{"id":8320,"date":"2026-03-09T04:08:00","date_gmt":"2026-03-08T19:08:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=8320"},"modified":"2026-03-07T17:17:35","modified_gmt":"2026-03-07T08:17:35","slug":"cuda-agent-large-scale-agentic-rl-for-high-performance-cuda-kernel-generation","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=8320","title":{"rendered":"CUDA Agent: Large-Scale Agentic RL for High-Performance CUDA Kernel Generation\u00a0"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>CUDA Agent: Large-Scale Agentic RL for High-Performance CUDA Kernel Generation\u00a0<\/strong>[51.7]<br>\u30ab\u30fc\u30cd\u30eb\u306e\u5c02\u9580\u77e5\u8b58\u30923\u3064\u306e\u30b3\u30f3\u30dd\u30fc\u30cd\u30f3\u30c8\u3067\u958b\u767a\u3059\u308b\u5927\u898f\u6a21\u306a\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u5f37\u5316\u5b66\u7fd2\u30b7\u30b9\u30c6\u30e0\u3067\u3042\u308b\u3002\u5f93\u6765\u306e\u65b9\u6cd5\u306b\u6bd4\u3079\u3001\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u306f\u30b9\u30b1\u30fc\u30e9\u30d6\u30eb\u306a\u30c7\u30fc\u30bf\u5408\u6210\u30d1\u30a4\u30d7\u30e9\u30a4\u30f3\u3068\u81ea\u52d5\u691c\u8a3c\u74b0\u5883\u3092\u6d3b\u7528\u3057\u3001\u5b89\u5b9a\u3057\u305f\u8a13\u7df4\u3092\u5b9f\u73fe\u3057\u305f\u3002KernelBench\u306e\u6027\u80fd\u30c6\u30b9\u30c8\u3067\u306f\u3001torch.compile\u306b\u5bfe\u3057\u3066\u6700\u592740%\u306e\u6027\u80fd\u5411\u4e0a\u3092\u9054\u6210\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2602.24286v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2602.24286v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Fri, 27 Feb 2026 18:58:05 GMT)<\/li>\n\n\n\n<li>\u300cWe introduced CUDA Agent, a large-scale agentic reinforcement learning system that endows large language models with the ability to generate and optimize CUDA kernels under realistic, execution-driven development workflows. By jointly scaling data synthesis, agent environments, and stability-oriented RL training, CUDA Agent moves LLMs beyond syntactic code generation toward hardware-aware performance optimization, achieving consistent gains over torch.compile and strong proprietary models on KernelBench.\u300d\u3068\u306e\u3053\u3068\u3002\u30d9\u30fc\u30b9\u30e2\u30c7\u30eb\u306fSeed1.6\u3001\u3053\u3053\u307e\u3067\u3084\u3089\u306a\u3044\u3068\u30fb\u30fb\u30fb\u3068\u3044\u3046\u70b9\u3001Skills\u3067\u7d50\u5408\u3059\u308b\u3088\u3046\u306a\u69cb\u6210\u3082\u8208\u5473\u6df1\u3044\u3002<\/li>\n\n\n\n<li>\u30d7\u30ed\u30b8\u30a7\u30af\u30c8\u30b5\u30a4\u30c8\u306f<a href=\"https:\/\/cuda-agent.github.io\/\">CUDA Agent | Large-Scale Agentic RL for CUDA Kernel Generation<\/a><\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[2],"tags":[468],"class_list":["post-8320","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-468"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/8320","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=8320"}],"version-history":[{"count":1,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/8320\/revisions"}],"predecessor-version":[{"id":8321,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/8320\/revisions\/8321"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=8320"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=8320"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=8320"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}