{"id":"https://openalex.org/W7162503198","doi":"https://doi.org/10.48550/arxiv.2605.26660","title":"WINDQuant: Weight-Informed Neural Decision-Making for Global Mixed-Precision LLM Quantization","display_name":"WINDQuant: Weight-Informed Neural Decision-Making for Global Mixed-Precision LLM Quantization","publication_year":2026,"publication_date":"2026-05-26","ids":{"openalex":"https://openalex.org/W7162503198","doi":"https://doi.org/10.48550/arxiv.2605.26660"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.26660","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.26660","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":"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"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2605.26660","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5137093137","display_name":"Phong Nam Huu Nguyen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Nguyen, Phong Nam Huu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137115677","display_name":"Khoi M. Le","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Le, Khoi M.","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5025645183","display_name":"Cong-Duy Nguyen","orcid":"https://orcid.org/0000-0002-0931-460X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Nguyen, Cong-Duy T","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5050386762","display_name":"Anh Tuan Luu","orcid":"https://orcid.org/0000-0002-1927-9895"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Luu, Anh Tuan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137162816","display_name":"Thong Thanh Nguyen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Nguyen, Thong Thanh","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5137141662","display_name":"Tho Quan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Quan, Tho","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":0.4189999997615814,"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":0.4189999997615814,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.1281999945640564,"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/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.09200000017881393,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/quantization","display_name":"Quantization (signal processing)","score":0.7705000042915344},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.5461999773979187},{"id":"https://openalex.org/keywords/reinforcement-learning","display_name":"Reinforcement learning","score":0.5322999954223633},{"id":"https://openalex.org/keywords/heuristic","display_name":"Heuristic","score":0.40230000019073486},{"id":"https://openalex.org/keywords/sensitivity","display_name":"Sensitivity (control systems)","score":0.36039999127388},{"id":"https://openalex.org/keywords/overhead","display_name":"Overhead (engineering)","score":0.35679998993873596},{"id":"https://openalex.org/keywords/controller","display_name":"Controller (irrigation)","score":0.3546000123023987}],"concepts":[{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.7705000042915344},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7261000275611877},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5461999773979187},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.5322999954223633},{"id":"https://openalex.org/C173801870","wikidata":"https://www.wikidata.org/wiki/Q201413","display_name":"Heuristic","level":2,"score":0.40230000019073486},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.37459999322891235},{"id":"https://openalex.org/C21200559","wikidata":"https://www.wikidata.org/wiki/Q7451068","display_name":"Sensitivity (control systems)","level":2,"score":0.36039999127388},{"id":"https://openalex.org/C2779960059","wikidata":"https://www.wikidata.org/wiki/Q7113681","display_name":"Overhead (engineering)","level":2,"score":0.35679998993873596},{"id":"https://openalex.org/C203479927","wikidata":"https://www.wikidata.org/wiki/Q5165939","display_name":"Controller (irrigation)","level":2,"score":0.3546000123023987},{"id":"https://openalex.org/C199833920","wikidata":"https://www.wikidata.org/wiki/Q612536","display_name":"Vector quantization","level":2,"score":0.3479999899864197},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3240000009536743},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.29409998655319214},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.29100000858306885},{"id":"https://openalex.org/C165064840","wikidata":"https://www.wikidata.org/wiki/Q1321061","display_name":"Matching (statistics)","level":2,"score":0.2849000096321106},{"id":"https://openalex.org/C147764199","wikidata":"https://www.wikidata.org/wiki/Q6865248","display_name":"Minification","level":2,"score":0.2768000066280365},{"id":"https://openalex.org/C137836250","wikidata":"https://www.wikidata.org/wiki/Q984063","display_name":"Optimization problem","level":2,"score":0.2653000056743622},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.26010000705718994},{"id":"https://openalex.org/C2984842247","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep neural networks","level":3,"score":0.25099998712539673}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.26660","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.26660","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2605.26660","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.26660","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":"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"},"sustainable_development_goals":[{"score":0.473344624042511,"id":"https://metadata.un.org/sdg/16","display_name":"Peace, Justice and strong institutions"}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Quantization":[0],"is":[1],"an":[2],"effective":[3],"approach":[4],"to":[5,84,90,157],"reduce":[6],"the":[7,22,102,135],"memory":[8],"footprint":[9],"and":[10,42,87,108,130],"inference":[11],"cost":[12],"of":[13,134],"large":[14],"language":[15],"models":[16,142],"(LLMs),":[17],"yet":[18],"maintaining":[19],"performance":[20,148],"in":[21,149],"ultra-low-bit":[23,71,150],"regime":[24],"remains":[25],"challenging.":[26],"Existing":[27],"post-training":[28],"methods":[29],"often":[30],"suffer":[31],"from":[32],"severe":[33],"accuracy":[34],"degradation,":[35],"while":[36,152],"quantization-aware":[37],"training":[38],"requires":[39],"costly":[40],"retraining":[41],"additional":[43],"resources.":[44],"Moreover,":[45],"most":[46],"mixed-precision":[47,137,169],"strategies":[48],"rely":[49],"on":[50,140],"coarse-grained":[51],"or":[52],"heuristic":[53],"sensitivity":[54],"analysis":[55],"that":[56,144],"overlooks":[57],"fine-grained":[58,91,109],"variations":[59],"within":[60,112],"weight":[61],"matrices.":[62],"We":[63],"propose":[64],"WINDQuant,":[65],"a":[66,95,115,164],"reinforcement-learning-based":[67],"allocation":[68],"controller":[69,166],"for":[70,167],"LLM":[72],"quantization.":[73,170],"Rather":[74],"than":[75],"introducing":[76],"another":[77],"low-level":[78],"quantization":[79,88],"operator,":[80],"WINDQuant":[81,105,145],"learns":[82],"how":[83],"assign":[85],"bit-widths":[86],"treatments":[89],"column":[92],"chunks":[93],"under":[94,114],"global":[96,116],"storage":[97],"budget.":[98],"By":[99],"operating":[100],"at":[101],"column-chunk":[103],"level,":[104],"enables":[106],"flexible":[107],"precision":[110],"assignment":[111],"layers":[113],"target":[117],"bit-width.":[118],"The":[119],"implementation":[120],"combines":[121],"PPO":[122],"with":[123],"activation-aware":[124],"calibration,":[125],"lightweight":[126],"per-unit":[127],"quantizer":[128],"fitting,":[129],"explicit":[131],"effective-bit":[132],"accounting":[133],"learned":[136],"plan.":[138],"Experiments":[139],"LLaMA":[141],"demonstrate":[143],"achieves":[146],"competitive":[147],"settings":[151],"reducing":[153],"optimization":[154],"overhead":[155],"relative":[156],"retraining-based":[158],"approaches,":[159],"highlighting":[160],"reinforcement":[161],"learning":[162],"as":[163],"practical":[165],"adaptive":[168]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-05-28T00:00:00"}