{"id":"https://openalex.org/W7214031986","doi":"https://doi.org/10.48550/arxiv.2609.24519","title":"AWE: Adaptive Weight Encoding for Exact Integer Matrix Products with Fewer GEMMs on FP4 Tensor Cores","display_name":"AWE: Adaptive Weight Encoding for Exact Integer Matrix Products with Fewer GEMMs on FP4 Tensor Cores","publication_year":2026,"publication_date":"2026-09-21","ids":{"openalex":"https://openalex.org/W7214031986","doi":"https://doi.org/10.48550/arxiv.2609.24519"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2609.24519","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.24519","pdf_url":"https://arxiv.org/pdf/2609.24519","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"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":"text"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2609.24519","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5080806431","display_name":"Shunichiro Hayashi","orcid":"https://orcid.org/0009-0000-8052-782X","observed_orcids":["https://orcid.org/0009-0000-8052-782X"]},"institutions":[{"id":"https://openalex.org/I60134161","display_name":"Nagoya University","ror":"https://ror.org/04chrp450","country_code":"JP","type":"education","lineage":["https://openalex.org/I60134161"]}],"countries":["JP"],"is_corresponding":false,"raw_author_name":"Hayashi, Shun-ichiro","raw_affiliation_strings":["Graduate School of Informatics Nagoya University Aichi , Japan"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Graduate School of Informatics Nagoya University Aichi , Japan","institution_ids":["https://openalex.org/I60134161"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5035972892","display_name":"Daichi Mukunoki","orcid":"https://orcid.org/0000-0002-0051-6811","observed_orcids":["https://orcid.org/0000-0002-0051-6811"]},"institutions":[{"id":"https://openalex.org/I60134161","display_name":"Nagoya University","ror":"https://ror.org/04chrp450","country_code":"JP","type":"education","lineage":["https://openalex.org/I60134161"]}],"countries":["JP"],"is_corresponding":false,"raw_author_name":"Mukunoki, Daichi","raw_affiliation_strings":["Information Technology Center Nagoya University Aichi , Japan"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Information Technology Center Nagoya University Aichi , Japan","institution_ids":["https://openalex.org/I60134161"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5149633453","display_name":"Tetsuya Hoshino","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I60134161","display_name":"Nagoya University","ror":"https://ror.org/04chrp450","country_code":"JP","type":"education","lineage":["https://openalex.org/I60134161"]}],"countries":["JP"],"is_corresponding":false,"raw_author_name":"Hoshino, Tetsuya","raw_affiliation_strings":["Information Technology Center Nagoya University Aichi , Japan"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Information Technology Center Nagoya University Aichi , Japan","institution_ids":["https://openalex.org/I60134161"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5152452960","display_name":"Takahiro Katagiri","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I60134161","display_name":"Nagoya University","ror":"https://ror.org/04chrp450","country_code":"JP","type":"education","lineage":["https://openalex.org/I60134161"]}],"countries":["JP"],"is_corresponding":false,"raw_author_name":"Katagiri, Takahiro","raw_affiliation_strings":["Information Technology Center Nagoya University Aichi , Japan"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Information Technology Center Nagoya University Aichi , Japan","institution_ids":["https://openalex.org/I60134161"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I60134161"],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":1,"citation_normalized_percentile":null,"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/T11697","display_name":"Numerical Methods and Algorithms","score":0.9922000169754028,"subfield":{"id":"https://openalex.org/subfields/1703","display_name":"Computational Theory and Mathematics"},"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/T11697","display_name":"Numerical Methods and Algorithms","score":0.9922000169754028,"subfield":{"id":"https://openalex.org/subfields/1703","display_name":"Computational Theory and Mathematics"},"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/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.003100000089034438,"subfield":{"id":"https://openalex.org/subfields/1708","display_name":"Hardware and Architecture"},"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/T11693","display_name":"Cryptography and Residue Arithmetic","score":0.002300000051036477,"subfield":{"id":"https://openalex.org/subfields/1710","display_name":"Information Systems"},"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/residue-number-system","display_name":"residue number system","score":0.6830000281333923}],"concepts":[{"id":"https://openalex.org/C54732982","wikidata":"https://www.wikidata.org/wiki/Q1415345","display_name":"Modulo","level":2,"score":0.6858999729156494},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.682699978351593},{"id":"https://openalex.org/C97137487","wikidata":"https://www.wikidata.org/wiki/Q729138","display_name":"Integer (computer science)","level":2,"score":0.6736999750137329},{"id":"https://openalex.org/C106487976","wikidata":"https://www.wikidata.org/wiki/Q685816","display_name":"Matrix (chemical analysis)","level":2,"score":0.5101000070571899},{"id":"https://openalex.org/C17349429","wikidata":"https://www.wikidata.org/wiki/Q1049914","display_name":"Matrix multiplication","level":3,"score":0.47780001163482666},{"id":"https://openalex.org/C55526617","wikidata":"https://www.wikidata.org/wiki/Q719375","display_name":"Operand","level":2,"score":0.476500004529953},{"id":"https://openalex.org/C114614502","wikidata":"https://www.wikidata.org/wiki/Q76592","display_name":"Combinatorics","level":1,"score":0.47189998626708984},{"id":"https://openalex.org/C201958917","wikidata":"https://www.wikidata.org/wiki/Q1367055","display_name":"Unimodular matrix","level":2,"score":0.4571000039577484},{"id":"https://openalex.org/C51255310","wikidata":"https://www.wikidata.org/wiki/Q1163016","display_name":"Tensor product","level":2,"score":0.4447999894618988},{"id":"https://openalex.org/C90673727","wikidata":"https://www.wikidata.org/wiki/Q901718","display_name":"Product (mathematics)","level":2,"score":0.42329999804496765},{"id":"https://openalex.org/C118615104","wikidata":"https://www.wikidata.org/wiki/Q121416","display_name":"Discrete mathematics","level":1,"score":0.4196000099182129},{"id":"https://openalex.org/C65236422","wikidata":"https://www.wikidata.org/wiki/Q173740","display_name":"Cartesian product","level":2,"score":0.3950999975204468},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.3944999873638153},{"id":"https://openalex.org/C62555958","wikidata":"https://www.wikidata.org/wiki/Q6042601","display_name":"Integer matrix","level":5,"score":0.3837999999523163},{"id":"https://openalex.org/C27753989","wikidata":"https://www.wikidata.org/wiki/Q284885","display_name":"Superposition principle","level":2,"score":0.3675999939441681},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.35179999470710754},{"id":"https://openalex.org/C46030957","wikidata":"https://www.wikidata.org/wiki/Q1238125","display_name":"Kronecker product","level":3,"score":0.34850001335144043},{"id":"https://openalex.org/C149810388","wikidata":"https://www.wikidata.org/wiki/Q5374873","display_name":"Emulation","level":2,"score":0.3352999985218048},{"id":"https://openalex.org/C132074034","wikidata":"https://www.wikidata.org/wiki/Q154210","display_name":"Congruence (geometry)","level":2,"score":0.3149999976158142},{"id":"https://openalex.org/C56086750","wikidata":"https://www.wikidata.org/wiki/Q6042592","display_name":"Integer programming","level":2,"score":0.30970001220703125},{"id":"https://openalex.org/C94020503","wikidata":"https://www.wikidata.org/wiki/Q12916","display_name":"Real number","level":2,"score":0.3034000098705292},{"id":"https://openalex.org/C94375191","wikidata":"https://www.wikidata.org/wiki/Q11205","display_name":"Arithmetic","level":1,"score":0.2973000109195709},{"id":"https://openalex.org/C164680029","wikidata":"https://www.wikidata.org/wiki/Q1055293","display_name":"Multiple","level":2,"score":0.2808000147342682},{"id":"https://openalex.org/C2778755073","wikidata":"https://www.wikidata.org/wiki/Q10858537","display_name":"Scale (ratio)","level":2,"score":0.2802000045776367},{"id":"https://openalex.org/C23230895","wikidata":"https://www.wikidata.org/wiki/Q104752","display_name":"Coprime integers","level":2,"score":0.27880001068115234},{"id":"https://openalex.org/C2778112365","wikidata":"https://www.wikidata.org/wiki/Q3511065","display_name":"Sequence (biology)","level":2,"score":0.26930001378059387},{"id":"https://openalex.org/C199163554","wikidata":"https://www.wikidata.org/wiki/Q1681619","display_name":"Univariate","level":3,"score":0.26739999651908875},{"id":"https://openalex.org/C203432157","wikidata":"https://www.wikidata.org/wiki/Q2498974","display_name":"Radical of an integer","level":4,"score":0.25440001487731934},{"id":"https://openalex.org/C77618280","wikidata":"https://www.wikidata.org/wiki/Q1155772","display_name":"Scheme (mathematics)","level":2,"score":0.2533999979496002},{"id":"https://openalex.org/C2781045450","wikidata":"https://www.wikidata.org/wiki/Q254569","display_name":"Backdoor","level":2,"score":0.2513999938964844}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2609.24519","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.24519","pdf_url":"https://arxiv.org/pdf/2609.24519","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"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":"text"},{"id":"doi:10.48550/arxiv.2609.24519","doi":"https://doi.org/10.48550/arxiv.2609.24519","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2609.24519","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2609.24519","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.24519","pdf_url":"https://arxiv.org/pdf/2609.24519","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"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":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[{"id":"https://openalex.org/F4320334764","display_name":"Japan Society for the Promotion of Science","ror":"https://ror.org/00hhkn466"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7214031986.pdf","grobid_xml":"https://content.openalex.org/works/W7214031986.grobid-xml"},"referenced_works_count":6,"referenced_works":["https://openalex.org/W4392271821","https://openalex.org/W4414786777","https://openalex.org/W4415064585","https://openalex.org/W7135043786","https://openalex.org/W7202119280","https://openalex.org/W7202362161"],"related_works":["https://openalex.org/W7202119280","https://openalex.org/W7212022796","https://openalex.org/W7204040582","https://openalex.org/W7204034780","https://openalex.org/W4415068415","https://openalex.org/W7214987274","https://openalex.org/W4414786777","https://openalex.org/W7104464230","https://openalex.org/W7135043786","https://openalex.org/W7214195921"],"abstract_inverted_index":{"Emulation":[0],"of":[1,139,180,189,232,243],"high-accuracy":[2],"floating-point":[3],"matrix":[4,31,114,159],"multiplication,":[5],"as":[6,126,135],"in":[7,213,218,252],"the":[8,12,41,44,61,127,167,174,183,187,190,228,237,240,268],"Ozaki":[9,233],"scheme,":[10],"splits":[11,84,94],"inputs":[13],"into":[14,85,98],"low-precision":[15],"components":[16],"and":[17,26,47,108,129,144,182,210,216,258],"multiplies":[18,109],"them":[19,110],"pairwise.":[20],"These":[21],"products":[22,209,215,242],"must":[23],"be":[24],"error-free,":[25],"each":[27,95,198],"is":[28,186],"an":[29],"integer":[30,83,158,172],"product":[32,185,253],"times":[33],"a":[34],"scale":[35],"factor.":[36],"FP4":[37,55,89,113,152,191],"Tensor":[38],"Cores":[39],"are":[40,154,177,246],"fastest":[42],"on":[43],"NVIDIA":[45],"B200":[46],"B300":[48],"but":[49],"cannot":[50],"hold":[51],"INT8":[52,96],"operands.":[53],"The":[54,119,220,250],"values":[56],"scaled":[57,193],"by":[58,194],"2":[59],"form":[60],"set":[62],"$S":[63],"=":[64],"\\{0,":[65],"\\pm1,":[66],"\\pm2,":[67],"\\pm3,":[68],"\\pm4,":[69],"\\pm6,":[70],"\\pm8,":[71],"\\pm12\\}$,":[72],"which":[73,226],"contains":[74],"every":[75],"residue":[76,229],"modulo":[77,224],"13,":[78,106],"so":[79],"with":[80,103,207,257],"carries":[81],"any":[82],"base-13":[86],"digits":[87,101],"that":[88],"can":[90],"store.":[91],"Prior":[92],"work":[93,245],"operand":[97],"3":[99],"such":[100,125],"(limbs)":[102],"weights":[104],"$(1,":[105],"169)$":[107],"pairwise,":[111],"9":[112],"multiplications":[115],"(GEMMs)":[116],"for":[117,156,205,236],"INT8$\\times$INT8.":[118],"classical":[120],"ways":[121],"to":[122,248],"reduce":[123],"products,":[124],"Karatsuba":[128],"Toom--Cook":[130],"methods,":[131],"do":[132],"not":[133],"apply":[134],"they":[136],"stand:":[137],"sums":[138],"limbs":[140,168],"reach":[141],"$\\pm":[142],"24$":[143],"leave":[145],"$S$.":[146],"This":[147],"paper":[148],"asks":[149],"how":[150],"many":[151],"GEMMs":[153,192],"needed":[155],"one":[157],"product.":[160],"We":[161,266],"propose":[162],"Adaptive":[163],"Weight":[164],"Encoding":[165],"(AWE):":[166],"take":[169],"freely":[170],"chosen":[171],"weights,":[173],"stored":[175],"planes":[176],"linear":[178],"combinations":[179],"limbs,":[181],"exact":[184],"sum":[188],"reconstruction":[195],"coefficients.":[196],"For":[197],"input":[199,263],"range,":[200],"we":[201],"searched":[202],"these":[203],"choices":[204],"encodings":[206,256,269],"fewer":[208],"found":[211],"INT8$\\times$INT8":[212],"6":[214],"INT4$\\times$INT8":[217],"4.":[219],"formulation":[221],"also":[222],"holds":[223],"$m$,":[225],"covers":[227],"number":[230],"systems":[231],"scheme":[234],"II:":[235],"FP64":[238],"significand,":[239],"75":[241],"prior":[244],"reduced":[247],"59.":[249],"boundary":[251],"count":[254],"between":[255],"without":[259],"residues":[260],"lies":[261],"near":[262],"width":[264],"15.":[265],"release":[267],"found.":[270]},"counts_by_year":[{"year":2026,"cited_by_count":1}],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-09-23T00:00:00"}
