{"id":"https://openalex.org/W7220925203","doi":"https://doi.org/10.48550/arxiv.2610.08231","title":"OSFP4: Joint Optimization of Diagonal Smoothing and Block Scales for NVFP4 Quantization","display_name":"OSFP4: Joint Optimization of Diagonal Smoothing and Block Scales for NVFP4 Quantization","publication_year":2026,"publication_date":"2026-10-06","ids":{"openalex":"https://openalex.org/W7220925203","doi":"https://doi.org/10.48550/arxiv.2610.08231"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2610.08231","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.08231","pdf_url":"https://arxiv.org/pdf/2610.08231","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/2610.08231","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5157899622","display_name":"Neriah Ben David","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"David, Neriah Ben","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5157973948","display_name":"Ori Meir","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Meir, Ori","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5004351362","display_name":"Or Ordentlich","orcid":"https://orcid.org/0000-0002-5791-7923","observed_orcids":["https://orcid.org/0000-0002-5791-7923"]},"institutions":[{"id":"https://openalex.org/I197251160","display_name":"Hebrew University of Jerusalem","ror":"https://ror.org/03qxff017","country_code":"IL","type":"education","lineage":["https://openalex.org/I197251160"]}],"countries":["IL"],"is_corresponding":false,"raw_author_name":"Ordentlich, Or","raw_affiliation_strings":["School of Computer Science and Engineering Hebrew University of Jerusalem Jerusalem , Israel"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"School of Computer Science and Engineering Hebrew University of Jerusalem Jerusalem , Israel","institution_ids":["https://openalex.org/I197251160"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"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":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":0.7279999852180481,"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":0.7279999852180481,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.1875,"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/T10363","display_name":"Low-power high-performance VLSI design","score":0.019300000742077827,"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/successive-interference-cancellation","display_name":"successive interference cancellation","score":0.6539999842643738},{"id":"https://openalex.org/keywords/gptq","display_name":"GPTQ","score":0.7829999923706055},{"id":"https://openalex.org/keywords/large-language-model-inference","display_name":"large language model inference","score":0.8140000104904175},{"id":"https://openalex.org/keywords/quantization-error-minimization","display_name":"quantization error minimization","score":0.746999979019165}],"concepts":[{"id":"https://openalex.org/C3770464","wikidata":"https://www.wikidata.org/wiki/Q775963","display_name":"Smoothing","level":2,"score":0.8080999851226807},{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.7738999724388123},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.6908000111579895},{"id":"https://openalex.org/C136625980","wikidata":"https://www.wikidata.org/wiki/Q663208","display_name":"Rounding","level":2,"score":0.5544000267982483},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.5189999938011169},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.49540001153945923},{"id":"https://openalex.org/C185429906","wikidata":"https://www.wikidata.org/wiki/Q1130160","display_name":"Estimator","level":2,"score":0.45509999990463257},{"id":"https://openalex.org/C130367717","wikidata":"https://www.wikidata.org/wiki/Q189791","display_name":"Diagonal","level":2,"score":0.45329999923706055},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.44130000472068787},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.4131999909877777},{"id":"https://openalex.org/C2777210771","wikidata":"https://www.wikidata.org/wiki/Q4927124","display_name":"Block (permutation group theory)","level":2,"score":0.3531000018119812},{"id":"https://openalex.org/C156273044","wikidata":"https://www.wikidata.org/wiki/Q4913766","display_name":"Bin","level":2,"score":0.34880000352859497},{"id":"https://openalex.org/C157972887","wikidata":"https://www.wikidata.org/wiki/Q463359","display_name":"Convex optimization","level":3,"score":0.3127000033855438},{"id":"https://openalex.org/C61005703","wikidata":"https://www.wikidata.org/wiki/Q2145211","display_name":"Round-off error","level":2,"score":0.2906000018119812},{"id":"https://openalex.org/C147764199","wikidata":"https://www.wikidata.org/wiki/Q6865248","display_name":"Minification","level":2,"score":0.28600001335144043},{"id":"https://openalex.org/C199833920","wikidata":"https://www.wikidata.org/wiki/Q612536","display_name":"Vector quantization","level":2,"score":0.2827000021934509},{"id":"https://openalex.org/C18555067","wikidata":"https://www.wikidata.org/wiki/Q8375051","display_name":"Joint (building)","level":2,"score":0.27239999175071716},{"id":"https://openalex.org/C70451592","wikidata":"https://www.wikidata.org/wiki/Q376493","display_name":"Dither","level":3,"score":0.2671999931335449},{"id":"https://openalex.org/C106487976","wikidata":"https://www.wikidata.org/wiki/Q685816","display_name":"Matrix (chemical analysis)","level":2,"score":0.2630000114440918},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.26249998807907104}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2610.08231","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.08231","pdf_url":"https://arxiv.org/pdf/2610.08231","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.2610.08231","doi":"https://doi.org/10.48550/arxiv.2610.08231","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2610.08231","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2610.08231","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.08231","pdf_url":"https://arxiv.org/pdf/2610.08231","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":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7220925203.pdf","grobid_xml":"https://content.openalex.org/works/W7220925203.grobid-xml"},"referenced_works_count":16,"referenced_works":["https://openalex.org/W4307934016","https://openalex.org/W4379260375","https://openalex.org/W4385326807","https://openalex.org/W4386231419","https://openalex.org/W4399115604","https://openalex.org/W4403565099","https://openalex.org/W4407632226","https://openalex.org/W4415796210","https://openalex.org/W4416956316","https://openalex.org/W4416969200","https://openalex.org/W7134059717","https://openalex.org/W7161008710","https://openalex.org/W7161033828","https://openalex.org/W7161152162","https://openalex.org/W7164019442","https://openalex.org/W7204747440"],"related_works":["https://openalex.org/W7161008710","https://openalex.org/W7164019442","https://openalex.org/W4416969200","https://openalex.org/W7166887532","https://openalex.org/W7172416727","https://openalex.org/W7206148958","https://openalex.org/W7204747440","https://openalex.org/W7140277917","https://openalex.org/W4406231048","https://openalex.org/W7127094184"],"abstract_inverted_index":{"NVFP4":[0,22,41,136],"is":[1,73,98,145],"an":[2],"attractive":[3],"datatype":[4],"for":[5,40],"large":[6],"language":[7],"model":[8],"(LLM)":[9],"inference,":[10],"offering":[11],"compact":[12],"storage":[13],"and":[14,38],"native":[15],"tensor-core":[16],"acceleration.":[17],"However,":[18],"preserving":[19],"accuracy":[20,120],"using":[21],"requires":[23,83],"careful":[24],"quantization.":[25],"In":[26],"this":[27],"work":[28],"we":[29],"develop":[30],"a":[31,49,102],"novel":[32],"quantization":[33,62,128],"scheme":[34],"called":[35],"Optimized":[36],"Smoothing":[37],"Scaling":[39],"(OSFP4).":[42],"For":[43],"each":[44],"linear":[45],"projection":[46],"it":[47],"uses":[48],"diagonal":[50],"smoothing":[51,89],"matrix":[52],"whose":[53],"entries":[54,90],"are":[55],"optimized":[56],"to":[57],"minimize":[58],"the":[59,69,88,94,108,117,122,126,140],"squared":[60],"matrix-product":[61],"error":[63],"under":[64],"NVFP4,":[65],"taking":[66],"into":[67],"account":[68],"rounding":[70],"procedure":[71],"that":[72,114],"used":[74],"(either":[75],"round-to-nearest,":[76],"or":[77],"GPTQ-style":[78],"successive":[79],"interference":[80],"cancellation).":[81],"This":[82],"performing":[84],"joint":[85],"optimization":[86],"on":[87,139],"as":[91,93],"well":[92],"block":[95],"scales,":[96],"which":[97],"facilitated":[99],"by":[100],"analyzing":[101],"multiplicative-dither":[103],"FP4":[104],"quantizer":[105],"instead":[106],"of":[107,134],"fixed":[109],"deterministic":[110],"one.":[111],"Experiments":[112],"show":[113],"OSFP4":[115],"achieves":[116],"highest":[118],"average":[119],"among":[121],"evaluated":[123],"competitors":[124],"in":[125,147],"corresponding":[127],"settings,":[129],"while":[130],"retaining":[131],"approximately":[132],"94-97\\%":[133],"vendor":[135],"prefill":[137],"throughput":[138],"measured":[141],"workloads.":[142],"Our":[143],"code":[144],"available":[146],"https://github.com/neriahbd/OSFP4":[148]},"counts_by_year":[],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-10-09T00:00:00"}