{"id":"https://openalex.org/W7134902181","doi":"https://doi.org/10.1145/3779212.3790237","title":"TPLA: Tensor Parallel Latent Attention for Efficient Disaggregated Prefill & Decode Inference","display_name":"TPLA: Tensor Parallel Latent Attention for Efficient Disaggregated Prefill & Decode Inference","publication_year":2026,"publication_date":"2026-03-10","ids":{"openalex":"https://openalex.org/W7134902181","doi":"https://doi.org/10.1145/3779212.3790237"},"language":null,"primary_location":{"id":"doi:10.1145/3779212.3790237","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3779212.3790237","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 31st ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 2","raw_type":"proceedings-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://doi.org/10.1145/3779212.3790237","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":null,"display_name":"Xiaojuan Tang","orcid":"https://orcid.org/0009-0002-0486-8651"},"institutions":[{"id":"https://openalex.org/I20231570","display_name":"Peking University","ror":"https://ror.org/02v51f717","country_code":"CN","type":"education","lineage":["https://openalex.org/I20231570"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Xiaojuan Tang","raw_affiliation_strings":["Peking University, Beijing, China"],"raw_orcid":"https://orcid.org/0009-0002-0486-8651","affiliations":[{"raw_affiliation_string":"Peking University, Beijing, China","institution_ids":["https://openalex.org/I20231570"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Fanxu Meng","orcid":"https://orcid.org/0009-0009-2382-556X"},"institutions":[{"id":"https://openalex.org/I20231570","display_name":"Peking University","ror":"https://ror.org/02v51f717","country_code":"CN","type":"education","lineage":["https://openalex.org/I20231570"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Fanxu Meng","raw_affiliation_strings":["Peking University, Beijing, China"],"raw_orcid":"https://orcid.org/0009-0009-2382-556X","affiliations":[{"raw_affiliation_string":"Peking University, Beijing, China","institution_ids":["https://openalex.org/I20231570"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128764606","display_name":"Pingzhi Tang","orcid":null},"institutions":[{"id":"https://openalex.org/I20231570","display_name":"Peking University","ror":"https://ror.org/02v51f717","country_code":"CN","type":"education","lineage":["https://openalex.org/I20231570"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Pingzhi Tang","raw_affiliation_strings":["Peking University, Beijing, China"],"raw_orcid":"https://orcid.org/0009-0001-7958-7144","affiliations":[{"raw_affiliation_string":"Peking University, Beijing, China","institution_ids":["https://openalex.org/I20231570"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128739179","display_name":"Yuxuan Wang","orcid":null},"institutions":[{"id":"https://openalex.org/I20231570","display_name":"Peking University","ror":"https://ror.org/02v51f717","country_code":"CN","type":"education","lineage":["https://openalex.org/I20231570"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yuxuan Wang","raw_affiliation_strings":["Peking University, Beijing, China"],"raw_orcid":"https://orcid.org/0009-0009-7552-307X","affiliations":[{"raw_affiliation_string":"Peking University, Beijing, China","institution_ids":["https://openalex.org/I20231570"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Di Yin","orcid":"https://orcid.org/0009-0006-7246-3102"},"institutions":[{"id":"https://openalex.org/I2250653659","display_name":"Tencent (China)","ror":"https://ror.org/00hhjss72","country_code":"CN","type":"company","lineage":["https://openalex.org/I2250653659"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Di Yin","raw_affiliation_strings":["Tencent YouTu Lab, Shanghai, China"],"raw_orcid":"https://orcid.org/0009-0006-7246-3102","affiliations":[{"raw_affiliation_string":"Tencent YouTu Lab, Shanghai, China","institution_ids":["https://openalex.org/I2250653659"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128706625","display_name":"Xing Sun","orcid":null},"institutions":[{"id":"https://openalex.org/I2250653659","display_name":"Tencent (China)","ror":"https://ror.org/00hhjss72","country_code":"CN","type":"company","lineage":["https://openalex.org/I2250653659"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Xing Sun","raw_affiliation_strings":["Tencent YouTu Lab, Shanghai, China"],"raw_orcid":"https://orcid.org/0000-0001-8132-9083","affiliations":[{"raw_affiliation_string":"Tencent YouTu Lab, Shanghai, China","institution_ids":["https://openalex.org/I2250653659"]}]},{"author_position":"last","author":{"id":null,"display_name":"Muhan Zhang","orcid":"https://orcid.org/0000-0002-7680-6401"},"institutions":[{"id":"https://openalex.org/I20231570","display_name":"Peking University","ror":"https://ror.org/02v51f717","country_code":"CN","type":"education","lineage":["https://openalex.org/I20231570"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Muhan Zhang","raw_affiliation_strings":["Peking University, Beijing, China"],"raw_orcid":"https://orcid.org/0000-0002-7680-6401","affiliations":[{"raw_affiliation_string":"Peking University, Beijing, China","institution_ids":["https://openalex.org/I20231570"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.23943456,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"2048","last_page":"2062"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.10289999842643738,"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/T10028","display_name":"Topic Modeling","score":0.10289999842643738,"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/T12303","display_name":"Tensor decomposition and applications","score":0.10199999809265137,"subfield":{"id":"https://openalex.org/subfields/2605","display_name":"Computational Mathematics"},"field":{"id":"https://openalex.org/fields/26","display_name":"Mathematics"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.094200000166893,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.5792999863624573},{"id":"https://openalex.org/keywords/context","display_name":"Context (archaeology)","score":0.5778999924659729},{"id":"https://openalex.org/keywords/representation","display_name":"Representation (politics)","score":0.5414999723434448},{"id":"https://openalex.org/keywords/tensor","display_name":"Tensor (intrinsic definition)","score":0.5242000222206116},{"id":"https://openalex.org/keywords/scheme","display_name":"Scheme (mathematics)","score":0.5080999732017517},{"id":"https://openalex.org/keywords/encoding","display_name":"Encoding (memory)","score":0.45649999380111694},{"id":"https://openalex.org/keywords/dimension","display_name":"Dimension (graph theory)","score":0.44369998574256897},{"id":"https://openalex.org/keywords/sequence","display_name":"Sequence (biology)","score":0.40779998898506165},{"id":"https://openalex.org/keywords/parallelism","display_name":"Parallelism (grammar)","score":0.39309999346733093}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7192999720573425},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5792999863624573},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.5778999924659729},{"id":"https://openalex.org/C2776359362","wikidata":"https://www.wikidata.org/wiki/Q2145286","display_name":"Representation (politics)","level":3,"score":0.5414999723434448},{"id":"https://openalex.org/C155281189","wikidata":"https://www.wikidata.org/wiki/Q3518150","display_name":"Tensor (intrinsic definition)","level":2,"score":0.5242000222206116},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.522599995136261},{"id":"https://openalex.org/C77618280","wikidata":"https://www.wikidata.org/wiki/Q1155772","display_name":"Scheme (mathematics)","level":2,"score":0.5080999732017517},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.45649999380111694},{"id":"https://openalex.org/C33676613","wikidata":"https://www.wikidata.org/wiki/Q13415176","display_name":"Dimension (graph theory)","level":2,"score":0.44369998574256897},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.4309999942779541},{"id":"https://openalex.org/C2778112365","wikidata":"https://www.wikidata.org/wiki/Q3511065","display_name":"Sequence (biology)","level":2,"score":0.40779998898506165},{"id":"https://openalex.org/C2781172179","wikidata":"https://www.wikidata.org/wiki/Q853109","display_name":"Parallelism (grammar)","level":2,"score":0.39309999346733093},{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.38989999890327454},{"id":"https://openalex.org/C60292330","wikidata":"https://www.wikidata.org/wiki/Q1014065","display_name":"Hadamard transform","level":2,"score":0.3433000147342682},{"id":"https://openalex.org/C192209626","wikidata":"https://www.wikidata.org/wiki/Q190909","display_name":"Focus (optics)","level":2,"score":0.33489999175071716},{"id":"https://openalex.org/C170133592","wikidata":"https://www.wikidata.org/wiki/Q1806883","display_name":"Latent semantic analysis","level":2,"score":0.3043000102043152},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.303600013256073},{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.2985999882221222},{"id":"https://openalex.org/C7366592","wikidata":"https://www.wikidata.org/wiki/Q1255620","display_name":"Dram","level":2,"score":0.27869999408721924},{"id":"https://openalex.org/C51167844","wikidata":"https://www.wikidata.org/wiki/Q4422623","display_name":"Latent variable","level":2,"score":0.275299996137619},{"id":"https://openalex.org/C159254197","wikidata":"https://www.wikidata.org/wiki/Q1144915","display_name":"Lexicographical order","level":2,"score":0.26409998536109924},{"id":"https://openalex.org/C2776190703","wikidata":"https://www.wikidata.org/wiki/Q488148","display_name":"Slicing","level":2,"score":0.2621000111103058},{"id":"https://openalex.org/C97931131","wikidata":"https://www.wikidata.org/wiki/Q5282087","display_name":"Discriminative model","level":2,"score":0.25920000672340393},{"id":"https://openalex.org/C2781020372","wikidata":"https://www.wikidata.org/wiki/Q533093","display_name":"On the fly","level":2,"score":0.257099986076355},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.25200000405311584}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1145/3779212.3790237","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3779212.3790237","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 31st ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 2","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.1145/3779212.3790237","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3779212.3790237","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 31st ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 2","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":7,"referenced_works":["https://openalex.org/W2890894339","https://openalex.org/W2946609015","https://openalex.org/W2969388332","https://openalex.org/W2998617917","https://openalex.org/W3194676777","https://openalex.org/W4390690464","https://openalex.org/W4396815229"],"related_works":[],"abstract_inverted_index":{"Multi-Head":[0],"Latent":[1],"Attention":[2,51],"(MLA),":[3],"introduced":[4],"in":[5,89],"DeepSeek-V2,":[6],"compresses":[7],"key\u2013value":[8],"states":[9],"into":[10],"a":[11,56,142,182],"low-rank":[12],"latent":[13,62,96,135],"vector":[14,19],"cKV,":[15,42],"caching":[16],"only":[17],"this":[18],"to":[20,93,156],"reduce":[21],"memory.":[22],"In":[23],"tensor":[24],"parallelism":[25],"(TP),":[26],"however,":[27],"attention":[28,72,87],"heads":[29],"are":[30],"computed":[31],"across":[32,69,137],"multiple":[33],"devices,":[34,70],"and":[35,64,77,123,127,154,172,177,191],"each":[36,65,75],"device":[37],"must":[38],"load":[39],"the":[40,44,61,79,94,104,148,166],"full":[41,95],"eroding":[43],"advantage":[45],"of":[46,121,201],"MLA":[47,114,149],"over":[48],"Grouped":[49],"Query":[50],"(GQA).":[52],"We":[53],"present":[54],"TPLA,":[55],"scheme":[57,145],"that":[58,146],"partitions":[59],"both":[60],"representation":[63],"head's":[66],"input":[67],"dimension":[68],"performs":[71],"independently":[73],"on":[74,189,199],"shard,":[76],"aggregates":[78],"results":[80],"with":[81,113,125],"an":[82],"all-reduce.":[83],"Unlike":[84],"GLA,":[85],"every":[86],"head":[88],"TPLA":[90,110,157,194],"still":[91],"attends":[92],"space,":[97],"preserving":[98],"MLA's":[99],"representational":[100],"capacity":[101],"while":[102,186],"reducing":[103,165],"per-device":[105,167],"KV":[106,168],"cache.":[107],"To":[108],"make":[109],"drop-in":[111],"compatible":[112],"checkpoints,":[115],"we":[116,140,174],"further":[117,197],"derive":[118],"orthogonal":[119],"reparameterizations":[120],"RMSNorm":[122],"softmax---instantiated":[124],"Hadamard":[126],"PCA":[128],"transforms---that":[129],"mitigate":[130],"cross-shard":[131],"discrepancies":[132],"when":[133],"slicing":[134],"vectors":[136],"devices.":[138],"Finally,":[139],"introduce":[141],"prefill-decode":[143],"separation":[144],"keeps":[147],"form":[150],"during":[151,158],"compute-bound":[152],"prefilling":[153],"switches":[155],"memory-bound":[159],"decoding,":[160],"minimizing":[161],"conversion-induced":[162],"error.":[163],"By":[164],"cache":[169],"for":[170],"DeepSeek-V3":[171],"Kimi-K2,":[173],"achieve":[175],"1.79x)":[176],"1.93)":[178],"speedups":[179],"respectively,":[180],"at":[181],"32K-token":[183],"context":[184],"length":[185],"maintaining":[187],"accuracy":[188],"commonsense":[190],"LongBench":[192],"benchmarks.":[193],"can":[195],"be":[196],"implemented":[198],"top":[200],"FlashAttention-3,":[202],"enabling":[203],"practical":[204],"end-to-end":[205],"acceleration.":[206]},"counts_by_year":[],"updated_date":"2026-03-27T05:58:40.876381","created_date":"2026-03-12T00:00:00"}