{"id":"https://openalex.org/W7164955587","doi":"https://doi.org/10.48550/arxiv.2606.15682","title":"ReQAT: Achieving Full-Precision Reasoning Accuracy with 4-bit Floating-Point Quantization-Aware Training","display_name":"ReQAT: Achieving Full-Precision Reasoning Accuracy with 4-bit Floating-Point Quantization-Aware Training","publication_year":2026,"publication_date":"2026-06-14","ids":{"openalex":"https://openalex.org/W7164955587","doi":"https://doi.org/10.48550/arxiv.2606.15682"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.15682","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.15682","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2606.15682","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5138740881","display_name":"Janghwan Lee","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lee, Janghwan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5014854780","display_name":"Sihwa Lee","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lee, Sihwa","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138730870","display_name":"Jinseok Kim","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kim, Jinseok","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5054973256","display_name":"Yongjik Kim","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kim, Yongjik","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5068625256","display_name":"Jieun Lim","orcid":"https://orcid.org/0000-0002-8985-1957"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lim, Jieun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5089478486","display_name":"Jinwook Oh","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Oh, Jinwook","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5113468057","display_name":"J H Choi","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Choi, Jungwook","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.3440000116825104,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.3440000116825104,"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/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.18170000612735748,"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/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.08510000258684158,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/initialization","display_name":"Initialization","score":0.7041000127792358},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.6848999857902527},{"id":"https://openalex.org/keywords/speedup","display_name":"Speedup","score":0.6263999938964844},{"id":"https://openalex.org/keywords/cache","display_name":"Cache","score":0.6241000294685364},{"id":"https://openalex.org/keywords/quantization","display_name":"Quantization (signal processing)","score":0.5580999851226807},{"id":"https://openalex.org/keywords/minification","display_name":"Minification","score":0.46860000491142273},{"id":"https://openalex.org/keywords/cpu-cache","display_name":"CPU cache","score":0.36730000376701355},{"id":"https://openalex.org/keywords/recall","display_name":"Recall","score":0.3610000014305115},{"id":"https://openalex.org/keywords/focus","display_name":"Focus (optics)","score":0.3382999897003174}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8009999990463257},{"id":"https://openalex.org/C114466953","wikidata":"https://www.wikidata.org/wiki/Q6034165","display_name":"Initialization","level":2,"score":0.7041000127792358},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.6848999857902527},{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.6263999938964844},{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.6241000294685364},{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.5580999851226807},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.49129998683929443},{"id":"https://openalex.org/C147764199","wikidata":"https://www.wikidata.org/wiki/Q6865248","display_name":"Minification","level":2,"score":0.46860000491142273},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.44510000944137573},{"id":"https://openalex.org/C189783530","wikidata":"https://www.wikidata.org/wiki/Q352090","display_name":"CPU cache","level":3,"score":0.36730000376701355},{"id":"https://openalex.org/C100660578","wikidata":"https://www.wikidata.org/wiki/Q18733","display_name":"Recall","level":2,"score":0.3610000014305115},{"id":"https://openalex.org/C192209626","wikidata":"https://www.wikidata.org/wiki/Q190909","display_name":"Focus (optics)","level":2,"score":0.3382999897003174},{"id":"https://openalex.org/C106301342","wikidata":"https://www.wikidata.org/wiki/Q4117933","display_name":"Entropy (arrow of time)","level":2,"score":0.32359999418258667},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.3192000091075897},{"id":"https://openalex.org/C157170001","wikidata":"https://www.wikidata.org/wiki/Q4781507","display_name":"Applications of artificial intelligence","level":2,"score":0.31839999556541443},{"id":"https://openalex.org/C81669768","wikidata":"https://www.wikidata.org/wiki/Q2359161","display_name":"Precision and recall","level":2,"score":0.31790000200271606},{"id":"https://openalex.org/C197792726","wikidata":"https://www.wikidata.org/wiki/Q191062","display_name":"Mnemonic","level":2,"score":0.3172000050544739},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.3142000138759613},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.2992999851703644},{"id":"https://openalex.org/C46686674","wikidata":"https://www.wikidata.org/wiki/Q466303","display_name":"Boosting (machine learning)","level":2,"score":0.2992999851703644},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.2870999872684479},{"id":"https://openalex.org/C195344581","wikidata":"https://www.wikidata.org/wiki/Q2555318","display_name":"Automated reasoning","level":2,"score":0.2800000011920929},{"id":"https://openalex.org/C177212765","wikidata":"https://www.wikidata.org/wiki/Q627335","display_name":"Workflow","level":2,"score":0.27639999985694885},{"id":"https://openalex.org/C23224414","wikidata":"https://www.wikidata.org/wiki/Q176769","display_name":"Hidden Markov model","level":2,"score":0.27129998803138733},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.26100000739097595},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.25189998745918274},{"id":"https://openalex.org/C52740198","wikidata":"https://www.wikidata.org/wiki/Q1539564","display_name":"Importance sampling","level":3,"score":0.25110000371932983},{"id":"https://openalex.org/C157764524","wikidata":"https://www.wikidata.org/wiki/Q1383412","display_name":"Throughput","level":3,"score":0.25040000677108765}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.15682","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.15682","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2606.15682","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.15682","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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"},"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],"Reasoning":[1],"Models":[2],"(LRMs)":[3],"achieve":[4],"strong":[5],"problem-solving":[6],"through":[7,78],"long":[8],"chain-of-thought,":[9],"but":[10,148],"their":[11],"deployment":[12],"is":[13],"constrained":[14],"by":[15],"the":[16,140],"high":[17],"cost":[18],"of":[19],"full-precision":[20],"inference":[21],"and":[22,39,50,69,123,164],"growing":[23],"KV":[24,40,133],"cache":[25,134],"footprints.":[26],"Microscaled":[27],"FP4":[28,32,58,90],"formats":[29],"enable":[30],"efficient":[31],"deployment;":[33],"however,":[34],"fully":[35],"quantizing":[36],"weights,":[37],"activations,":[38],"caches":[41],"(W4A4KV4)":[42],"causes":[43],"severe":[44],"reasoning":[45,79,103],"degradation":[46],"that":[47,57,76,129],"existing":[48],"PTQ":[49],"QAT":[51,98],"fail":[52],"to":[53,105,136,156],"recover.":[54],"We":[55],"identify":[56],"failures":[59],"concentrate":[60],"on":[61,82,108,160,166],"low-entropy":[62,110,121],"tokens--precise":[63],"symbolic":[64],"commitments":[65],"such":[66],"as":[67],"digits":[68],"operators--where":[70],"quantization":[71],"noise":[72],"inflates":[73],"sampling":[74],"errors":[75],"cascade":[77],"traces.":[80],"Based":[81],"this":[83],"insight,":[84],"we":[85],"propose":[86],"ReQAT,":[87],"a":[88,126],"reasoning-centric":[89],"training":[91,142],"framework":[92],"with":[93],"three":[94],"components:":[95],"(i)":[96],"Trace-Aligned":[97],"(TAQ),":[99],"which":[100,117],"revisits":[101],"identical":[102],"traces":[104],"focus":[106],"updates":[107],"critical":[109],"decisions;":[111],"(ii)":[112],"Selective":[113],"Entropy":[114],"Minimization":[115],"(SEM),":[116],"reinforces":[118],"confidence":[119],"at":[120],"positions;":[122],"(iii)":[124],"Q-FIT,":[125],"quantization-friendly":[127],"initialization":[128],"jointly":[130],"calibrates":[131],"RoPE-consistent":[132],"transformations":[135],"stabilize":[137],"QAT.":[138],"Under":[139],"same":[141],"budget,":[143],"ReQAT":[144],"not":[145],"only":[146],"recovers":[147],"surpasses":[149],"BF16":[150],"fine-tuning":[151],"accuracy,":[152],"while":[153],"delivering":[154],"up":[155],"3.9x":[157],"throughput":[158],"speedup":[159],"NVIDIA":[161],"DGX":[162],"Spark":[163],"3.1x":[165],"B200.":[167]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-06-17T00:00:00"}