{"id":"https://openalex.org/W4407848445","doi":"https://doi.org/10.1145/3696443.3708944","title":"CUrator: An Efficient LLM Execution Engine with Optimized Integration of CUDA Libraries","display_name":"CUrator: An Efficient LLM Execution Engine with Optimized Integration of CUDA Libraries","publication_year":2025,"publication_date":"2025-02-22","ids":{"openalex":"https://openalex.org/W4407848445","doi":"https://doi.org/10.1145/3696443.3708944"},"language":"en","primary_location":{"id":"doi:10.1145/3696443.3708944","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3696443.3708944","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 23rd ACM/IEEE International Symposium on Code Generation and Optimization","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://doi.org/10.1145/3696443.3708944","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5048932828","display_name":"Y. Lee","orcid":null},"institutions":[{"id":"https://openalex.org/I193775966","display_name":"Yonsei University","ror":"https://ror.org/01wjejq96","country_code":"KR","type":"education","lineage":["https://openalex.org/I193775966"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Yoon Noh Lee","raw_affiliation_strings":["Yonsei University, Seoul, Republic of Korea"],"raw_orcid":"https://orcid.org/0009-0002-4021-1091","affiliations":[{"raw_affiliation_string":"Yonsei University, Seoul, Republic of Korea","institution_ids":["https://openalex.org/I193775966"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5085503041","display_name":"Yongseung Yu","orcid":"https://orcid.org/0000-0001-9838-4310"},"institutions":[{"id":"https://openalex.org/I193775966","display_name":"Yonsei University","ror":"https://ror.org/01wjejq96","country_code":"KR","type":"education","lineage":["https://openalex.org/I193775966"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Yongseung Yu","raw_affiliation_strings":["Yonsei University, Seoul, Republic of Korea"],"raw_orcid":"https://orcid.org/0000-0001-9838-4310","affiliations":[{"raw_affiliation_string":"Yonsei University, Seoul, Republic of Korea","institution_ids":["https://openalex.org/I193775966"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5086980677","display_name":"Yongjun Park","orcid":"https://orcid.org/0000-0003-3725-0380"},"institutions":[{"id":"https://openalex.org/I193775966","display_name":"Yonsei University","ror":"https://ror.org/01wjejq96","country_code":"KR","type":"education","lineage":["https://openalex.org/I193775966"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Yongjun Park","raw_affiliation_strings":["Yonsei University, Seoul, Republic of Korea"],"raw_orcid":"https://orcid.org/0000-0003-3725-0380","affiliations":[{"raw_affiliation_string":"Yonsei University, Seoul, Republic of Korea","institution_ids":["https://openalex.org/I193775966"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I193775966"],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"cited_by_count":2,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"209","last_page":"224"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11181","display_name":"Advanced Data Storage Technologies","score":0.9944999814033508,"subfield":{"id":"https://openalex.org/subfields/1705","display_name":"Computer Networks and Communications"},"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/T11181","display_name":"Advanced Data Storage Technologies","score":0.9944999814033508,"subfield":{"id":"https://openalex.org/subfields/1705","display_name":"Computer Networks and Communications"},"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/T10715","display_name":"Distributed and Parallel Computing Systems","score":0.9912999868392944,"subfield":{"id":"https://openalex.org/subfields/1705","display_name":"Computer Networks and Communications"},"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":0.9847999811172485,"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/cuda","display_name":"CUDA","score":0.8805674910545349},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.706504762172699},{"id":"https://openalex.org/keywords/parallel-computing","display_name":"Parallel computing","score":0.47962531447410583},{"id":"https://openalex.org/keywords/computer-architecture","display_name":"Computer architecture","score":0.47286009788513184}],"concepts":[{"id":"https://openalex.org/C2778119891","wikidata":"https://www.wikidata.org/wiki/Q477690","display_name":"CUDA","level":2,"score":0.8805674910545349},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.706504762172699},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.47962531447410583},{"id":"https://openalex.org/C118524514","wikidata":"https://www.wikidata.org/wiki/Q173212","display_name":"Computer architecture","level":1,"score":0.47286009788513184}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1145/3696443.3708944","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3696443.3708944","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 23rd ACM/IEEE International Symposium on Code Generation and Optimization","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.1145/3696443.3708944","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3696443.3708944","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 23rd ACM/IEEE International Symposium on Code Generation and Optimization","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":21,"referenced_works":["https://openalex.org/W1965628535","https://openalex.org/W2100218206","https://openalex.org/W2122080725","https://openalex.org/W2580538010","https://openalex.org/W2804500013","https://openalex.org/W2953264125","https://openalex.org/W2963989532","https://openalex.org/W2990138404","https://openalex.org/W3012249773","https://openalex.org/W3014127403","https://openalex.org/W3103018381","https://openalex.org/W3105696640","https://openalex.org/W3208285274","https://openalex.org/W4221062244","https://openalex.org/W4307789874","https://openalex.org/W4321636575","https://openalex.org/W4385438314","https://openalex.org/W4390097705","https://openalex.org/W4393949386","https://openalex.org/W4396822035","https://openalex.org/W6912037511"],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2899084033","https://openalex.org/W2748952813","https://openalex.org/W4399354997","https://openalex.org/W3062287","https://openalex.org/W2380390332","https://openalex.org/W2005148983","https://openalex.org/W2012954338","https://openalex.org/W2096672917","https://openalex.org/W2392023973"],"abstract_inverted_index":{"Large":[0],"Language":[1],"Models":[2],"(LLMs)":[3],"have":[4,118,128],"recently":[5],"emerged":[6],"as":[7,49,167],"a":[8,13,54,196],"state-of-the-art":[9],"learning":[10,113],"model":[11,122],"with":[12,83,207],"wide":[14],"range":[15],"of":[16,56,98,105,133,178,210,266,310],"applications":[17],"in":[18,64,101,203,278],"diverse":[19,134],"computing":[20,66,136],"environments.":[21,67,137],"Among":[22],"the":[23,29,31,38,44,70,77,84,96,103,131,182,190,204,208,247,275,279,286,292,297,306],"various":[24,111,179,311,314],"computational":[25],"operations":[26],"that":[27,147,261,291],"comprise":[28],"LLM,":[30],"GEneral":[32],"Matrix":[33],"Multiplication":[34],"(GEMM)":[35],"operation":[36,42],"is":[37],"most":[39],"frequently":[40],"utilized":[41],"within":[43,76,217],"LLM.":[45,219],"GEMM":[46,62],"libraries":[47,159],"such":[48,166],"cuBLAS":[50,156,253],"and":[51,157,170,236,254,268,281],"CUTLASS":[52,71,158,191,205,233,241,255],"provide":[53,296],"variety":[55],"optimization":[57,224],"techniques":[58],"to":[59,86,185,285],"achieve":[60,149],"optimal":[61,150,212],"performance":[63,104,153,309],"GPU-enabled":[65],"In":[68],"particular,":[69],"open-source":[72],"library":[73,206],"for":[74,89,124,163,199,214,232,240,271,300],"GPUs":[75,162],"CUDA":[78],"programming":[79],"environment":[80],"provides":[81],"users":[82],"capability":[85],"optimize":[87],"templates":[88],"high":[90],"performance.":[91],"Previous":[92],"research":[93],"has":[94],"demonstrated":[95],"effectiveness":[97],"CUTLASS-based":[99],"GEMMs":[100,216],"improving":[102],"real-world":[106],"deep":[107,112],"neural":[108],"networks":[109],"on":[110,160,181,274,313],"platforms.":[114],"However,":[115],"these":[116],"studies":[117],"not":[119],"considered":[120],"different":[121,161],"parameters":[123,202],"modern":[125,164],"LLMs":[126,165,180,273,312],"nor":[127],"they":[129],"explored":[130],"impact":[132],"GPU":[135,277],"This":[138],"paper":[139],"presents":[140],"CUrator,":[141],"an":[142],"efficient":[143],"LLM":[144,152],"execution":[145],"engine":[146],"can":[148,295],"end-to-end":[151,308],"using":[154],"both":[155],"BERT,":[168],"GPT,":[169],"Llama.":[171],"CUrator":[172,220,245,262,293],"first":[173],"generates":[174],"CUTLASS-/cuBLAS-friendly":[175],"graph":[176],"IRs":[177],"TVM":[183],"framework":[184,294],"maximize":[186],"mapping":[187,192,250],"coverage.":[188],"On":[189],"path,":[193],"it":[194],"performs":[195],"comprehensive":[197],"search":[198],"programmable":[200],"tuning":[201,302],"objective":[209],"deriving":[211],"kernels":[213],"all":[215],"each":[218],"further":[221],"introduces":[222],"two":[223],"techniques:":[225],"1)":[226],"build-time":[227],"reduction":[228],"key":[229],"initialization":[230],"support":[231,239],"Split-K":[234,238],"GEMMs,":[235],"2)":[237],"Batch":[242],"GEMMs.":[243],"Finally,":[244],"selects":[246],"best":[248,298],"performing":[249],"path":[251],"between":[252],"paths.":[256],"The":[257],"experimental":[258],"results":[259],"show":[260],"achieves":[263],"inference":[264],"speedups":[265],"1.50\u00d7":[267],"4.99\u00d7,":[269],"respectively,":[270],"representative":[272],"A100":[276],"single":[280],"half":[282],"precision,":[283],"compared":[284],"baseline.":[287],"We":[288],"strongly":[289],"believe":[290],"direction":[299],"next-generation":[301],"frameworks":[303],"by":[304],"showing":[305],"maximum":[307],"GPUs.":[315]},"counts_by_year":[{"year":2026,"cited_by_count":1},{"year":2025,"cited_by_count":1}],"updated_date":"2026-07-14T23:27:15.235271","created_date":"2025-10-10T00:00:00"}