{"id":"https://openalex.org/W4415428196","doi":"https://doi.org/10.3233/faia251188","title":"PoT-PTQ:Two-Step Power-of-Two Post-Training for LLMs","display_name":"PoT-PTQ:Two-Step Power-of-Two Post-Training for LLMs","publication_year":2025,"publication_date":"2025-10-21","ids":{"openalex":"https://openalex.org/W4415428196","doi":"https://doi.org/10.3233/faia251188"},"language":null,"primary_location":{"id":"doi:10.3233/faia251188","is_oa":true,"landing_page_url":"https://doi.org/10.3233/faia251188","pdf_url":null,"source":{"id":"https://openalex.org/S4210201731","display_name":"Frontiers in artificial intelligence and applications","issn_l":"0922-6389","issn":["0922-6389","1879-8314"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"journal"},"license":"cc-by-nc","license_id":"https://openalex.org/licenses/cc-by-nc","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Frontiers in Artificial Intelligence and Applications","raw_type":"book-chapter"},"type":"book-chapter","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"hybrid","oa_url":"https://doi.org/10.3233/faia251188","any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5100352792","display_name":"Xinyu Wang","orcid":"https://orcid.org/0000-0002-3676-2408"},"institutions":[{"id":"https://openalex.org/I5023651","display_name":"McGill University","ror":"https://ror.org/01pxwe438","country_code":"CA","type":"education","lineage":["https://openalex.org/I5023651"]}],"countries":["CA"],"is_corresponding":true,"raw_author_name":"Xinyu Wang","raw_affiliation_strings":["McGill University, Canada"],"affiliations":[{"raw_affiliation_string":"McGill University, Canada","institution_ids":["https://openalex.org/I5023651"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5052855957","display_name":"Vahid Partovi Nia","orcid":"https://orcid.org/0000-0001-6673-4224"},"institutions":[{"id":"https://openalex.org/I4210116209","display_name":"The Ark","ror":"https://ror.org/02dnv4714","country_code":"IE","type":"archive","lineage":["https://openalex.org/I4210116209"]},{"id":"https://openalex.org/I4210159102","display_name":"Huawei Technologies (Sweden)","ror":"https://ror.org/0500fyd17","country_code":"SE","type":"company","lineage":["https://openalex.org/I2250955327","https://openalex.org/I4210159102"]}],"countries":["IE","SE"],"is_corresponding":false,"raw_author_name":"Vahid Partovi Nia","raw_affiliation_strings":["Noah\u2019s Ark Lab, Canada"],"affiliations":[{"raw_affiliation_string":"Noah\u2019s Ark Lab, Canada","institution_ids":["https://openalex.org/I4210159102","https://openalex.org/I4210116209"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101593400","display_name":"Peng L\u00fc","orcid":"https://orcid.org/0000-0002-3138-9501"},"institutions":[{"id":"https://openalex.org/I70931966","display_name":"Universit\u00e9 de Montr\u00e9al","ror":"https://ror.org/0161xgx34","country_code":"CA","type":"education","lineage":["https://openalex.org/I70931966"]}],"countries":["CA"],"is_corresponding":false,"raw_author_name":"Peng Lu","raw_affiliation_strings":["Universit\u00e9 de Montr\u00e9al, Canada"],"affiliations":[{"raw_affiliation_string":"Universit\u00e9 de Montr\u00e9al, Canada","institution_ids":["https://openalex.org/I70931966"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100754509","display_name":"Jiayi Huang","orcid":"https://orcid.org/0000-0002-5153-506X"},"institutions":[{"id":"https://openalex.org/I4210164802","display_name":"Mila - Quebec Artificial Intelligence Institute","ror":"https://ror.org/05c22rx21","country_code":"CA","type":"facility","lineage":["https://openalex.org/I4210164802"]},{"id":"https://openalex.org/I70931966","display_name":"Universit\u00e9 de Montr\u00e9al","ror":"https://ror.org/0161xgx34","country_code":"CA","type":"education","lineage":["https://openalex.org/I70931966"]}],"countries":["CA"],"is_corresponding":false,"raw_author_name":"Jerry Huang","raw_affiliation_strings":["Mila \u2013 Quebec AI Institute, Canada","Universit\u00e9 de Montr\u00e9al, Canada"],"affiliations":[{"raw_affiliation_string":"Mila \u2013 Quebec AI Institute, Canada","institution_ids":["https://openalex.org/I4210164802"]},{"raw_affiliation_string":"Universit\u00e9 de Montr\u00e9al, Canada","institution_ids":["https://openalex.org/I70931966"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5062087251","display_name":"Xiao-Wen Chang","orcid":"https://orcid.org/0000-0001-9251-2959"},"institutions":[{"id":"https://openalex.org/I5023651","display_name":"McGill University","ror":"https://ror.org/01pxwe438","country_code":"CA","type":"education","lineage":["https://openalex.org/I5023651"]}],"countries":["CA"],"is_corresponding":false,"raw_author_name":"Xiao-Wen Chang","raw_affiliation_strings":["McGill University, Canada"],"affiliations":[{"raw_affiliation_string":"McGill University, Canada","institution_ids":["https://openalex.org/I5023651"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5019715118","display_name":"Boxing Chen","orcid":"https://orcid.org/0000-0002-3170-4858"},"institutions":[{"id":"https://openalex.org/I4210116209","display_name":"The Ark","ror":"https://ror.org/02dnv4714","country_code":"IE","type":"archive","lineage":["https://openalex.org/I4210116209"]},{"id":"https://openalex.org/I4210159102","display_name":"Huawei Technologies (Sweden)","ror":"https://ror.org/0500fyd17","country_code":"SE","type":"company","lineage":["https://openalex.org/I2250955327","https://openalex.org/I4210159102"]}],"countries":["IE","SE"],"is_corresponding":false,"raw_author_name":"Boxing Chen","raw_affiliation_strings":["Noah\u2019s Ark Lab, Canada"],"affiliations":[{"raw_affiliation_string":"Noah\u2019s Ark Lab, Canada","institution_ids":["https://openalex.org/I4210159102","https://openalex.org/I4210116209"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5020617875","display_name":"Yugui Cui","orcid":null},"institutions":[{"id":"https://openalex.org/I4210116209","display_name":"The Ark","ror":"https://ror.org/02dnv4714","country_code":"IE","type":"archive","lineage":["https://openalex.org/I4210116209"]},{"id":"https://openalex.org/I4210159102","display_name":"Huawei Technologies (Sweden)","ror":"https://ror.org/0500fyd17","country_code":"SE","type":"company","lineage":["https://openalex.org/I2250955327","https://openalex.org/I4210159102"]}],"countries":["IE","SE"],"is_corresponding":false,"raw_author_name":"Yufei Cui","raw_affiliation_strings":["Noah\u2019s Ark Lab, Canada"],"affiliations":[{"raw_affiliation_string":"Noah\u2019s Ark Lab, Canada","institution_ids":["https://openalex.org/I4210159102","https://openalex.org/I4210116209"]}]}],"institutions":[],"countries_distinct_count":3,"institutions_distinct_count":7,"corresponding_author_ids":["https://openalex.org/A5100352792"],"corresponding_institution_ids":["https://openalex.org/I5023651"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.50500925,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.6908000111579895,"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":0.6908000111579895,"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/T11201","display_name":"Metallurgy and Material Forming","score":0.602400004863739,"subfield":{"id":"https://openalex.org/subfields/2211","display_name":"Mechanics of Materials"},"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":0.8901000022888184},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.644599974155426},{"id":"https://openalex.org/keywords/software-deployment","display_name":"Software deployment","score":0.4749000072479248},{"id":"https://openalex.org/keywords/integer","display_name":"Integer (computer science)","score":0.47269999980926514},{"id":"https://openalex.org/keywords/floating-point","display_name":"Floating point","score":0.3880999982357025},{"id":"https://openalex.org/keywords/natural-language","display_name":"Natural language","score":0.32519999146461487},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.32429999113082886}],"concepts":[{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.8901000022888184},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.644599974155426},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5554999709129333},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.5185999870300293},{"id":"https://openalex.org/C105339364","wikidata":"https://www.wikidata.org/wiki/Q2297740","display_name":"Software deployment","level":2,"score":0.4749000072479248},{"id":"https://openalex.org/C97137487","wikidata":"https://www.wikidata.org/wiki/Q729138","display_name":"Integer (computer science)","level":2,"score":0.47269999980926514},{"id":"https://openalex.org/C84211073","wikidata":"https://www.wikidata.org/wiki/Q117879","display_name":"Floating point","level":2,"score":0.3880999982357025},{"id":"https://openalex.org/C195324797","wikidata":"https://www.wikidata.org/wiki/Q33742","display_name":"Natural language","level":2,"score":0.32519999146461487},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.32429999113082886},{"id":"https://openalex.org/C121040770","wikidata":"https://www.wikidata.org/wiki/Q215675","display_name":"Quantum entanglement","level":3,"score":0.3066999912261963},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.30079999566078186},{"id":"https://openalex.org/C139676723","wikidata":"https://www.wikidata.org/wiki/Q1193832","display_name":"Sign (mathematics)","level":2,"score":0.2978000044822693},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.29670000076293945},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.2903999984264374},{"id":"https://openalex.org/C2987812609","wikidata":"https://www.wikidata.org/wiki/Q194158","display_name":"Bit rate","level":2,"score":0.28450000286102295},{"id":"https://openalex.org/C28719098","wikidata":"https://www.wikidata.org/wiki/Q44946","display_name":"Point (geometry)","level":2,"score":0.2768999934196472},{"id":"https://openalex.org/C165838908","wikidata":"https://www.wikidata.org/wiki/Q736777","display_name":"Calibration","level":2,"score":0.2720000147819519},{"id":"https://openalex.org/C94375191","wikidata":"https://www.wikidata.org/wiki/Q11205","display_name":"Arithmetic","level":1,"score":0.2653000056743622},{"id":"https://openalex.org/C41426520","wikidata":"https://www.wikidata.org/wiki/Q1192065","display_name":"Point estimation","level":2,"score":0.26510000228881836}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.3233/faia251188","is_oa":true,"landing_page_url":"https://doi.org/10.3233/faia251188","pdf_url":null,"source":{"id":"https://openalex.org/S4210201731","display_name":"Frontiers in artificial intelligence and applications","issn_l":"0922-6389","issn":["0922-6389","1879-8314"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"journal"},"license":"cc-by-nc","license_id":"https://openalex.org/licenses/cc-by-nc","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Frontiers in Artificial Intelligence and Applications","raw_type":"book-chapter"}],"best_oa_location":{"id":"doi:10.3233/faia251188","is_oa":true,"landing_page_url":"https://doi.org/10.3233/faia251188","pdf_url":null,"source":{"id":"https://openalex.org/S4210201731","display_name":"Frontiers in artificial intelligence and applications","issn_l":"0922-6389","issn":["0922-6389","1879-8314"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"journal"},"license":"cc-by-nc","license_id":"https://openalex.org/licenses/cc-by-nc","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Frontiers in Artificial Intelligence and Applications","raw_type":"book-chapter"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Large":[0],"Language":[1],"Models":[2],"(LLMs)":[3],"have":[4],"demonstrated":[5],"remarkable":[6],"performance":[7,138],"across":[8],"various":[9],"natural":[10],"language":[11],"processing":[12],"(NLP)":[13],"tasks.":[14],"However,":[15],"their":[16],"deployment":[17],"is":[18,30,61],"challenging":[19],"due":[20],"to":[21,34,176,192],"the":[22,64,105,108,119,145,165,170],"substantial":[23],"computational":[24],"resources":[25],"required.":[26],"Power-of-two":[27],"(PoT)":[28],"quantization":[29,43,79,120,163],"a":[31,76,113,123,133,181,187],"general":[32],"tool":[33],"counteract":[35],"this":[36],"difficulty.":[37],"Albeit":[38],"previous":[39],"works":[40],"on":[41,48,57,180,186],"PoT":[42,141,162],"can":[44],"be":[45],"efficiently":[46],"dequantized":[47],"CPUs":[49],"using":[50,132],"fixed-point":[51],"addition,":[52],"it":[53],"showed":[54],"less":[55],"effectiveness":[56],"GPUs.":[58],"The":[59,137],"reason":[60],"entanglement":[62],"of":[63,107,139],"sign":[65],"bit":[66,69],"and":[67,94,127,158,174,184],"sequential":[68],"manipulations":[70],"needed":[71],"for":[72,81,169],"dequantization.":[73,102,195],"We":[74],"propose":[75],"novel":[77],"POT":[78],"framework":[80],"LLM":[82],"weights":[83],"that":[84],"(i)":[85,117],"outperforms":[86],"state-of-the-art":[87,147],"accuracy":[88,106],"in":[89,148],"extremely":[90],"low-precision":[91],"number":[92],"formats,":[93],"(ii)":[95,128],"enables":[96],"faster":[97],"inference":[98,173],"through":[99],"more":[100],"efficient":[101],"To":[103],"maintain":[104],"quantized":[109],"model,":[110],"we":[111],"introduce":[112],"two-step":[114],"post-training":[115,142],"algorithm:":[116],"initialize":[118],"scales":[121,131],"with":[122],"robust":[124],"starting":[125],"point,":[126],"refine":[129],"these":[130],"minimal":[134],"calibration":[135],"set.":[136],"our":[140],"algorithm":[143],"surpasses":[144],"current":[146],"integer":[149,194],"quantization,":[150],"particularly":[151],"at":[152],"low":[153],"precisions":[154],"such":[155],"as":[156],"2-":[157],"3-bit":[159],"formats.":[160],"Our":[161],"accelerates":[164],"dequantization":[166],"step":[167],"required":[168],"floating":[171],"point":[172],"leads":[175],"3.67\u00d7":[177],"speed":[178],"up":[179],"NVIDIA":[182,188],"V100,":[183],"1.63\u00d7":[185],"RTX":[189],"4090,":[190],"compared":[191],"uniform":[193]},"counts_by_year":[],"updated_date":"2026-03-07T16:01:11.037858","created_date":"2025-10-24T00:00:00"}