{"id":"https://openalex.org/W4412887850","doi":"https://doi.org/10.18653/v1/2025.findings-acl.996","title":"Outlier-weighed Layerwise Sampling for LLM Fine-tuning","display_name":"Outlier-weighed Layerwise Sampling for LLM Fine-tuning","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4412887850","doi":"https://doi.org/10.18653/v1/2025.findings-acl.996"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2025.findings-acl.996","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.996","pdf_url":"https://aclanthology.org/2025.findings-acl.996.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.996.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5101469911","display_name":"Pengxiang Li","orcid":"https://orcid.org/0000-0003-4414-6186"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Pengxiang Li","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5039943423","display_name":"Yin Lu","orcid":"https://orcid.org/0000-0002-9568-7428"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lu Yin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5089605068","display_name":"Xiaowei Gao","orcid":"https://orcid.org/0000-0003-3273-7499"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xiaowei Gao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5100415814","display_name":"Shiwei Liu","orcid":"https://orcid.org/0000-0003-0976-7163"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shiwei Liu","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":13.2299,"has_fulltext":true,"cited_by_count":3,"citation_normalized_percentile":{"value":0.98954857,"is_in_top_1_percent":false,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":94,"max":97},"biblio":{"volume":null,"issue":null,"first_page":"19460","last_page":"19473"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T12169","display_name":"Non-Destructive Testing Techniques","score":0.9896000027656555,"subfield":{"id":"https://openalex.org/subfields/2210","display_name":"Mechanical Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T12169","display_name":"Non-Destructive Testing Techniques","score":0.9896000027656555,"subfield":{"id":"https://openalex.org/subfields/2210","display_name":"Mechanical Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T11338","display_name":"Advancements in Photolithography Techniques","score":0.984000027179718,"subfield":{"id":"https://openalex.org/subfields/2208","display_name":"Electrical and Electronic Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T11607","display_name":"Microwave and Dielectric Measurement Techniques","score":0.9585000276565552,"subfield":{"id":"https://openalex.org/subfields/2208","display_name":"Electrical and Electronic Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/outlier","display_name":"Outlier","score":0.7260879278182983},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.6068345308303833},{"id":"https://openalex.org/keywords/sampling","display_name":"Sampling (signal processing)","score":0.5130389928817749},{"id":"https://openalex.org/keywords/anomaly-detection","display_name":"Anomaly detection","score":0.4516368508338928},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.4124772250652313},{"id":"https://openalex.org/keywords/computer-vision","display_name":"Computer vision","score":0.20531520247459412}],"concepts":[{"id":"https://openalex.org/C79337645","wikidata":"https://www.wikidata.org/wiki/Q779824","display_name":"Outlier","level":2,"score":0.7260879278182983},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6068345308303833},{"id":"https://openalex.org/C140779682","wikidata":"https://www.wikidata.org/wiki/Q210868","display_name":"Sampling (signal processing)","level":3,"score":0.5130389928817749},{"id":"https://openalex.org/C739882","wikidata":"https://www.wikidata.org/wiki/Q3560506","display_name":"Anomaly detection","level":2,"score":0.4516368508338928},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4124772250652313},{"id":"https://openalex.org/C31972630","wikidata":"https://www.wikidata.org/wiki/Q844240","display_name":"Computer vision","level":1,"score":0.20531520247459412},{"id":"https://openalex.org/C106131492","wikidata":"https://www.wikidata.org/wiki/Q3072260","display_name":"Filter (signal processing)","level":2,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2025.findings-acl.996","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.996","pdf_url":"https://aclanthology.org/2025.findings-acl.996.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.996","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.996","pdf_url":"https://aclanthology.org/2025.findings-acl.996.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":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4412887850.pdf","grobid_xml":"https://content.openalex.org/works/W4412887850.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/W2499612753","https://openalex.org/W3111802945","https://openalex.org/W2946096271","https://openalex.org/W2295423552","https://openalex.org/W1598471830","https://openalex.org/W3107369729"],"abstract_inverted_index":{"The":[0],"rapid":[1],"advancements":[2],"in":[3,23,107],"Large":[4],"Language":[5],"Models":[6],"(LLMs)":[7],"have":[8,34],"revolutionized":[9],"various":[10,123],"natural":[11],"language":[12],"processing":[13],"tasks.However,":[14],"the":[15,60,98,117,149],"substantial":[16],"size":[17],"of":[18,64,100,179],"LLMs":[19,175],"presents":[20],"significant":[21],"challenges":[22],"training":[24],"or":[25],"fine-tuning.While":[26],"parameter-efficient":[27],"approaches":[28],"such":[29],"as":[30],"low-rank":[31,113],"adaptation":[32],"(LoRA)":[33],"gained":[35],"popularity,":[36],"they":[37],"often":[38],"compromise":[39],"performance":[40],"compared":[41],"to":[42,71,80,142,172],"full-rank":[43],"fine-tuning.In":[44],"this":[45],"paper,":[46],"we":[47,110],"propose":[48],"Outlier-weighed":[49],"Layerwise":[50],"Sampling":[51],"(OWS),":[52],"a":[53,88,104,143,153,159],"new":[54],"memory-efficient":[55],"finetuning":[56],"approach,":[57],"inspired":[58],"by":[59],"layerwise":[61],"outlier":[62],"distribution":[63],"LLMs.Unlike":[65],"LoRA,":[66],"which":[67],"adds":[68],"extra":[69],"adapters":[70],"all":[72],"layers,":[73],"OWS":[74,131],"strategically":[75],"assigns":[76],"higher":[77],"sampling":[78,86],"probabilities":[79],"layers":[81,90,102],"with":[82,176],"more":[83,167],"outliers,":[84],"selectively":[85],"only":[87,177],"few":[89],"and":[91,127,158],"fine-tuning":[92],"their":[93],"pre-trained":[94],"weights.To":[95],"further":[96,115],"increase":[97],"number":[99],"fine-tuned":[101],"without":[103],"proportional":[105],"rise":[106],"memory":[108,168],"costs,":[109],"incorporate":[111],"gradient":[112],"projection,":[114],"boosting":[116],"approach's":[118],"performance.Our":[119],"extensive":[120],"experiments":[121],"across":[122],"architectures,":[124],"including":[125,136],"LLaMa2,":[126],"Mistral,":[128],"demonstrate":[129],"that":[130],"consistently":[132],"outperforms":[133],"baseline":[134],"approaches,":[135],"full":[137],"fine-tuning.Specifically,":[138],"it":[139],"achieves":[140],"up":[141],"1.1%":[144],"average":[145],"accuracy":[146],"gain":[147],"on":[148,156,163],"Commonsense":[150],"Reasoning":[151],"benchmark,":[152],"3.0%":[154],"improvement":[155],"MMLU,":[157],"notable":[160],"10%":[161],"boost":[162],"MT-Bench,":[164],"while":[165],"being":[166],"efficient.OWS":[169],"allows":[170],"us":[171],"fine-tune":[173],"7B":[174],"21GB":[178],"memory.Our":[180],"code":[181],"is":[182],"available":[183],"at":[184],"https:":[185],"//github.com/pixeli99/OWS.":[186]},"counts_by_year":[{"year":2026,"cited_by_count":1},{"year":2025,"cited_by_count":2}],"updated_date":"2026-07-29T14:22:42.915294","created_date":"2025-10-10T00:00:00"}