,"landing_page_url":"http://repository.hkust.edu.hk/ir/Record/1783.1-168686","pdf_url":null,"source":{"id":"https://openalex.org/S4306401796","display_name":"Rare & Special e-Zone (The Hong Kong University of Science and Technology)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I200769079","host_organization_name":"Hong Kong University of Science and Technology","host_organization_lineage":["https://openalex.org/I200769079"],"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":"Conference paper"},{"id":"pmh:oai:figshare.com:article/30025237","is_oa":true,"landing_page_url":"https://figshare.com/articles/conference_contribution/BayesKD_Bayesian_Knowledge_Distillation_for_Compact_LLMs_in_Constrained_Fine-tuning_Scenarios/30025237","pdf_url":null,"source":{"id":"https://openalex.org/S4377196282","display_name":"Figshare","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I4210132348","host_organization_name":"Figshare (United Kingdom)","host_organization_lineage":["https://openalex.org/I4210132348"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Conference contribution"},{"id":"pmh:oai:pure.atira.dk:openaire/490dc7ec-cc5a-4d36-9df9-50ae0a183718","is_oa":true,"landing_page_url":"https://research.birmingham.ac.uk/en/publications/490dc7ec-cc5a-4d36-9df9-50ae0a183718","pdf_url":null,"source":{"id":"https://openalex.org/S4306402634","display_name":"University of Birmingham Research Portal (University of Birmingham)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I79619799","host_organization_name":"University of Birmingham","host_organization_lineage":["https://openalex.org/I79619799"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"Li, W, Li, L, Lee, M, Sun, S, Zhang, L, Xue, W & Guo, Y 2025, Bayeskd : Bayesian knowledge distillation for compact llms in constrained fine-tuning scenarios. in W Che, J Nabende, E Shutova & M T Pilehvar (eds), Findings of the Association for Computational Linguistics : ACL 2025. Association for Computational Linguistics, ACL, pp. 138-152, 63rd Annual Meeting of the Association for Computational Linguistics, Vienna, Austria, 27/07/25. < https://aclanthology.org/2025.findings-acl.7/ >","raw_type":"contributionToPeriodical"}],"best_oa_location":{"id":"doi:10.18653/v1/2025.findings-acl.7","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.7","pdf_url":"https://aclanthology.org/2025.findings-acl.7.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":[{"score":0.4699999988079071,"id":"https://metadata.un.org/sdg/2","display_name":"Zero hunger"}],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4412888924.pdf","grobid_xml":"https://content.openalex.org/works/W4412888924.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/W2390279801","https://openalex.org/W4391913857","https://openalex.org/W2358668433","https://openalex.org/W4396701345","https://openalex.org/W2376932109","https://openalex.org/W2001405890","https://openalex.org/W4396696052"],"abstract_inverted_index":{"Large":[0],"language":[1],"models":[2,126],"(LLMs)":[3],"have":[4],"revolutionized":[5],"various":[6,166],"domains":[7],"with":[8],"their":[9,13,129],"remarkable":[10],"capabilities,":[11],"but":[12],"massive":[14],"parameter":[15,191],"sizes":[16],"pose":[17],"significant":[18],"challenges":[19],"for":[20,61,188],"finetuning":[21,66],"and":[22,38,77,95,124,173,183,193],"inference,":[23],"especially":[24],"in":[25,32,64,81,198],"resourceconstrained":[26],"environments.Conventional":[27],"compression":[28],"methods":[29,72,164],"often":[30],"result":[31],"substantial":[33],"performance":[34],"degradation":[35],"within":[36],"LLMs":[37,63],"struggle":[39],"to":[40,79,141],"restore":[41],"model":[42,152],"quality":[43],"during":[44],"fine-tuning.To":[45],"address":[46],"this":[47],"challenge,":[48],"we":[49,136],"present":[50],"Bayesian":[51,96,139],"Knowledge":[52,92,115],"Distillation":[53,97],"(BayesKD),":[54],"a":[55,133],"novel":[56],"distillation":[57,71],"framework":[58],"meticulously":[59],"designed":[60],"compact":[62],"resource-constrained":[65],"scenarios.Departing":[67],"from":[68],"conventional":[69],"LLM":[70,83],"that":[73,159],"introduce":[74],"timeconsuming":[75],"paradigms":[76],"fail":[78],"generalize":[80],"compressed":[82],"fine-tuning":[84],"scenarios,":[85],"our":[86,100,175],"BayesKD":[87,160,176],"develops":[88],"the":[89,106,109,114,119,122,189,199],"Logits":[90,101],"Dual-Scaling,":[91],"Alignment":[93,116],"Module,":[94],"Optimization.In":[98],"particular,":[99],"Dual-Scaling":[102],"strategy":[103],"adaptively":[104],"aligns":[105],"strength":[107],"of":[108,181],"teacher's":[110],"knowledge":[111,130],"transfer,":[112],"while":[113],"Module":[117],"bridges":[118],"gap":[120],"between":[121],"teacher":[123],"student":[125],"by":[127],"projecting":[128],"representations":[131],"into":[132],"shared":[134],"interval.Additionally,":[135],"employ":[137],"Logits-Aware":[138],"Optimization":[140],"swiftly":[142],"identify":[143],"optimal":[144],"settings":[145],"based":[146],"on":[147,165],"these":[148],"strategies,":[149],"thereby":[150],"enhancing":[151],"performance.Extensive":[153],"experiments":[154],"across":[155],"diverse":[156],"tasks":[157],"demonstrate":[158],"consistently":[161],"outperforms":[162],"baseline":[163],"state-of-the-art":[167],"LLMs,":[168],"including":[169],"LLaMA,":[170],"Qwen2,":[171],"Bloom,":[172],"Vicuna.Notably,":[174],"achieves":[177],"average":[178],"accuracy":[179],"gains":[180],"2.99%":[182],"4.05%":[184],"over":[185],"standard":[186],"KD":[187],"8B":[190],"LLaMA":[192],"Qwen2":[194],"model.Codes":[195],"are":[196],"available":[197],"supplementary":[200],"materials.":[201]},"counts_by_year":[{"year":2026,"cited_by_count":3}],"updated_date":"2026-08-25T07:29:55.448023","created_date":"2025-10-10T00:00:00"}