{"id":"https://openalex.org/W4408062366","doi":"https://doi.org/10.5220/0013159100003890","title":"A Mixed Quantization Approach for Data-Free Quantization of LLMs","display_name":"A Mixed Quantization Approach for Data-Free Quantization of LLMs","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4408062366","doi":"https://doi.org/10.5220/0013159100003890"},"language":"en","primary_location":{"id":"doi:10.5220/0013159100003890","is_oa":true,"landing_page_url":"https://doi.org/10.5220/0013159100003890","pdf_url":null,"source":null,"license":"cc-by-nc-nd","license_id":"https://openalex.org/licenses/cc-by-nc-nd","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 17th International Conference on Agents and Artificial Intelligence","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://doi.org/10.5220/0013159100003890","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5102430197","display_name":"Feng Zhang","orcid":"https://orcid.org/0000-0003-2586-5505"},"institutions":[{"id":"https://openalex.org/I39854758","display_name":"Auckland University of Technology","ror":"https://ror.org/01zvqw119","country_code":"NZ","type":"education","lineage":["https://openalex.org/I39854758"]}],"countries":["NZ"],"is_corresponding":false,"raw_author_name":"Feng Zhang","raw_affiliation_strings":["Auckland University of Technology, Auckland, New Zealand"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Auckland University of Technology, Auckland, New Zealand","institution_ids":["https://openalex.org/I39854758"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100630042","display_name":"Yanbin Liu","orcid":"https://orcid.org/0000-0003-4724-8065"},"institutions":[{"id":"https://openalex.org/I39854758","display_name":"Auckland University of Technology","ror":"https://ror.org/01zvqw119","country_code":"NZ","type":"education","lineage":["https://openalex.org/I39854758"]}],"countries":["NZ"],"is_corresponding":false,"raw_author_name":"Yanbin Liu","raw_affiliation_strings":["Auckland University of Technology, Auckland, New Zealand"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Auckland University of Technology, Auckland, New Zealand","institution_ids":["https://openalex.org/I39854758"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5020475433","display_name":"Weihua Li","orcid":"https://orcid.org/0000-0001-7679-8360"},"institutions":[{"id":"https://openalex.org/I39854758","display_name":"Auckland University of Technology","ror":"https://ror.org/01zvqw119","country_code":"NZ","type":"education","lineage":["https://openalex.org/I39854758"]}],"countries":["NZ"],"is_corresponding":false,"raw_author_name":"Weihua Li","raw_affiliation_strings":["Auckland University of Technology, Auckland, New Zealand"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Auckland University of Technology, Auckland, New Zealand","institution_ids":["https://openalex.org/I39854758"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100395117","display_name":"Xiaodan Wang","orcid":"https://orcid.org/0000-0003-0216-5135"},"institutions":[{"id":"https://openalex.org/I1793135","display_name":"Yanbian University","ror":"https://ror.org/039xnh269","country_code":"CN","type":"education","lineage":["https://openalex.org/I1793135"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Xiaodan Wang","raw_affiliation_strings":["Yanbian University, Jilin, China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Yanbian University, Jilin, China","institution_ids":["https://openalex.org/I1793135"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5002048333","display_name":"Quan Bai","orcid":null},"institutions":[{"id":"https://openalex.org/I129801699","display_name":"University of Tasmania","ror":"https://ror.org/01nfmeh72","country_code":"AU","type":"education","lineage":["https://openalex.org/I129801699"]}],"countries":["AU"],"is_corresponding":false,"raw_author_name":"Quan Bai","raw_affiliation_strings":["University of Tasmania, Hobart, Australia"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Tasmania, Hobart, Australia","institution_ids":["https://openalex.org/I129801699"]}]}],"institutions":[],"countries_distinct_count":3,"institutions_distinct_count":3,"corresponding_author_ids":[],"corresponding_institution_ids":[],"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":"353","last_page":"363"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10688","display_name":"Image and Signal Denoising Methods","score":0.8374999761581421,"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"}},"topics":[{"id":"https://openalex.org/T10688","display_name":"Image and Signal Denoising Methods","score":0.8374999761581421,"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"}},{"id":"https://openalex.org/T10901","display_name":"Advanced Data Compression Techniques","score":0.8174999952316284,"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"}},{"id":"https://openalex.org/T10522","display_name":"Medical Imaging Techniques and Applications","score":0.804099977016449,"subfield":{"id":"https://openalex.org/subfields/2741","display_name":"Radiology, Nuclear Medicine and Imaging"},"field":{"id":"https://openalex.org/fields/27","display_name":"Medicine"},"domain":{"id":"https://openalex.org/domains/4","display_name":"Health Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/quantization","display_name":"Quantization (signal processing)","score":0.8295514583587646},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.48026561737060547},{"id":"https://openalex.org/keywords/vector-quantization","display_name":"Vector quantization","score":0.4650465250015259},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.1978815793991089},{"id":"https://openalex.org/keywords/algorithm","display_name":"Algorithm","score":0.14912700653076172}],"concepts":[{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.8295514583587646},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.48026561737060547},{"id":"https://openalex.org/C199833920","wikidata":"https://www.wikidata.org/wiki/Q612536","display_name":"Vector quantization","level":2,"score":0.4650465250015259},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.1978815793991089},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.14912700653076172}],"mesh":[],"locations_count":2,"locations":[{"id":"doi:10.5220/0013159100003890","is_oa":true,"landing_page_url":"https://doi.org/10.5220/0013159100003890","pdf_url":null,"source":null,"license":"cc-by-nc-nd","license_id":"https://openalex.org/licenses/cc-by-nc-nd","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 17th International Conference on Agents and Artificial Intelligence","raw_type":"proceedings-article"},{"id":"pmh:oai:openrepository.aut.ac.nz:10292/18885","is_oa":true,"landing_page_url":"http://hdl.handle.net/10292/18885","pdf_url":null,"source":{"id":"https://openalex.org/S4306401809","display_name":"Tuwhera (Auckland University of Technology)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I39854758","host_organization_name":"Auckland University of Technology","host_organization_lineage":["https://openalex.org/I39854758"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by-nc-nd","license_id":"https://openalex.org/licenses/cc-by-nc-nd","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Conference Contribution"}],"best_oa_location":{"id":"doi:10.5220/0013159100003890","is_oa":true,"landing_page_url":"https://doi.org/10.5220/0013159100003890","pdf_url":null,"source":null,"license":"cc-by-nc-nd","license_id":"https://openalex.org/licenses/cc-by-nc-nd","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 17th International Conference on Agents and Artificial Intelligence","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"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/W3209251257"],"abstract_inverted_index":{"Large":[0],"Language":[1],"Models":[2],"(LLMs)":[3],"have":[4,47],"demonstrated":[5],"significant":[6,34],"capabilities":[7],"in":[8,114],"intelligent":[9],"activities":[10],"such":[11],"as":[12],"natural":[13],"language":[14],"comprehension,":[15],"content":[16],"generation,":[17],"and":[18,23,40,78,195],"knowledge":[19],"retrieval.":[20],"However,":[21],"training":[22],"deploying":[24],"these":[25],"models":[26],"require":[27],"substantial":[28],"computation":[29],"resources,":[30],"setting":[31],"up":[32],"a":[33,94,125,158,189],"barrier":[35],"for":[36,93,132],"developing":[37],"AI":[38],"applications":[39],"conducting":[41],"research.":[42],"Various":[43],"model":[44],"compression":[45],"techniques":[46],"been":[48,60],"developed":[49],"to":[50,67,88,111,117,130,157],"address":[51],"the":[52,73,102,133,139,146,170],"demanding":[53],"computational":[54],"resource":[55],"issue.":[56],"Nonetheless,":[57],"there":[58],"has":[59],"limited":[61],"exploration":[62],"into":[63,145],"high-level":[64],"quantization":[65,118,127,135,141,193],"strategy":[66],"offer":[68],"better":[69,95],"flexibility":[70],"of":[71,105,192],"balancing":[72],"trade-off":[74],"between":[75],"memory":[76,142,182],"usage":[77,183],"accuracy.":[79],"We":[80],"propose":[81],"an":[82],"effective":[83],"mixed-quantization":[84],"method":[85,167],"named":[86],"MXQ":[87],"bridge":[89],"this":[90],"research":[91],"gap":[92],"memory-accuracy":[96,197],"balance.":[97],"Specifically,":[98],"we":[99,123],"observe":[100],"that":[101,165],"weight":[103],"distributions":[104],"LLMs":[106],"vary":[107],"considerably":[108],"from":[109],"layer":[110],"layer,":[112],"resulting":[113],"different":[115],"tolerances":[116],"errors.":[119],"Motivated":[120],"by":[121,155],"this,":[122],"derive":[124],"novel":[126],"optimisation":[128],"formulation":[129,150],"solve":[131],"layer-wise":[134],"parameters,":[136],"while":[137],"enforcing":[138],"overall":[140],"consumption":[143],"budget":[144,178],"constraints.":[147],"The":[148],"new":[149],"can":[151,168],"be":[152],"efficiently":[153],"solved":[154],"converting":[156],"mixed":[159],"integer":[160],"programming":[161],"problem.":[162],"Experiments":[163],"shows":[164],"our":[166],"achieve":[169],"1%":[171],"accuracy":[172],"loss":[173],"goal":[174],"with":[175],"additional":[176],"bit":[177],"or":[179],"further":[180],"reduce":[181],"on":[184],"Llama":[185],"models.":[186],"This":[187],"unlocks":[188],"wide":[190],"range":[191],"options":[194],"simplifies":[196],"trade-off.":[198]},"counts_by_year":[],"updated_date":"2026-07-14T23:27:15.235271","created_date":"2025-10-10T00:00:00"}