{"id":"https://openalex.org/W7214988073","doi":"https://doi.org/10.48550/arxiv.2609.36760","title":"QuantMLA: Function-Aligned Dual-Path Quantization for Low-Bit MLA KV Caching","display_name":"QuantMLA: Function-Aligned Dual-Path Quantization for Low-Bit MLA KV Caching","publication_year":2026,"publication_date":"2026-09-29","ids":{"openalex":"https://openalex.org/W7214988073","doi":"https://doi.org/10.48550/arxiv.2609.36760"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2609.36760","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.36760","pdf_url":"https://arxiv.org/pdf/2609.36760","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/2609.36760","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5122136707","display_name":"Zunhai Su","orcid":"https://orcid.org/0009-0006-2011-2044","observed_orcids":["https://orcid.org/0009-0006-2011-2044"]},"institutions":[{"id":"https://openalex.org/I4405259410","display_name":"Meituan (China)","ror":"https://ror.org/0272kpg92","country_code":"CN","type":"company","lineage":["https://openalex.org/I4405259410"]},{"id":"https://openalex.org/I889458895","display_name":"University of Hong Kong","ror":"https://ror.org/02zhqgq86","country_code":"HK","type":"education","lineage":["https://openalex.org/I889458895"]}],"countries":["CN","HK"],"is_corresponding":false,"raw_author_name":"Su, Zunhai","raw_affiliation_strings":["Meituan LongCat Team","The University of Hong Kong"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Meituan LongCat Team","institution_ids":["https://openalex.org/I4405259410"]},{"raw_affiliation_string":"The University of Hong Kong","institution_ids":["https://openalex.org/I889458895"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5146476691","display_name":"Yuxuan Sun","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I4405259410","display_name":"Meituan (China)","ror":"https://ror.org/0272kpg92","country_code":"CN","type":"company","lineage":["https://openalex.org/I4405259410"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Sun, Yuxuan","raw_affiliation_strings":["Meituan LongCat Team"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Meituan LongCat Team","institution_ids":["https://openalex.org/I4405259410"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133356299","display_name":"Jianchao Tan","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I4405259410","display_name":"Meituan (China)","ror":"https://ror.org/0272kpg92","country_code":"CN","type":"company","lineage":["https://openalex.org/I4405259410"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Tan, Jianchao","raw_affiliation_strings":["Meituan LongCat Team"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Meituan LongCat Team","institution_ids":["https://openalex.org/I4405259410"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5153637930","display_name":"Tao Zhang","orcid":"https://orcid.org/0000-0002-6272-4069","observed_orcids":["https://orcid.org/0000-0002-6272-4069"]},"institutions":[{"id":"https://openalex.org/I90610280","display_name":"South China University of Technology","ror":"https://ror.org/0530pts50","country_code":"CN","type":"education","lineage":["https://openalex.org/I90610280"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Zhang, Tao","raw_affiliation_strings":["South China University of Technology"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"South China University of Technology","institution_ids":["https://openalex.org/I90610280"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5036320354","display_name":"Ruihan Hu","orcid":"https://orcid.org/0000-0001-8525-2503","observed_orcids":["https://orcid.org/0000-0001-8525-2503","https://orcid.org/0009-0003-5286-9373"]},"institutions":[{"id":"https://openalex.org/I204983213","display_name":"Harbin Institute of Technology","ror":"https://ror.org/01yqg2h08","country_code":"CN","type":"education","lineage":["https://openalex.org/I204983213"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Hu, Ruihan","raw_affiliation_strings":["Harbin Institute of Technology"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Harbin Institute of Technology","institution_ids":["https://openalex.org/I204983213"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5153621656","display_name":"Yuchen Xie","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I4405259410","display_name":"Meituan (China)","ror":"https://ror.org/0272kpg92","country_code":"CN","type":"company","lineage":["https://openalex.org/I4405259410"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Xie, Yuchen","raw_affiliation_strings":["Meituan LongCat Team"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Meituan LongCat Team","institution_ids":["https://openalex.org/I4405259410"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5153650427","display_name":"Xunliang Cai","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I4405259410","display_name":"Meituan (China)","ror":"https://ror.org/0272kpg92","country_code":"CN","type":"company","lineage":["https://openalex.org/I4405259410"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Cai, Xunliang","raw_affiliation_strings":["Meituan LongCat Team"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Meituan LongCat Team","institution_ids":["https://openalex.org/I4405259410"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5072120057","display_name":"Ngai Wong","orcid":"https://orcid.org/0000-0002-3026-0108","observed_orcids":["https://orcid.org/0000-0002-3026-0108"]},"institutions":[{"id":"https://openalex.org/I889458895","display_name":"University of Hong Kong","ror":"https://ror.org/02zhqgq86","country_code":"HK","type":"education","lineage":["https://openalex.org/I889458895"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Wong, Ngai","raw_affiliation_strings":["The University of Hong Kong"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"The University of Hong Kong","institution_ids":["https://openalex.org/I889458895"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":4,"corresponding_author_ids":[],"corresponding_institution_ids":[],"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/T10036","display_name":"Advanced Neural Network Applications","score":0.8213000297546387,"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":0.8213000297546387,"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":0.030400000512599945,"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"}},{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.023399999365210533,"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/multi-head-latent-attention","display_name":"multi-head latent attention","score":0.9860000014305115},{"id":"https://openalex.org/keywords/kv-caching","display_name":"KV caching","score":0.7609999775886536},{"id":"https://openalex.org/keywords/low-bit-quantization","display_name":"low-bit quantization","score":0.8240000009536743},{"id":"https://openalex.org/keywords/rope","display_name":"rope","score":0.6439999938011169},{"id":"https://openalex.org/keywords/int4-quantization","display_name":"INT4 quantization","score":0.6129999756813049},{"id":"https://openalex.org/keywords/cache-compression","display_name":"cache compression","score":0.47699999809265137},{"id":"https://openalex.org/keywords/dequantization","display_name":"dequantization","score":0.4729999899864197},{"id":"https://openalex.org/keywords/large-language-model-inference","display_name":"large language model inference","score":0.5680000185966492}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7867000102996826},{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.7853999733924866},{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.4950999915599823},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.47749999165534973},{"id":"https://openalex.org/C199833920","wikidata":"https://www.wikidata.org/wiki/Q612536","display_name":"Vector quantization","level":2,"score":0.4648999869823456},{"id":"https://openalex.org/C189783530","wikidata":"https://www.wikidata.org/wiki/Q352090","display_name":"CPU cache","level":3,"score":0.4546999931335449},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.44679999351501465},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.42559999227523804},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.42559999227523804},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.385699987411499},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.37599998712539673},{"id":"https://openalex.org/C74193536","wikidata":"https://www.wikidata.org/wiki/Q574844","display_name":"Kernel (algebra)","level":2,"score":0.36730000376701355},{"id":"https://openalex.org/C38556500","wikidata":"https://www.wikidata.org/wiki/Q13404475","display_name":"Cache algorithms","level":4,"score":0.36230000853538513},{"id":"https://openalex.org/C43126263","wikidata":"https://www.wikidata.org/wiki/Q128751","display_name":"Source code","level":2,"score":0.3610000014305115},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.3521000146865845},{"id":"https://openalex.org/C201148951","wikidata":"https://www.wikidata.org/wiki/Q5015976","display_name":"Cache coloring","level":4,"score":0.3422999978065491},{"id":"https://openalex.org/C78548338","wikidata":"https://www.wikidata.org/wiki/Q2493","display_name":"Data compression","level":2,"score":0.3174999952316284},{"id":"https://openalex.org/C204241405","wikidata":"https://www.wikidata.org/wiki/Q461499","display_name":"Transformation (genetics)","level":3,"score":0.3174000084400177},{"id":"https://openalex.org/C165064840","wikidata":"https://www.wikidata.org/wiki/Q1321061","display_name":"Matching (statistics)","level":2,"score":0.3131999969482422},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.29760000109672546},{"id":"https://openalex.org/C2985998994","wikidata":"https://www.wikidata.org/wiki/Q3644502","display_name":"Source model","level":2,"score":0.295199990272522},{"id":"https://openalex.org/C2779960059","wikidata":"https://www.wikidata.org/wiki/Q7113681","display_name":"Overhead (engineering)","level":2,"score":0.27549999952316284},{"id":"https://openalex.org/C47487241","wikidata":"https://www.wikidata.org/wiki/Q5227230","display_name":"Data access","level":2,"score":0.2721000015735626},{"id":"https://openalex.org/C157764524","wikidata":"https://www.wikidata.org/wiki/Q1383412","display_name":"Throughput","level":3,"score":0.26249998807907104},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.26190000772476196},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.2612999975681305},{"id":"https://openalex.org/C180016635","wikidata":"https://www.wikidata.org/wiki/Q2712821","display_name":"Compression (physics)","level":2,"score":0.25029999017715454}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2609.36760","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.36760","pdf_url":"https://arxiv.org/pdf/2609.36760","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.2609.36760","doi":"https://doi.org/10.48550/arxiv.2609.36760","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2609.36760","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:2609.36760","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.36760","pdf_url":"https://arxiv.org/pdf/2609.36760","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/W7214988073.pdf","grobid_xml":"https://content.openalex.org/works/W7214988073.grobid-xml"},"referenced_works_count":31,"referenced_works":["https://openalex.org/W2794325560","https://openalex.org/W2950700477","https://openalex.org/W2963748441","https://openalex.org/W2989588035","https://openalex.org/W2996908057","https://openalex.org/W3134642945","https://openalex.org/W3177813494","https://openalex.org/W3210904070","https://openalex.org/W4362598949","https://openalex.org/W4387321091","https://openalex.org/W4396815229","https://openalex.org/W4396912812","https://openalex.org/W4403565099","https://openalex.org/W4405903187","https://openalex.org/W4406840497","https://openalex.org/W4407208746","https://openalex.org/W4412944812","https://openalex.org/W4414281281","https://openalex.org/W4414359395","https://openalex.org/W4414846785","https://openalex.org/W4415796640","https://openalex.org/W4416102280","https://openalex.org/W4416117793","https://openalex.org/W4416135561","https://openalex.org/W4416691570","https://openalex.org/W4416915242","https://openalex.org/W7077044503","https://openalex.org/W7103188877","https://openalex.org/W7128678202","https://openalex.org/W7161742294","https://openalex.org/W7161811629"],"related_works":["https://openalex.org/W7222635619","https://openalex.org/W7221828960","https://openalex.org/W4407819997","https://openalex.org/W7171964880","https://openalex.org/W7138903718","https://openalex.org/W7171503847","https://openalex.org/W4417282901","https://openalex.org/W7165750449","https://openalex.org/W4416330253","https://openalex.org/W7204789226"],"abstract_inverted_index":{"Multi-Head":[0],"Latent":[1],"Attention":[2],"(MLA)":[3],"enables":[4],"expressive":[5],"multi-head":[6],"attention":[7,124,190,199],"with":[8,24,100,156],"compact":[9],"caches":[10,155],"for":[11,68],"its":[12],"content":[13,107,152,163],"and":[14,27,50,111,126,153,181,195],"decoupled":[15],"RoPE":[16,154,170],"paths,":[17],"yet":[18],"cache":[19,164,172,203],"memory":[20],"still":[21],"scales":[22],"linearly":[23],"context":[25],"length":[26],"batch":[28],"size.":[29],"In":[30],"this":[31,60],"work,":[32],"we":[33,62],"establish":[34],"a":[35,65,128,186,212],"systematic":[36],"model":[37,86,137],"of":[38,55,122,150],"MLA's":[39],"dual-path":[40,70],"quantization":[41],"errors,":[42,113],"characterizing":[43],"their":[44],"distinct":[45],"effects":[46],"on":[47,131,178],"attention-output":[48,103],"distortion":[49],"explaining":[51],"the":[52,106,119,123,144,151,162,169],"pronounced":[53],"amplification":[54],"RoPE-path":[56],"errors.":[57],"Guided":[58],"by":[59],"analysis,":[61],"introduce":[63],"QuantMLA,":[64],"function-aligned":[66,101],"framework":[67],"low-bit":[69,188],"quantization.":[71],"We":[72,184],"derive":[73],"path-specific":[74,98],"transformation":[75,91],"spaces":[76],"that":[77,192],"preserve":[78],"full-precision":[79],"computation":[80],"while":[81,114,167,211],"remaining":[82],"fully":[83],"fusible":[84],"into":[85,198],"parameters":[87],"offline,":[88],"eliminating":[89],"online":[90],"overhead.":[92],"Within":[93],"these":[94],"spaces,":[95],"QuantMLA":[96,139],"learns":[97],"transformations":[99],"objectives:":[102],"reconstruction":[104,117],"captures":[105],"path's":[108],"coupled":[109],"matching":[110],"aggregation":[112],"positional":[115],"QK":[116],"preserves":[118],"RoPE-induced":[120],"component":[121],"logits":[125],"admits":[127],"theoretical":[129],"bound":[130],"output":[132,220],"distortion.":[133],"Across":[134],"four":[135],"MLA":[136,189],"families,":[138],"enables,":[140],"to":[141,165],"our":[142],"knowledge,":[143],"first":[145],"reported":[146],"joint":[147],"INT4":[148,174],"caching":[149],"minimal":[157],"accuracy":[158],"degradation.":[159],"Further":[160],"compressing":[161],"INT2":[166],"retaining":[168],"key":[171],"at":[173,208],"maintains":[175],"competitive":[176],"performance":[177],"challenging":[179],"reasoning":[180],"code":[182,225],"benchmarks.":[183],"develop":[185],"native":[187],"kernel":[191],"integrates":[193],"unpacking":[194],"dequantization":[196],"directly":[197],"computation.":[200],"The":[201,224],"physical":[202],"layout":[204],"provides":[205],"3.59x":[206],"compression":[207],"128K":[209],"context,":[210],"cache-pressure":[213],"serving":[214],"workload":[215],"achieves":[216],"5.168x":[217],"higher":[218],"whole-job":[219],"throughput":[221],"than":[222],"BF16.":[223],"will":[226],"be":[227],"released":[228],"upon":[229],"acceptance.":[230]},"counts_by_year":[{"year":2026,"cited_by_count":1}],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-10-01T00:00:00"}