{"id":"https://openalex.org/W7130570081","doi":"https://doi.org/10.1109/apccas67402.2025.11378253","title":"RoPE on the Fast Track: Latency-Optimized LLM Inference on GPUs with Low-Rank QKV Factorization and Rotary Positional Embedding","display_name":"RoPE on the Fast Track: Latency-Optimized LLM Inference on GPUs with Low-Rank QKV Factorization and Rotary Positional Embedding","publication_year":2025,"publication_date":"2025-10-12","ids":{"openalex":"https://openalex.org/W7130570081","doi":"https://doi.org/10.1109/apccas67402.2025.11378253"},"language":null,"primary_location":{"id":"doi:10.1109/apccas67402.2025.11378253","is_oa":false,"landing_page_url":"https://doi.org/10.1109/apccas67402.2025.11378253","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2025 IEEE Asia Pacific Conference on Circuits and Systems (APCCAS)","raw_type":"proceedings-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":false,"oa_status":"closed","oa_url":null,"any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5126367865","display_name":"Gilhyeon Lee","orcid":null},"institutions":[{"id":"https://openalex.org/I118373667","display_name":"Seoul National University of Science and Technology","ror":"https://ror.org/00chfja07","country_code":"KR","type":"education","lineage":["https://openalex.org/I118373667"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Gilhyeon Lee","raw_affiliation_strings":["Seoul National University of Science and Technology,Department of Electrical and Information Engineering and Research Center for Electrical and Information Technology,Seoul,Korea"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Seoul National University of Science and Technology,Department of Electrical and Information Engineering and Research Center for Electrical and Information Technology,Seoul,Korea","institution_ids":["https://openalex.org/I118373667"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5126439231","display_name":"Jihoon Jang","orcid":null},"institutions":[{"id":"https://openalex.org/I118373667","display_name":"Seoul National University of Science and Technology","ror":"https://ror.org/00chfja07","country_code":"KR","type":"education","lineage":["https://openalex.org/I118373667"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Jihoon Jang","raw_affiliation_strings":["Seoul National University of Science and Technology,Department of Electrical and Information Engineering and Research Center for Electrical and Information Technology,Seoul,Korea"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Seoul National University of Science and Technology,Department of Electrical and Information Engineering and Research Center for Electrical and Information Technology,Seoul,Korea","institution_ids":["https://openalex.org/I118373667"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5126440595","display_name":"Kyungmin Goh","orcid":null},"institutions":[{"id":"https://openalex.org/I118373667","display_name":"Seoul National University of Science and Technology","ror":"https://ror.org/00chfja07","country_code":"KR","type":"education","lineage":["https://openalex.org/I118373667"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Kyungmin Goh","raw_affiliation_strings":["Seoul National University of Science and Technology,Department of Electrical and Information Engineering and Research Center for Electrical and Information Technology,Seoul,Korea"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Seoul National University of Science and Technology,Department of Electrical and Information Engineering and Research Center for Electrical and Information Technology,Seoul,Korea","institution_ids":["https://openalex.org/I118373667"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5126427575","display_name":"Hyun Kim","orcid":null},"institutions":[{"id":"https://openalex.org/I118373667","display_name":"Seoul National University of Science and Technology","ror":"https://ror.org/00chfja07","country_code":"KR","type":"education","lineage":["https://openalex.org/I118373667"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Hyun Kim","raw_affiliation_strings":["Seoul National University of Science and Technology,Department of Electrical and Information Engineering and Research Center for Electrical and Information Technology,Seoul,Korea"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Seoul National University of Science and Technology,Department of Electrical and Information Engineering and Research Center for Electrical and Information Technology,Seoul,Korea","institution_ids":["https://openalex.org/I118373667"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I118373667"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.81071469,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"1","last_page":"5"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.20649999380111694,"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":0.20649999380111694,"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/T10201","display_name":"Speech Recognition and Synthesis","score":0.15389999747276306,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.14409999549388885,"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/latency","display_name":"Latency (audio)","score":0.6241000294685364},{"id":"https://openalex.org/keywords/embedding","display_name":"Embedding","score":0.5784000158309937},{"id":"https://openalex.org/keywords/decoding-methods","display_name":"Decoding methods","score":0.5127000212669373},{"id":"https://openalex.org/keywords/execution-time","display_name":"Execution time","score":0.5105000138282776},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.5099999904632568},{"id":"https://openalex.org/keywords/factorization","display_name":"Factorization","score":0.4507000148296356},{"id":"https://openalex.org/keywords/normalization","display_name":"Normalization (sociology)","score":0.4433000087738037},{"id":"https://openalex.org/keywords/limiting","display_name":"Limiting","score":0.43790000677108765}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8195000290870667},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.7318999767303467},{"id":"https://openalex.org/C82876162","wikidata":"https://www.wikidata.org/wiki/Q17096504","display_name":"Latency (audio)","level":2,"score":0.6241000294685364},{"id":"https://openalex.org/C41608201","wikidata":"https://www.wikidata.org/wiki/Q980509","display_name":"Embedding","level":2,"score":0.5784000158309937},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.5127000212669373},{"id":"https://openalex.org/C2989134064","wikidata":"https://www.wikidata.org/wiki/Q288510","display_name":"Execution time","level":2,"score":0.5105000138282776},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5099999904632568},{"id":"https://openalex.org/C187834632","wikidata":"https://www.wikidata.org/wiki/Q188804","display_name":"Factorization","level":2,"score":0.4507000148296356},{"id":"https://openalex.org/C136886441","wikidata":"https://www.wikidata.org/wiki/Q926129","display_name":"Normalization (sociology)","level":2,"score":0.4433000087738037},{"id":"https://openalex.org/C188198153","wikidata":"https://www.wikidata.org/wiki/Q1613840","display_name":"Limiting","level":2,"score":0.43790000677108765},{"id":"https://openalex.org/C74912251","wikidata":"https://www.wikidata.org/wiki/Q6815727","display_name":"Memory footprint","level":2,"score":0.4049000144004822},{"id":"https://openalex.org/C2779960059","wikidata":"https://www.wikidata.org/wiki/Q7113681","display_name":"Overhead (engineering)","level":2,"score":0.4041999876499176},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.39489999413490295},{"id":"https://openalex.org/C162269090","wikidata":"https://www.wikidata.org/wiki/Q1156047","display_name":"Rope","level":2,"score":0.37119999527931213},{"id":"https://openalex.org/C2781357197","wikidata":"https://www.wikidata.org/wiki/Q5757597","display_name":"High memory","level":2,"score":0.35010001063346863},{"id":"https://openalex.org/C42935608","wikidata":"https://www.wikidata.org/wiki/Q190411","display_name":"Field-programmable gate array","level":2,"score":0.3400999903678894},{"id":"https://openalex.org/C2776834041","wikidata":"https://www.wikidata.org/wiki/Q25346349","display_name":"Execution model","level":2,"score":0.32120001316070557},{"id":"https://openalex.org/C175291020","wikidata":"https://www.wikidata.org/wiki/Q1156822","display_name":"Offset (computer science)","level":2,"score":0.3199000060558319},{"id":"https://openalex.org/C63540848","wikidata":"https://www.wikidata.org/wiki/Q3140932","display_name":"Fault tolerance","level":2,"score":0.3140000104904175},{"id":"https://openalex.org/C3826847","wikidata":"https://www.wikidata.org/wiki/Q188768","display_name":"FLOPS","level":2,"score":0.3059999942779541},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.2865999937057495},{"id":"https://openalex.org/C33676613","wikidata":"https://www.wikidata.org/wiki/Q13415176","display_name":"Dimension (graph theory)","level":2,"score":0.26570001244544983},{"id":"https://openalex.org/C149635348","wikidata":"https://www.wikidata.org/wiki/Q193040","display_name":"Embedded system","level":1,"score":0.26570001244544983},{"id":"https://openalex.org/C165005293","wikidata":"https://www.wikidata.org/wiki/Q1074500","display_name":"Chip","level":2,"score":0.2606000006198883},{"id":"https://openalex.org/C42355184","wikidata":"https://www.wikidata.org/wiki/Q1361088","display_name":"Matrix decomposition","level":3,"score":0.2538999915122986}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/apccas67402.2025.11378253","is_oa":false,"landing_page_url":"https://doi.org/10.1109/apccas67402.2025.11378253","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2025 IEEE Asia Pacific Conference on Circuits and Systems (APCCAS)","raw_type":"proceedings-article"}],"best_oa_location":null,"sustainable_development_goals":[],"awards":[{"id":"https://openalex.org/G284294426","display_name":null,"funder_award_id":"2020M3H2A1078119","funder_id":"https://openalex.org/F4320322120","funder_display_name":"National Research Foundation of Korea"}],"funders":[{"id":"https://openalex.org/F4320321681","display_name":"Ministry of Trade, Industry and Energy","ror":"https://ror.org/008nkqk13"},{"id":"https://openalex.org/F4320322120","display_name":"National Research Foundation of Korea","ror":"https://ror.org/013aysd81"}],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":16,"referenced_works":["https://openalex.org/W2122080725","https://openalex.org/W2954698171","https://openalex.org/W2964110616","https://openalex.org/W3133730715","https://openalex.org/W3189166979","https://openalex.org/W4285225959","https://openalex.org/W4367043644","https://openalex.org/W4385245566","https://openalex.org/W4385570645","https://openalex.org/W4388979610","https://openalex.org/W4393318691","https://openalex.org/W4393949386","https://openalex.org/W4401402799","https://openalex.org/W4406892532","https://openalex.org/W4409307140","https://openalex.org/W4410949500"],"related_works":[],"abstract_inverted_index":{"In":[0],"large":[1],"language":[2],"model":[3,52,165],"(LLM)":[4],"inference,":[5],"rotary":[6],"positional":[7],"embedding":[8],"(RoPE)":[9],"contributes":[10],"significantly":[11],"to":[12,17,65,99,114,136,199],"overall":[13],"decoding":[14,173],"latency":[15,174],"due":[16],"accumulated":[18],"overhead":[19],"from":[20],"numerous":[21],"micro-kernel":[22],"invocations":[23],"and":[24,30,43,85,110,139,186],"its":[25,141],"sequential":[26],"execution":[27,143,154],"after":[28],"Q":[29],"K":[31],"generation,":[32],"which":[33],"prevents":[34],"effective":[35],"parallelization.":[36],"Additionally,":[37],"the":[38,51,69,100,117,148,200],"<tex":[39,95,178,187],"xmlns:mml=\"http://www.w3.org/1998/Math/MathML\"":[40,96,179,188],"xmlns:xlink=\"http://www.w3.org/1999/xlink\">$\\mathrm{Q},":[41],"\\mathrm{K}$</tex>,":[42],"V":[44],"(QKV)":[45],"generation":[46,102],"weights":[47,138],"in":[48,60,68,122],"modern":[49],"LLMs-where":[50],"dimension":[53],"reaches":[54],"several":[55,58],"thou-sand-are":[56],"each":[57],"gigabytes":[59],"size,":[61],"making":[62],"it":[63,135],"impractical":[64],"store":[66],"them":[67],"on-chip":[70],"memory":[71,105],"of":[72,116],"FPGAs":[73],"or":[74],"custom":[75],"ASICs.":[76],"This":[77,150],"necessitates":[78],"repeated":[79],"off-chip":[80],"accesses,":[81],"limiting":[82],"hardware":[83],"utilization":[84],"inference":[86],"speed.":[87],"To":[88],"address":[89],"these":[90],"bottlenecks,":[91],"we":[92,125],"apply":[93],"rank":[94],"xmlns:xlink=\"http://www.w3.org/1999/xlink\">$=8$</tex>":[97],"factorization":[98],"QKV":[101],"weights,":[103],"reducing":[104],"usage":[106],"by":[107,133,156,177],"over":[108],"99%":[109],"lowering":[111],"per-token":[112],"FLOPs":[113],"1.41%":[115],"baseline,":[118],"with":[119,144],"minimal":[120],"loss":[121],"accuracy.":[123],"Furthermore,":[124],"reformulate":[126],"RoPE":[127,131],"as":[128],"a":[129,163],"Weight-side":[130],"mechanism":[132],"pre-applying":[134],"low-rank":[137],"enabling":[140,157],"parallel":[142],"RMS":[145],"normalization":[146],"on":[147,166],"GPU.":[149],"structural":[151],"change":[152],"improves":[153],"efficiency":[155],"concurrent":[158],"computation.":[159],"Experimental":[160],"results":[161],"using":[162],"Llama3based":[164],"an":[167],"RTX":[168],"3090":[169],"GPU":[170],"show":[171],"that":[172],"is":[175],"reduced":[176],"xmlns:xlink=\"http://www.w3.org/1999/xlink\">$1.14":[180],"\\times$</tex>":[181,190],"for":[182,191],"batch":[183,192],"size":[184,193],"1":[185],"xmlns:xlink=\"http://www.w3.org/1999/xlink\">$1.17":[189],"2,":[194],"while":[195],"maintaining":[196],"comparable":[197],"accuracy":[198],"baseline.":[201]},"counts_by_year":[],"updated_date":"2026-06-26T08:34:08.712188","created_date":"2026-02-20T00:00:00"}