{"id":"https://openalex.org/W7161024485","doi":"https://doi.org/10.48550/arxiv.2605.11581","title":"Ada-MK: Adaptive MegaKernel Optimization via Automated DAG-based Search for LLM Inference","display_name":"Ada-MK: Adaptive MegaKernel Optimization via Automated DAG-based Search for LLM Inference","publication_year":2026,"publication_date":"2026-05-12","ids":{"openalex":"https://openalex.org/W7161024485","doi":"https://doi.org/10.48550/arxiv.2605.11581"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.11581","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.11581","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,"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.2605.11581","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5043102204","display_name":"Wenxin Dong","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Dong, Wenxin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136073346","display_name":"Mingqing Hu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hu, Mingqing","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136040482","display_name":"Guanghui Yu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yu, Guanghui","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136007517","display_name":"Qiang Fu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Fu, Qiang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136018204","display_name":"Peng Xu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xu, Peng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136039747","display_name":"Hui Xu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xu, Hui","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136031941","display_name":"Yue Xing","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xing, Yue","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5006865469","display_name":"Xuewu Jiao","orcid":"https://orcid.org/0009-0004-6530-4774"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jiao, Xuewu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136081963","display_name":"Shuanglong Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Shuanglong","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5136032779","display_name":"Lin Liu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Lin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"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/T10036","display_name":"Advanced Neural Network Applications","score":0.31299999356269836,"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":0.31299999356269836,"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/T14347","display_name":"Big Data and Digital Economy","score":0.07649999856948853,"subfield":{"id":"https://openalex.org/subfields/1710","display_name":"Information Systems"},"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/T11704","display_name":"Mobile Crowdsensing and Crowdsourcing","score":0.0714000016450882,"subfield":{"id":"https://openalex.org/subfields/1706","display_name":"Computer Science Applications"},"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":0.5519999861717224},{"id":"https://openalex.org/keywords/plug-in","display_name":"Plug-in","score":0.5090000033378601},{"id":"https://openalex.org/keywords/software-portability","display_name":"Software portability","score":0.47929999232292175},{"id":"https://openalex.org/keywords/scheduling","display_name":"Scheduling (production processes)","score":0.474700003862381},{"id":"https://openalex.org/keywords/latency","display_name":"Latency (audio)","score":0.4361000061035156},{"id":"https://openalex.org/keywords/software-deployment","display_name":"Software deployment","score":0.41179999709129333},{"id":"https://openalex.org/keywords/kernel","display_name":"Kernel (algebra)","score":0.39739999175071716},{"id":"https://openalex.org/keywords/overhead","display_name":"Overhead (engineering)","score":0.39469999074935913},{"id":"https://openalex.org/keywords/asynchronous-communication","display_name":"Asynchronous communication","score":0.3937000036239624}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8492000102996826},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.5541999936103821},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5519999861717224},{"id":"https://openalex.org/C4924752","wikidata":"https://www.wikidata.org/wiki/Q184148","display_name":"Plug-in","level":2,"score":0.5090000033378601},{"id":"https://openalex.org/C63000827","wikidata":"https://www.wikidata.org/wiki/Q3080428","display_name":"Software portability","level":2,"score":0.47929999232292175},{"id":"https://openalex.org/C206729178","wikidata":"https://www.wikidata.org/wiki/Q2271896","display_name":"Scheduling (production processes)","level":2,"score":0.474700003862381},{"id":"https://openalex.org/C82876162","wikidata":"https://www.wikidata.org/wiki/Q17096504","display_name":"Latency (audio)","level":2,"score":0.4361000061035156},{"id":"https://openalex.org/C105339364","wikidata":"https://www.wikidata.org/wiki/Q2297740","display_name":"Software deployment","level":2,"score":0.41179999709129333},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.40700000524520874},{"id":"https://openalex.org/C74193536","wikidata":"https://www.wikidata.org/wiki/Q574844","display_name":"Kernel (algebra)","level":2,"score":0.39739999175071716},{"id":"https://openalex.org/C2779960059","wikidata":"https://www.wikidata.org/wiki/Q7113681","display_name":"Overhead (engineering)","level":2,"score":0.39469999074935913},{"id":"https://openalex.org/C151319957","wikidata":"https://www.wikidata.org/wiki/Q752739","display_name":"Asynchronous communication","level":2,"score":0.3937000036239624},{"id":"https://openalex.org/C15569618","wikidata":"https://www.wikidata.org/wiki/Q3561421","display_name":"Liveness","level":2,"score":0.3822000026702881},{"id":"https://openalex.org/C26713055","wikidata":"https://www.wikidata.org/wiki/Q245962","display_name":"Implementation","level":2,"score":0.3601999878883362},{"id":"https://openalex.org/C169590947","wikidata":"https://www.wikidata.org/wiki/Q47506","display_name":"Compiler","level":2,"score":0.3587999939918518},{"id":"https://openalex.org/C198370458","wikidata":"https://www.wikidata.org/wiki/Q586459","display_name":"Type inference","level":3,"score":0.32519999146461487},{"id":"https://openalex.org/C149635348","wikidata":"https://www.wikidata.org/wiki/Q193040","display_name":"Embedded system","level":1,"score":0.30970001220703125},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.2994999885559082},{"id":"https://openalex.org/C34388435","wikidata":"https://www.wikidata.org/wiki/Q2267362","display_name":"Bounded function","level":2,"score":0.2980000078678131},{"id":"https://openalex.org/C46743427","wikidata":"https://www.wikidata.org/wiki/Q1341685","display_name":"Inference engine","level":3,"score":0.2939000129699707},{"id":"https://openalex.org/C157764524","wikidata":"https://www.wikidata.org/wiki/Q1383412","display_name":"Throughput","level":3,"score":0.28529998660087585},{"id":"https://openalex.org/C193702766","wikidata":"https://www.wikidata.org/wiki/Q1414548","display_name":"Concurrency","level":2,"score":0.27630001306533813},{"id":"https://openalex.org/C124304363","wikidata":"https://www.wikidata.org/wiki/Q673661","display_name":"Abstraction","level":2,"score":0.27619999647140503},{"id":"https://openalex.org/C79403827","wikidata":"https://www.wikidata.org/wiki/Q3988","display_name":"Real-time computing","level":1,"score":0.2757999897003174},{"id":"https://openalex.org/C43521106","wikidata":"https://www.wikidata.org/wiki/Q2165493","display_name":"Pipeline (software)","level":2,"score":0.27399998903274536},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.27000001072883606},{"id":"https://openalex.org/C199360897","wikidata":"https://www.wikidata.org/wiki/Q9143","display_name":"Programming language","level":1,"score":0.26930001378059387},{"id":"https://openalex.org/C2779639559","wikidata":"https://www.wikidata.org/wiki/Q7661178","display_name":"Symbolic execution","level":3,"score":0.2689000070095062},{"id":"https://openalex.org/C110251889","wikidata":"https://www.wikidata.org/wiki/Q1569697","display_name":"Model checking","level":2,"score":0.26829999685287476},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.2623000144958496},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.25600001215934753}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.11581","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.11581","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,"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.2605.11581","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.11581","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,"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":{"When":[0],"large":[1],"language":[2],"models":[3],"(LLMs)":[4],"serve":[5],"real-time":[6],"inference":[7,47,186],"in":[8,108,233],"commercial":[9,235],"online":[10,236],"advertising":[11,237],"systems,":[12],"end-to-end":[13,46],"latency":[14],"must":[15],"be":[16,134],"strictly":[17],"bounded":[18],"to":[19,90,137,212],"the":[20,28,119,173],"millisecond":[21],"range.":[22],"Yet":[23],"every":[24],"token":[25],"generated":[26],"during":[27],"decode":[29],"phase":[30],"triggers":[31],"thousands":[32],"of":[33,45,123,231],"kernel":[34,37],"launches,":[35],"and":[36,53,76,93,129,181,217],"launch":[38,51],"overhead":[39,52],"alone":[40],"can":[41,133],"account":[42],"for":[43],"14.6%":[44],"time.":[48,139],"MegaKernel":[49,68,125,190,232],"eliminates":[50],"inter-operator":[54],"HBM":[55],"round-trips":[56],"by":[57,163,210],"fusing":[58],"multiple":[59],"operators":[60],"into":[61,194],"a":[62,71,115,124,148,183,192,234],"single":[63],"persistent":[64],"kernel.":[65],"However,":[66],"existing":[67],"implementations":[69],"face":[70],"fundamental":[72],"tension":[73],"between":[74],"portability":[75],"efficiency":[77],"on":[78,141],"resource-constrained":[79],"GPUs":[80],"such":[81],"as":[82,191],"NVIDIA":[83,204],"Ada:":[84],"hand-tuned":[85],"solutions":[86],"are":[87,106],"tightly":[88],"coupled":[89],"specific":[91],"architectures":[92],"lack":[94],"portability,":[95],"while":[96],"auto-compiled":[97],"approaches":[98],"introduce":[99],"runtime":[100,130,179],"dynamic":[101,131],"scheduling":[102],"whose":[103],"branch":[104],"penalties":[105],"unacceptable":[107],"latency-critical":[109],"settings.":[110],"We":[111],"observe":[112],"that":[113,157,171,188],"under":[114],"fixed":[116],"deployment":[117,230],"configuration,":[118],"optimal":[120,174],"execution":[121,175],"path":[122],"is":[126],"uniquely":[127],"determined,":[128],"decision-making":[132],"entirely":[135],"hoisted":[136],"compile":[138],"Building":[140],"this":[142],"insight,":[143],"we":[144],"propose":[145],"Ada-MK:":[146],"(1)":[147],"three-dimensional":[149],"shared-memory":[150],"constraint":[151],"model":[152],"combined":[153],"with":[154,199],"K-dimension":[155],"splitting":[156],"reduces":[158],"peak":[159],"shared":[160],"memory":[161],"usage":[162],"50%;":[164],"(2)":[165],"MLIR-based":[166],"fine-grained":[167],"DAG":[168],"offline":[169],"search":[170],"solidifies":[172],"path,":[176],"completely":[177],"eliminating":[178],"branching;":[180],"(3)":[182],"heterogeneous":[184],"hybrid":[185],"engine":[187],"embeds":[189],"plugin":[193],"TensorRT-LLM,":[195],"combining":[196],"high-throughput":[197],"Prefill":[198],"low-latency":[200],"Decode.":[201],"On":[202],"an":[203],"L20,":[205],"Ada-MK":[206],"improves":[207],"single-batch":[208],"throughput":[209],"up":[211],"23.6%":[213],"over":[214,219],"vanilla":[215],"TensorRT-LLM":[216],"50.2%":[218],"vLLM,":[220],"achieving":[221],"positive":[222],"gains":[223],"across":[224],"all":[225],"tested":[226],"scenarios--the":[227],"first":[228],"industrial":[229],"system.":[238]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-05-14T00:00:00"}