{"id":"https://openalex.org/W4412888880","doi":"https://doi.org/10.18653/v1/2025.findings-acl.29","title":"Variable Layerwise Quantization: A Simple and Effective Approach to Quantize LLMs","display_name":"Variable Layerwise Quantization: A Simple and Effective Approach to Quantize LLMs","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4412888880","doi":"https://doi.org/10.18653/v1/2025.findings-acl.29"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2025.findings-acl.29","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.29","pdf_url":"https://aclanthology.org/2025.findings-acl.29.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.29.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5036739449","display_name":"Razvan-Gabriel Dumitru","orcid":null},"institutions":[{"id":"https://openalex.org/I4210167039","display_name":"ServiceNow (United States)","ror":"https://ror.org/05xr0bc04","country_code":"US","type":"company","lineage":["https://openalex.org/I4210167039"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Razvan-Gabriel Dumitru","raw_affiliation_strings":["ServiceNow Research University of Arizona ServiceNow Research ServiceNow Research Universit di Bologna ServiceNow Research University of Arizona"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"ServiceNow Research University of Arizona ServiceNow Research ServiceNow Research Universit di Bologna ServiceNow Research University of Arizona","institution_ids":["https://openalex.org/I4210167039"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5025247775","display_name":"Vikas Yadav","orcid":"https://orcid.org/0000-0003-0378-6216"},"institutions":[{"id":"https://openalex.org/I4210167039","display_name":"ServiceNow (United States)","ror":"https://ror.org/05xr0bc04","country_code":"US","type":"company","lineage":["https://openalex.org/I4210167039"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Vikas Yadav","raw_affiliation_strings":["ServiceNow Research University of Arizona ServiceNow Research ServiceNow Research Universit di Bologna ServiceNow Research University of Arizona"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"ServiceNow Research University of Arizona ServiceNow Research ServiceNow Research Universit di Bologna ServiceNow Research University of Arizona","institution_ids":["https://openalex.org/I4210167039"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5086568118","display_name":"Rishabh Maheshwary","orcid":null},"institutions":[{"id":"https://openalex.org/I4210167039","display_name":"ServiceNow (United States)","ror":"https://ror.org/05xr0bc04","country_code":"US","type":"company","lineage":["https://openalex.org/I4210167039"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Rishabh Maheshwary","raw_affiliation_strings":["ServiceNow Research University of Arizona ServiceNow Research ServiceNow Research Universit di Bologna ServiceNow Research University of Arizona"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"ServiceNow Research University of Arizona ServiceNow Research ServiceNow Research Universit di Bologna ServiceNow Research University of Arizona","institution_ids":["https://openalex.org/I4210167039"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5114847592","display_name":"Paul Ioan Clotan","orcid":null},"institutions":[{"id":"https://openalex.org/I4210167039","display_name":"ServiceNow (United States)","ror":"https://ror.org/05xr0bc04","country_code":"US","type":"company","lineage":["https://openalex.org/I4210167039"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Paul Ioan Clotan","raw_affiliation_strings":["ServiceNow Research University of Arizona ServiceNow Research ServiceNow Research Universit di Bologna ServiceNow Research University of Arizona"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"ServiceNow Research University of Arizona ServiceNow Research ServiceNow Research Universit di Bologna ServiceNow Research University of Arizona","institution_ids":["https://openalex.org/I4210167039"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5016214507","display_name":"Sathwik Tejaswi Madhusudhan","orcid":null},"institutions":[{"id":"https://openalex.org/I4210167039","display_name":"ServiceNow (United States)","ror":"https://ror.org/05xr0bc04","country_code":"US","type":"company","lineage":["https://openalex.org/I4210167039"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Sathwik Tejaswi Madhusudhan","raw_affiliation_strings":["ServiceNow Research University of Arizona ServiceNow Research ServiceNow Research Universit di Bologna ServiceNow Research University of Arizona"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"ServiceNow Research University of Arizona ServiceNow Research ServiceNow Research Universit di Bologna ServiceNow Research University of Arizona","institution_ids":["https://openalex.org/I4210167039"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5047699502","display_name":"Mihai Surdeanu","orcid":"https://orcid.org/0000-0001-6956-8030"},"institutions":[{"id":"https://openalex.org/I4210167039","display_name":"ServiceNow (United States)","ror":"https://ror.org/05xr0bc04","country_code":"US","type":"company","lineage":["https://openalex.org/I4210167039"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Mihai Surdeanu","raw_affiliation_strings":["ServiceNow Research University of Arizona ServiceNow Research ServiceNow Research Universit di Bologna ServiceNow Research University of Arizona"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"ServiceNow Research University of Arizona ServiceNow Research ServiceNow Research Universit di Bologna ServiceNow Research University of Arizona","institution_ids":["https://openalex.org/I4210167039"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I4210167039"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.29823063,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"534","last_page":"550"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T13114","display_name":"Image Processing Techniques and Applications","score":0.34369999170303345,"subfield":{"id":"https://openalex.org/subfields/2214","display_name":"Media Technology"},"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/T13114","display_name":"Image Processing Techniques and Applications","score":0.34369999170303345,"subfield":{"id":"https://openalex.org/subfields/2214","display_name":"Media Technology"},"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/quantization","display_name":"Quantization (signal processing)","score":0.660767138004303},{"id":"https://openalex.org/keywords/simple","display_name":"Simple (philosophy)","score":0.6303805112838745},{"id":"https://openalex.org/keywords/variable","display_name":"Variable (mathematics)","score":0.6049221158027649},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.4645874798297882},{"id":"https://openalex.org/keywords/mathematics","display_name":"Mathematics","score":0.26943767070770264},{"id":"https://openalex.org/keywords/algorithm","display_name":"Algorithm","score":0.2533605694770813},{"id":"https://openalex.org/keywords/mathematical-analysis","display_name":"Mathematical analysis","score":0.10606560111045837},{"id":"https://openalex.org/keywords/philosophy","display_name":"Philosophy","score":0.06991580128669739},{"id":"https://openalex.org/keywords/epistemology","display_name":"Epistemology","score":0.05883869528770447}],"concepts":[{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.660767138004303},{"id":"https://openalex.org/C2780586882","wikidata":"https://www.wikidata.org/wiki/Q7520643","display_name":"Simple (philosophy)","level":2,"score":0.6303805112838745},{"id":"https://openalex.org/C182365436","wikidata":"https://www.wikidata.org/wiki/Q50701","display_name":"Variable (mathematics)","level":2,"score":0.6049221158027649},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.4645874798297882},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.26943767070770264},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.2533605694770813},{"id":"https://openalex.org/C134306372","wikidata":"https://www.wikidata.org/wiki/Q7754","display_name":"Mathematical analysis","level":1,"score":0.10606560111045837},{"id":"https://openalex.org/C138885662","wikidata":"https://www.wikidata.org/wiki/Q5891","display_name":"Philosophy","level":0,"score":0.06991580128669739},{"id":"https://openalex.org/C111472728","wikidata":"https://www.wikidata.org/wiki/Q9471","display_name":"Epistemology","level":1,"score":0.05883869528770447}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2025.findings-acl.29","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.29","pdf_url":"https://aclanthology.org/2025.findings-acl.29.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.29","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.29","pdf_url":"https://aclanthology.org/2025.findings-acl.29.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":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4412888880.pdf","grobid_xml":"https://content.openalex.org/works/W4412888880.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/W2382521049","https://openalex.org/W2390279801","https://openalex.org/W1585007175","https://openalex.org/W4391913857","https://openalex.org/W2144385241","https://openalex.org/W2358668433","https://openalex.org/W4396701345"],"abstract_inverted_index":{"We":[0],"present":[1,129],"a":[2,12,64,87,123],"simple":[3],"meta":[4],"quantization":[5,27,138,145,160,182,198,213,220],"approach":[6,192],"that":[7,95,105,190],"quantizes":[8],"different":[9,18,69,107],"layers":[10,34,42,55,108,155,234],"of":[11,24,54,63,86,92,154,229],"large":[13],"language":[14],"model":[15,151],"(LLM)":[16],"at":[17,109],"bit":[19,37],"levels,":[20],"and":[21,39,217],"is":[22,79,102,193,215],"independent":[23],"the":[25,31,52,58,61,75,81,84,90,149,227],"underlying":[26],"technique.Specifically,":[28],"we":[29,128],"quantize":[30],"most":[32],"important":[33,41],"to":[35,43,50,148,179,195,203,221,236],"higher":[36],"precision":[38],"less":[40],"lower":[44,159,204,222],"bits.We":[45],"propose":[46],"two":[47],"effective":[48],"strategies":[49],"measure":[51],"importance":[53,62,85,115,178],"within":[56],"LLMs:":[57],"first":[59],"measures":[60],"layer":[65,88,93,177],"based":[66],"on":[67],"how":[68],"its":[70],"output":[71],"embeddings":[72,77],"are":[73,96,156],"from":[74,134],"input":[76],"(higher":[78],"better);":[80],"second":[82],"estimates":[83],"using":[89,161,171],"number":[91],"weights":[94],"much":[97],"larger":[98,230],"than":[99,209],"average":[100],"(smaller":[101],"better).We":[103],"show":[104],"quantizing":[106],"varying":[110],"bits":[111,205,223],"as":[112],"per":[113],"our":[114,135,162,191],"scores":[116],"results":[117],"in":[118,158,226],"minimal":[119],"performance":[120,142],"drop":[121],"with":[122,232,239],"far":[124],"more":[125,233],"compressed":[126],"model.Finally,":[127],"several":[130],"practical":[131],"key":[132],"takeaways":[133],"variable":[136,144],"layer-wise":[137],"experiments:":[139],"(a)":[140],"LLM":[141],"under":[143],"remains":[146],"close":[147],"original":[150],"until":[152,167],"25-50%":[153],"moved":[157,170],"proposed":[163],"ordering":[164],"but":[165],"only":[166],"5-10%":[168],"if":[169],"no":[172],"specific":[173],"ordering;":[174],"(b)":[175],"Adding":[176],"inherently":[180],"dynamic":[181,197],"techniques":[183],"can":[184],"further":[185],"improve":[186],"their":[187],"performance,":[188],"showing":[189],"complementary":[194],"other":[196],"methods;":[199],"(c)":[200],"Quantizing":[201],"LLMs":[202,231,238],"performs":[206],"substantially":[207],"better":[208,225],"pruning":[210],"unless":[211],"extreme":[212],"(2-bit)":[214],"used;":[216],"(d)":[218],"Layer-wise":[219],"works":[224],"case":[228],"compared":[235],"smaller":[237],"fewer":[240],"layers.":[241]},"counts_by_year":[],"updated_date":"2026-07-29T14:22:42.915294","created_date":"2025-10-10T00:00:00"}