{"id":"https://openalex.org/W7161540980","doi":"https://doi.org/10.48550/arxiv.2605.15250","title":"GQLA: Group-Query Latent Attention for Hardware-Adaptive Large Language Model Decoding","display_name":"GQLA: Group-Query Latent Attention for Hardware-Adaptive Large Language Model Decoding","publication_year":2026,"publication_date":"2026-05-14","ids":{"openalex":"https://openalex.org/W7161540980","doi":"https://doi.org/10.48550/arxiv.2605.15250"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.15250","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.15250","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.2605.15250","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5136368742","display_name":"Fanxu Meng","orcid":null},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Meng, Fanxu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":["https://openalex.org/A5136368742"],"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/T10181","display_name":"Natural Language Processing Techniques","score":0.2515000104904175,"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"}},"topics":[{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.2515000104904175,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.15399999916553497,"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/T10028","display_name":"Topic Modeling","score":0.07400000095367432,"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/path","display_name":"Path (computing)","score":0.7476999759674072},{"id":"https://openalex.org/keywords/decoding-methods","display_name":"Decoding methods","score":0.6710000038146973},{"id":"https://openalex.org/keywords/set","display_name":"Set (abstract data type)","score":0.605400025844574},{"id":"https://openalex.org/keywords/cache","display_name":"Cache","score":0.5702999830245972},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.5673999786376953},{"id":"https://openalex.org/keywords/encoding","display_name":"Encoding (memory)","score":0.45089998841285706},{"id":"https://openalex.org/keywords/tensor","display_name":"Tensor (intrinsic definition)","score":0.44749999046325684},{"id":"https://openalex.org/keywords/parallelism","display_name":"Parallelism (grammar)","score":0.4357999861240387},{"id":"https://openalex.org/keywords/task","display_name":"Task (project management)","score":0.4016000032424927}],"concepts":[{"id":"https://openalex.org/C2777735758","wikidata":"https://www.wikidata.org/wiki/Q817765","display_name":"Path (computing)","level":2,"score":0.7476999759674072},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7307999730110168},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.6710000038146973},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.605400025844574},{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.5702999830245972},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5673999786376953},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.45089998841285706},{"id":"https://openalex.org/C155281189","wikidata":"https://www.wikidata.org/wiki/Q3518150","display_name":"Tensor (intrinsic definition)","level":2,"score":0.44749999046325684},{"id":"https://openalex.org/C2781172179","wikidata":"https://www.wikidata.org/wiki/Q853109","display_name":"Parallelism (grammar)","level":2,"score":0.4357999861240387},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.4016000032424927},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.38999998569488525},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3549000024795532},{"id":"https://openalex.org/C2780980858","wikidata":"https://www.wikidata.org/wiki/Q110022","display_name":"Dual (grammatical number)","level":2,"score":0.34880000352859497},{"id":"https://openalex.org/C199360897","wikidata":"https://www.wikidata.org/wiki/Q9143","display_name":"Programming language","level":1,"score":0.3382999897003174},{"id":"https://openalex.org/C2777210771","wikidata":"https://www.wikidata.org/wiki/Q4927124","display_name":"Block (permutation group theory)","level":2,"score":0.33570000529289246},{"id":"https://openalex.org/C32638748","wikidata":"https://www.wikidata.org/wiki/Q5437051","display_name":"Fast path","level":4,"score":0.33329999446868896},{"id":"https://openalex.org/C127162648","wikidata":"https://www.wikidata.org/wiki/Q16858953","display_name":"Channel (broadcasting)","level":2,"score":0.3246999979019165},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.3149999976158142},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.3075000047683716},{"id":"https://openalex.org/C159254197","wikidata":"https://www.wikidata.org/wiki/Q1144915","display_name":"Lexicographical order","level":2,"score":0.3043000102043152},{"id":"https://openalex.org/C2780791683","wikidata":"https://www.wikidata.org/wiki/Q846785","display_name":"Action (physics)","level":2,"score":0.30309998989105225},{"id":"https://openalex.org/C77618280","wikidata":"https://www.wikidata.org/wiki/Q1155772","display_name":"Scheme (mathematics)","level":2,"score":0.30230000615119934},{"id":"https://openalex.org/C115874739","wikidata":"https://www.wikidata.org/wiki/Q825377","display_name":"Critical path method","level":2,"score":0.30160000920295715},{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.28049999475479126},{"id":"https://openalex.org/C59656382","wikidata":"https://www.wikidata.org/wiki/Q191536","display_name":"Conjunction (astronomy)","level":2,"score":0.27619999647140503},{"id":"https://openalex.org/C12725497","wikidata":"https://www.wikidata.org/wiki/Q810247","display_name":"Baseline (sea)","level":2,"score":0.2750999927520752},{"id":"https://openalex.org/C94375191","wikidata":"https://www.wikidata.org/wiki/Q11205","display_name":"Arithmetic","level":1,"score":0.2727999985218048},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.27250000834465027},{"id":"https://openalex.org/C2779662365","wikidata":"https://www.wikidata.org/wiki/Q5416694","display_name":"Event (particle physics)","level":2,"score":0.26330000162124634},{"id":"https://openalex.org/C2779439359","wikidata":"https://www.wikidata.org/wiki/Q317088","display_name":"Commodity","level":2,"score":0.2623000144958496},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.2556000053882599}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.15250","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.15250","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.2605.15250","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.15250","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":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Multi-head":[0],"Latent":[1,74],"Attention":[2,75],"(MLA),":[3],"the":[4,20,52,68,92,113,117,135,157,184,191,195,204],"attention":[5],"used":[6],"in":[7],"DeepSeek-V2/V3,":[8],"jointly":[9],"compresses":[10,183],"keys":[11],"and":[12,18,55,101,142],"values":[13],"into":[14,168,176],"a":[15,77,102,106,128,172,177],"low-rank":[16],"latent":[17],"matches":[19,116],"H100":[21,139],"roofline":[22],"almost":[23],"perfectly.":[24],"Its":[25],"trained":[26,83],"weights,":[27],"however,":[28],"expose":[29,85],"only":[30],"one":[31],"decoding":[32,89],"path":[33,97,104,114,197],"-":[34,39,120,126],"an":[35,95],"absorbed":[36],"MQA":[37],"form":[38],"which":[40,170],"ties":[41],"efficient":[42],"inference":[43,64],"to":[44,99,151,188],"H100-class":[45],"compute-bandwidth":[46],"ratios,":[47],"forfeits":[48],"tensor":[49,154],"parallelism":[50,155],"along":[51],"head":[53],"axis,":[54],"yields":[56],"no":[57,121,123],"Multi-Token":[58],"Prediction":[59],"(MTP)":[60],"gain":[61],"on":[62,156,180,194,203],"commodity":[63],"GPUs":[65],"such":[66],"as":[67],"export-restricted":[69],"H20.":[70],"We":[71],"propose":[72],"Group-Query":[73],"(GQLA),":[76],"minimal":[78],"modification":[79],"of":[80,131,137,190],"MLA":[81],"whose":[82],"weights":[84,133],"two":[86],"algebraically":[87],"equivalent":[88],"paths":[90],"over":[91],"same":[93],"parameters:":[94],"MQA-absorb":[96,196],"identical":[98],"MLA's,":[100],"GQA":[103,158,174,192],"with":[105],"per-group":[107,205],"expanded":[108],"cache.":[109],"The":[110],"runtime":[111],"picks":[112],"that":[115],"target":[118],"hardware":[119],"retraining,":[122],"custom":[124],"kernels":[125],"so":[127],"single":[129],"set":[130],"GQLA":[132,178],"pins":[134],"rooflines":[136],"both":[138],"(MQA-absorb,":[140],"s_q=1)":[141],"H20":[143],"(GQA":[144],"+":[145],"MTP,":[146],"s_q=2),":[147],"while":[148,198],"supporting":[149],"up":[150],"8-way":[152],"zero-redundancy":[153],"path.":[159,206],"To":[160],"avoid":[161],"pretraining":[162],"from":[163],"scratch":[164],"we":[165],"extend":[166],"TransMLA":[167],"TransGQLA,":[169],"converts":[171],"pretrained":[173],"checkpoint":[175],"model;":[179],"LLaMA-3-8B":[181],"it":[182],"per-token":[185],"KV":[186],"cache":[187],"28.125%":[189],"baseline":[193],"structurally":[199],"preserving":[200],"GQA-level":[201],"traffic":[202]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-05-19T00:00:00"}