{"id":6714,"date":"2025-05-12T03:48:00","date_gmt":"2025-05-11T18:48:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=6714"},"modified":"2025-05-10T16:52:18","modified_gmt":"2025-05-10T07:52:18","slug":"pangu-ultra-moe-how-to-train-your-big-moe-on-ascend-npus","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=6714","title":{"rendered":"Pangu Ultra MoE: How to Train Your Big MoE on Ascend NPUs"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Pangu Ultra MoE: How to Train Your Big MoE on Ascend NPUs\u00a0<\/strong>[111.7]<br>\u30df\u30ad\u30c1\u30e3\u30fc\u30fb\u30aa\u30d6\u30fb\u30a8\u30ad\u30b9\u30d1\u30fc\u30c8(MoE)\u30681\u5146\u8fd1\u3044\u30d1\u30e9\u30e1\u30fc\u30bf\u3092\u6301\u3064\u758e\u5927\u8a00\u8a9e\u30e2\u30c7\u30eb(LLM)\u304c\u3001\u6700\u3082\u6709\u80fd\u306a\u8a00\u8a9e\u30e2\u30c7\u30eb\u306e\u9818\u57df\u3092\u652f\u914d\u3057\u3066\u3044\u308b\u3002 \u672c\u7a3f\u3067\u306f,Ascend NPU\u4e0a\u3067\u305d\u306e\u3088\u3046\u306a\u30b9\u30b1\u30fc\u30eb\u3092\u5229\u7528\u3059\u308b\u30ec\u30b7\u30d4\u3092\u660e\u3089\u304b\u306b\u3059\u308b\u3053\u3068\u3092\u76ee\u7684\u3068\u3057\u3066\u3044\u308b\u3002 \u4e3b\u306a\u76ee\u7684\u306f\u3001\u52d5\u7684\u30b9\u30d1\u30fc\u30b9\u30e2\u30c7\u30eb\u69cb\u9020\u4e0b\u3067\u306e\u30b3\u30f3\u30d4\u30e5\u30fc\u30c6\u30a3\u30f3\u30b0\u30ea\u30bd\u30fc\u30b9\u306e\u3088\u308a\u826f\u3044\u4f7f\u7528\u3068\u3001\u5b9f\u969b\u306e\u30cf\u30fc\u30c9\u30a6\u30a7\u30a2\u3067\u671f\u5f85\u3055\u308c\u308b\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u5411\u4e0a\u306e\u5b9f\u73fe\u3067\u3042\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2505.04519v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2505.04519v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Wed, 07 May 2025 15:46:36 GMT)<\/li>\n\n\n\n<li><a href=\"https:\/\/devneko.jp\/wordpress\/?p=6561\">Llama 4, Nemotron-H, Pangu Ultra, Kimi-VL, Kimi-VL-Thinking, Deep Coder \u2013 arXiv\u6700\u65b0\u8ad6\u6587\u306e\u7d39\u4ecb<\/a>\u306b\u3082\u95a2\u9023\u3059\u308bPangu Ultra\u306e\u4e3b\u306b\u5b9f\u88c5\u306b\u95a2\u3059\u308b\u8ad6\u6587\u3002<\/li>\n\n\n\n<li>\u300cOur system optimizations focus on Expert Parallelism and memory management, significantly lowering communication and activation overhead across 6K NPUs. These innovations enable a 30.0% MFU, demonstrating Ascend NPUs\u2019 capability to support full-scale training of large-scale sparse LLMs, e g , Pangu Ultra MoE, with comparable performance as DeepSeek R1.\u300d\u3068\u306e\u3053\u3068\u3067NVIDIA\u306eGPU\u306b\u983c\u3089\u305a\u3068\u3082\u6700\u5148\u7aef\u30e2\u30c7\u30eb\u3092\u69cb\u7bc9\u53ef\u80fd\u3068\u4e3b\u5f35\u3057\u3066\u3044\u308b\u3088\u3046\u306b\u898b\u3048\u308b\u3002<\/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":[223],"class_list":["post-6714","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-llm"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6714","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=6714"}],"version-history":[{"count":1,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6714\/revisions"}],"predecessor-version":[{"id":6715,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6714\/revisions\/6715"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=6714"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=6714"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=6714"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}