{"id":"https://openalex.org/W7133503979","doi":"https://doi.org/10.48550/arxiv.2603.02637","title":"StitchCUDA: An Automated Multi-Agents End-to-End GPU Programing Framework with Rubric-based Agentic Reinforcement Learning","display_name":"StitchCUDA: An Automated Multi-Agents End-to-End GPU Programing Framework with Rubric-based Agentic Reinforcement Learning","publication_year":2026,"publication_date":"2026-03-03","ids":{"openalex":"https://openalex.org/W7133503979","doi":"https://doi.org/10.48550/arxiv.2603.02637"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2603.02637","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.02637","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"article"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2603.02637","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5090166559","display_name":"Shiyang Li","orcid":"https://orcid.org/0000-0003-3865-2197"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Shiyang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128126392","display_name":"Zijian Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Zijian","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128070792","display_name":"Winson Chen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Winson","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128117458","display_name":"Yuebo Luo","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Luo, Yuebo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128116210","display_name":"Mingyi Hong","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hong, Mingyi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5002899157","display_name":"Caiwen Ding","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ding, Caiwen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":6,"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/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.4106999933719635,"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"}},"topics":[{"id":"https://openalex.org/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.4106999933719635,"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"}},{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.2295999974012375,"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/T10462","display_name":"Reinforcement Learning in Robotics","score":0.05869999900460243,"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/reinforcement-learning","display_name":"Reinforcement learning","score":0.6916000247001648},{"id":"https://openalex.org/keywords/speedup","display_name":"Speedup","score":0.6345999836921692},{"id":"https://openalex.org/keywords/cuda","display_name":"CUDA","score":0.6312999725341797},{"id":"https://openalex.org/keywords/kernel","display_name":"Kernel (algebra)","score":0.5859000086784363},{"id":"https://openalex.org/keywords/correctness","display_name":"Correctness","score":0.5224999785423279},{"id":"https://openalex.org/keywords/coding","display_name":"Coding (social sciences)","score":0.42739999294281006},{"id":"https://openalex.org/keywords/code","display_name":"Code (set theory)","score":0.38260000944137573},{"id":"https://openalex.org/keywords/focus","display_name":"Focus (optics)","score":0.3630000054836273}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8756999969482422},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.6916000247001648},{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.6345999836921692},{"id":"https://openalex.org/C2778119891","wikidata":"https://www.wikidata.org/wiki/Q477690","display_name":"CUDA","level":2,"score":0.6312999725341797},{"id":"https://openalex.org/C74193536","wikidata":"https://www.wikidata.org/wiki/Q574844","display_name":"Kernel (algebra)","level":2,"score":0.5859000086784363},{"id":"https://openalex.org/C55439883","wikidata":"https://www.wikidata.org/wiki/Q360812","display_name":"Correctness","level":2,"score":0.5224999785423279},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.446399986743927},{"id":"https://openalex.org/C179518139","wikidata":"https://www.wikidata.org/wiki/Q5140297","display_name":"Coding (social sciences)","level":2,"score":0.42739999294281006},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.4196999967098236},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.38260000944137573},{"id":"https://openalex.org/C192209626","wikidata":"https://www.wikidata.org/wiki/Q190909","display_name":"Focus (optics)","level":2,"score":0.3630000054836273},{"id":"https://openalex.org/C50630238","wikidata":"https://www.wikidata.org/wiki/Q971505","display_name":"General-purpose computing on graphics processing units","level":3,"score":0.3625999987125397},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.36160001158714294},{"id":"https://openalex.org/C187191949","wikidata":"https://www.wikidata.org/wiki/Q1138496","display_name":"Profiling (computer programming)","level":2,"score":0.3467999994754791},{"id":"https://openalex.org/C34165917","wikidata":"https://www.wikidata.org/wiki/Q188267","display_name":"Programming paradigm","level":2,"score":0.33980000019073486},{"id":"https://openalex.org/C2776999362","wikidata":"https://www.wikidata.org/wiki/Q2349274","display_name":"Planner","level":2,"score":0.3093000054359436},{"id":"https://openalex.org/C43126263","wikidata":"https://www.wikidata.org/wiki/Q128751","display_name":"Source code","level":2,"score":0.3077000081539154},{"id":"https://openalex.org/C199360897","wikidata":"https://www.wikidata.org/wiki/Q9143","display_name":"Programming language","level":1,"score":0.2921999990940094},{"id":"https://openalex.org/C139571649","wikidata":"https://www.wikidata.org/wiki/Q1156793","display_name":"Program optimization","level":3,"score":0.2897999882698059},{"id":"https://openalex.org/C108583219","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep learning","level":2,"score":0.2678999900817871},{"id":"https://openalex.org/C519991488","wikidata":"https://www.wikidata.org/wiki/Q28865","display_name":"Python (programming language)","level":2,"score":0.2596000134944916},{"id":"https://openalex.org/C43521106","wikidata":"https://www.wikidata.org/wiki/Q2165493","display_name":"Pipeline (software)","level":2,"score":0.2547000050544739},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.25360000133514404},{"id":"https://openalex.org/C133162039","wikidata":"https://www.wikidata.org/wiki/Q1061077","display_name":"Code generation","level":3,"score":0.2533999979496002},{"id":"https://openalex.org/C2777904410","wikidata":"https://www.wikidata.org/wiki/Q7397","display_name":"Software","level":2,"score":0.2515000104904175}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2603.02637","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.02637","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"article"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2603.02637","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.02637","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Modern":[0],"machine":[1],"learning":[2,116],"(ML)":[3],"workloads":[4],"increasingly":[5],"rely":[6],"on":[7,19,32,41,173,183],"GPUs,":[8],"yet":[9],"achieving":[10],"high":[11],"end-to-end":[12,49,68,108,184],"performance":[13,97],"remains":[14],"challenging":[15],"due":[16],"to":[17,48,78,86,142],"dependencies":[18],"both":[20],"GPU":[21,34,69,109,185],"kernel":[22,35,150],"efficiency":[23],"and":[24,44,90,96,123,131,154,196],"host-side":[25],"settings.":[26],"Although":[27],"LLM-based":[28],"methods":[29],"show":[30,175],"promise":[31],"automated":[33],"generation,":[36,71],"prior":[37],"works":[38],"mainly":[39],"focus":[40],"single-kernel":[42],"optimization":[43],"do":[45],"not":[46],"extend":[47],"programs,":[50],"hindering":[51],"practical":[52],"deployment.":[53],"To":[54,101],"address":[55],"the":[56,104,138,193,199],"challenge,":[57],"in":[58,107],"this":[59],"work,":[60],"we":[61,155],"propose":[62],"StitchCUDA,":[63],"a":[64,76,83,91],"multi-agent":[65,194],"framework":[66],"for":[67,93],"program":[70],"with":[72,127,188],"three":[73],"specialized":[74],"agents:":[75],"Planner":[77],"orchestrate":[79],"whole":[80],"system":[81],"design,":[82],"Coder":[84,139],"dedicated":[85],"implementing":[87],"it":[88],"step-by-step,":[89],"Verifier":[92],"correctness":[94],"check":[95],"profiling":[98],"using":[99],"Nsys/NCU.":[100],"fundamentally":[102],"improve":[103],"Coder's":[105,159],"ability":[106],"programming,":[110],"StitchCUDA":[111,177],"integrates":[112],"rubric-based":[113],"agentic":[114],"reinforcement":[115],"over":[117,192],"two":[118],"atomic":[119],"skills,":[120],"task-to-code":[121],"generation":[122],"feedback-driven":[124],"code":[125,166],"optimization,":[126],"combined":[128],"rubric":[129],"reward":[130,133,160],"rule-based":[132],"from":[134],"real":[135],"executions.":[136],"Therefore,":[137],"learns":[140],"how":[141],"implement":[143],"advanced":[144],"CUDA":[145],"programming":[146,186],"techniques":[147],"(e.g.,":[148,162],"custom":[149],"fusion,":[151],"cublas":[152],"epilogue),":[153],"also":[156],"effectively":[157],"prevent":[158],"hacking":[161],"just":[163],"copy":[164],"PyTorch":[165],"or":[167],"hardcoding":[168],"output)":[169],"during":[170],"benchmarking.":[171],"Experiments":[172],"KernelBench":[174],"that":[176],"achieves":[178],"nearly":[179],"100%":[180],"success":[181],"rate":[182],"tasks,":[187],"1.72x":[189],"better":[190],"speedup":[191],"baseline":[195],"2.73x":[197],"than":[198],"RL":[200],"model":[201],"baselines.":[202]},"counts_by_year":[],"updated_date":"2026-06-11T09:08:48.828518","created_date":"2026-03-05T00:00:00"}