{"id":5007,"date":"2026-05-20T09:00:28","date_gmt":"2026-05-20T01:00:28","guid":{"rendered":"https:\/\/1ow.top\/index.php\/2026\/05\/20\/%e5%9b%bd%e4%ba%a7gpu%e7%bb%84%e4%ba%86%e4%b8%aa%e5%bc%80%e6%ba%90%e5%b1%80%ef%bc%8c%e6%8a%8asglang%e7%ad%89%e6%a0%b8%e5%bf%83%e5%bc%80%e5%8f%91%e8%80%85%e9%83%bd%e6%91%87%e6%9d%a5%e4%ba%86%ef%bc%81\/"},"modified":"2026-05-20T09:00:28","modified_gmt":"2026-05-20T01:00:28","slug":"%e5%9b%bd%e4%ba%a7gpu%e7%bb%84%e4%ba%86%e4%b8%aa%e5%bc%80%e6%ba%90%e5%b1%80%ef%bc%8c%e6%8a%8asglang%e7%ad%89%e6%a0%b8%e5%bf%83%e5%bc%80%e5%8f%91%e8%80%85%e9%83%bd%e6%91%87%e6%9d%a5%e4%ba%86%ef%bc%81","status":"publish","type":"post","link":"https:\/\/1ow.top\/index.php\/2026\/05\/20\/%e5%9b%bd%e4%ba%a7gpu%e7%bb%84%e4%ba%86%e4%b8%aa%e5%bc%80%e6%ba%90%e5%b1%80%ef%bc%8c%e6%8a%8asglang%e7%ad%89%e6%a0%b8%e5%bf%83%e5%bc%80%e5%8f%91%e8%80%85%e9%83%bd%e6%91%87%e6%9d%a5%e4%ba%86%ef%bc%81\/","title":{"rendered":"\u56fd\u4ea7GPU\u7ec4\u4e86\u4e2a\u5f00\u6e90\u5c40\uff0c\u628aSGLang\u7b49\u6838\u5fc3\u5f00\u53d1\u8005\u90fd\u6447\u6765\u4e86\uff01"},"content":{"rendered":"
\"\u56fd\u4ea7GPU\u7ec4\u4e86\u4e2a\u5f00\u6e90\u5c40\uff0c\u628aSGLang\u7b49\u6838\u5fc3\u5f00\u53d1\u8005\u90fd\u6447\u6765\u4e86\uff01\"\/<\/div><\/figure>\n
\"\u56fd\u4ea7GPU\u7ec4\u4e86\u4e2a\u5f00\u6e90\u5c40\uff0c\u628aSGLang\u7b49\u6838\u5fc3\u5f00\u53d1\u8005\u90fd\u6447\u6765\u4e86\uff01\"\/<\/div><\/figure>\n
\"\u56fd\u4ea7GPU\u7ec4\u4e86\u4e2a\u5f00\u6e90\u5c40\uff0c\u628aSGLang\u7b49\u6838\u5fc3\u5f00\u53d1\u8005\u90fd\u6447\u6765\u4e86\uff01\"\/<\/div><\/figure>\n
\"\u56fd\u4ea7GPU\u7ec4\u4e86\u4e2a\u5f00\u6e90\u5c40\uff0c\u628aSGLang\u7b49\u6838\u5fc3\u5f00\u53d1\u8005\u90fd\u6447\u6765\u4e86\uff01\"\/<\/div><\/figure>\n

\u91d1\u78ca \u53d1\u81ea \u51f9\u975e\u5bfa<\/p>\n

\u6ca1\u6709\u5927\u5382\u9ad8\u7ba1\u7ad9\u53f0\uff0c\u4e00\u5c4b\u5b50\u5374\u6324\u6ee1\u4e86\u5f00\u6e90\u5708\u7684\u719f\u9762\u5b54\u3002<\/p>\n

\u968f\u4fbf\u5f80\u53f0\u4e0b\u626b\u4e00\u773c\uff0c\u5c31\u80fd\u5bf9\u4e0a\u597d\u51e0\u4e2aGitHub\u4e0a\u7684\u660e\u661fID\uff1a<\/p>\n

\u8fd9\u573a\u770b\u4f3c\u662f\u5f00\u6e90\u5708\u6781\u5ba2\u4eec\u7684\u9762\u57fa\u4f1a\uff0c\u5374\u7740\u5b9e\u662f\u6709\u70b9\u53cd\u5dee\u5728\u8eab\u4e0a\u7684\u2014\u2014<\/p>\n

\u6d3b\u52a8\u7684\u6512\u5c40\u8005\uff0c\u662f\u56fd\u4ea7GPU\u73a9\u5bb6\uff0c\u6469\u5c14\u7ebf\u7a0b\u3002<\/p>\n

\u8fd9\u4e8b\u786e\u5b9e\u6709\u70b9\u610f\u601d\u3002<\/p>\n

\u56e0\u4e3a\u8fc7\u53bb\u63d0\u5230\u56fd\u4ea7GPU\uff0c\u5916\u754c\u6700\u5bb9\u6613\u60f3\u5230\u7684\u5173\u952e\u8bcd\uff0c\u5f80\u5f80\u8fd8\u662f\u786c\u4ef6\u53c2\u6570\u3001\u663e\u5b58\u5bb9\u91cf\u3001\u7b97\u529b\u6307\u6807\u3001\u751f\u6001\u66ff\u4ee3\u3001\u6a21\u578b\u80fd\u4e0d\u80fd\u8dd1\u8d77\u6765\u3002<\/p>\n

\u4f46\u8fd9\u573aSGLang \u00d7 MUSA Meetup\u771f\u6b63\u629b\u51fa\u7684\u95ee\u9898\u5df2\u7ecf\u53d8\u4e86\uff1a<\/p>\n

\u600e\u4e48\u8ba9\u56fd\u4ea7GPU\u771f\u6b63\u8fdb\u5165\u5927\u6a21\u578b\u63a8\u7406\u7684\u4e3b\u6d41\u5f00\u6e90\u5de5\u7a0b\u94fe\u8def\uff1f<\/p>\n

\u8bf4\u5f97\u66f4\u76f4\u63a5\u4e00\u70b9\uff0c\u5c31\u662f\u8ba9SGLang\u3001Triton\/FlagOS\u3001TileLang\u3001Mooncake\u3001KVCache\u3001P\/D\u5206\u79bb\u3001\u5206\u5e03\u5f0f\u901a\u4fe1\u3001CI\/CD\u3001upstream PR\u8fd9\u4e9b\u4e1c\u897f\uff0c\u80fd\u591f\u56f4\u7740\u56fd\u4ea7GPU\u4e00\u8d77\u8f6c\u8d77\u6765\u3002<\/p>\n

\u6709\u4e00\u8bf4\u4e00\uff0c\u5728\u6574\u4f53\u542c\u4e0b\u6765\u4e4b\u540e\uff0c\u6709\u4e00\u4e2a\u975e\u5e38\u76f4\u89c2\u7684\u611f\u53d7\u3002<\/p>\n

\u90a3\u5c31\u662f\u56fd\u4ea7GPU\u7684\u7ade\u4e89\uff0c\u5df2\u7ecf\u4e0d\u53ea\u662f\u82af\u7247\u53c2\u6570\u4e4b\u4e89\uff0c\u8f6c\u800c\u5f00\u59cb\u8fc8\u5411\u751f\u6001\u5750\u6807\u4e4b\u4e89\u3002<\/p>\n

\u4e3a\u4ec0\u4e48\u8fd9\u4e48\u8bf4\uff1f\u6211\u4eec\u7ee7\u7eed\u5f80\u4e0b\u770b\u3002<\/p>\n

\u5148\u770b\u8fd9\u573aMeetup\u672c\u8eab\u3002<\/p>\n

\u5b83\u7684\u4e3b\u9898\u5f88\u660e\u786e\uff1aSGLang \u00d7 MUSA\u3002<\/p>\n

SGLang\u662f\u5f53\u4e0b\u5927\u6a21\u578b\u63a8\u7406serving\u9886\u57df\u5173\u6ce8\u5ea6\u5f88\u9ad8\u7684\u5f00\u6e90\u6846\u67b6\uff0c\u9762\u5411LLM\u548c\u591a\u6a21\u6001\u6a21\u578b\uff0c\u6838\u5fc3\u76ee\u6807\u662f\u4f4e\u5ef6\u8fdf\u3001\u9ad8\u541e\u5410\uff0c\u8986\u76d6\u4ece\u5355\u5361\u5230\u5927\u89c4\u6a21\u5206\u5e03\u5f0f\u96c6\u7fa4\u7684\u90e8\u7f72\u573a\u666f\u3002<\/p>\n

\u8fd9\u7c7b\u6846\u67b6\u4e4b\u6240\u4ee5\u91cd\u8981\uff0c\u662f\u56e0\u4e3a\u4eca\u5929\u7684\u5927\u6a21\u578b\u843d\u5730\uff0c\u65e9\u5c31\u4e0d\u662f\u201c\u6a21\u578b\u8bad\u7ec3\u597d\u4e86\uff0c\u653e\u4e0a\u53bb\u8dd1\u201d\u8fd9\u4e48\u7b80\u5355\u3002<\/p>\n

\u771f\u6b63\u8fdb\u5165\u751f\u4ea7\u73af\u5883\u540e\uff0c\u7cfb\u7edf\u8981\u5904\u7406\u7684\u662f\u4e00\u6574\u5957\u590d\u6742\u95ee\u9898\u3002<\/p>\n

\u4f8b\u5982prefill\u548cdecode\u600e\u4e48\u62c6\uff0cKVCache\u600e\u4e48\u590d\u7528\uff0c\u957f\u4e0a\u4e0b\u6587\u600e\u4e48\u7701\u94b1\uff0c\u591a\u8f6e\u5bf9\u8bdd\u600e\u4e48\u964dTTFT\uff0c\u5927\u89c4\u6a21\u96c6\u7fa4\u600e\u4e48\u8c03\u5ea6\uff0c\u65b0\u6a21\u578b\u53d1\u5e03\u540e\u600e\u4e48day-0 support\uff0c\u51fa\u4e86\u6027\u80fdgap\u600e\u4e48\u5b9a\u4f4d\u5230\u5177\u4f53kernel\u3002<\/p>\n

\u7b2c\u4e00\u4e2a\u4e0a\u53f0\u7684\u662fSGLang\u6838\u5fc3\u5f00\u53d1\u8005BBuf\u3002<\/p>\n

\u8fd9\u4e2a\u5728GitHub\u4e0a\u62e5\u670927k\u661f\u7684\u5f00\u6e90\u63a8\u7406\u6846\u67b6\uff0c\u73b0\u5728\u5df2\u7ecf\u662f\u5168\u7403\u5f00\u53d1\u8005\u90e8\u7f72\u5927\u6a21\u578b\u7684\u9996\u9009\u3002<\/p>\n

\u4ed6\u5e26\u6765\u7684SGLang 2026 Q2 Roadmap\uff0c\u6bcf\u4e00\u6761\u90fd\u8e29\u5728\u884c\u4e1a\u7684\u75db\u70b9\u4e0a\uff1a<\/p>\n

\u6700\u8ba9\u4eba\u5370\u8c61\u6df1\u523b\u7684\u662f\u4ed6\u5c55\u793a\u7684\u4e00\u7ec4\u6570\u636e\u3002<\/p>\n

SGLang\u901a\u8fc7P\/D\u5206\u79bb\u67b6\u6784\uff0c\u572812\u4e2aH100\u8282\u70b9\u4e0a\u8dd1\u51fa\u4e8652.3k\u8f93\u5165token\/s\/node\u300122.3k\u8f93\u51fatoken\/s\/node\u7684\u6210\u7ee9\uff0c\u6bd4DeepSeek\u5b98\u65b9API\u8fd8\u4fbf\u5b9c5\u500d\uff0c\u8fd9\u4e2a\u7ed3\u679c\u5df2\u7ecf\u88ab\u5168\u740310\u591a\u4e2a\u56e2\u961f\u590d\u73b0\u3002<\/p>\n

\u7d27\u63a5\u7740\u4e0a\u53f0\u7684\u6469\u5c14\u7ebf\u7a0bContributorR0CKSTAR\uff0c\u5e26\u6765\u4e86\u5168\u573a\u6700\u786c\u6838\u7684\u5de5\u7a0b\u5b9e\u8df5\u5206\u4eab\u3002<\/p>\n

\u4ed6\u7528\u4e00\u53e5\u8bdd\u603b\u7ed3\u4e86\u8fc7\u53bb\u534a\u5e74\u7684\u5de5\u4f5c\uff1a<\/p>\n

SGLang on MUSA\u5df2\u7ecf\u5b8c\u6210\u4e86\u4ece\u73af\u5883\u6784\u5efa\u5230CI\u6d4b\u8bd5\u7684\u5168\u94fe\u8def\u6253\u901a\u3002<\/p>\n

\u8fd9\u610f\u5473\u7740\u4ec0\u4e48\uff1f<\/p>\n

\u73b0\u5728\u4f60\u53ea\u8981\u514b\u9686SGLang\u7684\u5b98\u65b9\u4ed3\u5e93\uff0c\u5b89\u88c5sgl-kernel\u548csglang\uff0c\u5c31\u80fd\u5728\u6469\u5c14\u7ebf\u7a0bMTT S5000\u663e\u5361\u4e0a\u76f4\u63a5\u8fd0\u884c\u51e0\u4e4e\u6240\u6709\u4e3b\u6d41\u5927\u6a21\u578b\u3002<\/p>\n

DeepSeek\u3001\u901a\u4e49\u5343\u95ee3.5\u3001GLM-4.5\u3001FLUX\u3001Wan\u8fd9\u4e9b\u70ed\u95e8\u6a21\u578b\uff0c\u90fd\u5df2\u7ecf\u5b8c\u6210\u4e86\u6df1\u5ea6\u4f18\u5316\u3002<\/p>\n

\u4ed6\u7279\u522b\u63d0\u5230\u4e86MUSA\u7684\u4e09\u5c42CUDA\u517c\u5bb9\u6808\u3002<\/p>\n

\u8fc7\u53bb\u9002\u914d\u4e00\u4e2a\u63a8\u7406\u6846\u67b6\u8981\u6539\u51e0\u5343\u884c\u4ee3\u7801\uff0c\u73b0\u5728\u53ea\u8981\u5728\u5f00\u5934\u52a0\u4e00\u884cimport torchada\uff0c99%\u7684CUDA\u4ee3\u7801\u5c31\u80fd\u76f4\u63a5\u8fd0\u884c\u3002\u8fd9\u4e2a\u770b\u4f3c\u7b80\u5355\u7684\u6539\u52a8\uff0c\u89e3\u51b3\u4e86\u56fd\u4ea7GPU\u751f\u6001\u7684\u4e00\u5927\u75db\u70b9\u3002<\/p>\n

\u636e\u4e86\u89e3\uff0c\u622a\u81f35\u670812\u65e5\uff0c\u6469\u5c14\u7ebf\u7a0b\u5728SGLang\u4e3b\u7ebf\u7d2f\u8ba1\u63d0\u4ea447\u4e2aPR\uff0c\u5176\u4e2d41\u4e2a\u5df2\u5408\u5165\uff0c\u5b8c\u6210\u4e86\u4ece\u73af\u5883\u6784\u5efa\u5230\u5206\u5e03\u5f0f\u63a8\u7406\u7684\u5168\u94fe\u8def\u6253\u901a\u3002<\/p>\n

\u667a\u6e90\u7684\u8096\u822a\u5219\u5e26\u6765\u4e86DeepSeek V4\u5728MUSA\u4e0a\u7684Day0 \u9002\u914d\u6210\u679c\u3002<\/p>\n

\u901a\u8fc7FlagOS\u7684Triton\u7b97\u5b50\u4f18\u5316\u548c\u6469\u5c14\u7ebf\u7a0b\u7684SQMMA\u5f20\u91cf\u52a0\u901f\u5f15\u64ce\uff0c\u4ed6\u4eec\u628aDeepSeek V4\u7684\u9996token\u5ef6\u8fdf\u964d\u4f4e\u4e8656.7%\uff0c\u541e\u5410\u91cf\u63d0\u5347\u4e8623%\u3002<\/p>\n

\u5bf9\u6b64\uff0c\u8096\u822a\u8868\u793a\uff1a<\/p>\n

\u6211\u4eec\u6ca1\u6709\u505a\u4ec0\u4e48\u9ed1\u9b54\u6cd5\uff0c\u5c31\u662f\u628a\u4e24\u4e2a\u6700\u5173\u952e\u7684\u7b97\u5b50\u4f18\u5316\u5230\u4e86\u6781\u81f4\u3002<\/p>\n

FP8 \u77e9\u9635\u4e58\u7b97\u5b50\u5e73\u5747\u52a0\u901f8.85\u500d\uff0c\u7a00\u758f\u6ce8\u610f\u529b\u7b97\u5b50\u5e73\u5747\u52a0\u901f6.01\u500d\uff0c\u8fd9\u4e24\u4e2a\u5360\u4e86\u63a8\u7406\u65f6\u95f480%\u7684\u7b97\u5b50\u4e00\u4f18\u5316\uff0c\u7aef\u5230\u7aef\u6027\u80fd\u81ea\u7136\u5c31\u4e0a\u53bb\u4e86\u3002<\/p>\n

TileLang\u7ef4\u62a4\u8005\u5510\u6b63\u4e3e\u7684\u5206\u4eab\uff0c\u5219\u8ba9\u6240\u6709\u4eba\u770b\u5230\u4e86\u4e0b\u4e00\u4ee3\u7b97\u5b50\u7f16\u7a0b\u7684\u672a\u6765\u3002<\/p>\n

\u8fd9\u4e2a2025\u5e742\u6708\u624d\u5f00\u6e90\u7684\u9879\u76ee\uff0c\u77ed\u77ed\u4e00\u5e74\u591a\u5c31\u6536\u83b7\u4e866k\u661f\u548c133\u4f4d\u8d21\u732e\u8005\uff0c\u8fdeDeepSeek V4\u7684\u6838\u5fc3kernel\u90fd\u662f\u7528TileLang\u5199\u7684\uff0c\u6b63\u5982\u5510\u6b63\u4e3e\u6240\u8bf4\uff1a<\/p>\n

\u7528TileLang\u5199FlashAttention\uff0c\u53ea\u898150\u884cPython\u4ee3\u7801\uff0c\u6027\u80fd\u548c\u4e13\u5bb6\u624b\u5199\u7684CUDA\u4e00\u6a21\u4e00\u6837\u3002<\/p>\n

\u5e76\u4e14\u4ece\u4ed6\u5728\u73b0\u573a\u5c55\u793a\u7684\u5bf9\u6bd4\u56fe\u6765\u770b\uff0c\u540c\u6837\u7684GEMM\u7b97\u5b50\uff0cTileLang\u752815\u884c\u4ee3\u7801\u8fbe\u5230\u4e86CUTLASS\u7684\u6027\u80fd\uff0c\u4ee3\u7801\u91cf\u51cf\u5c11\u4e8690%\u3002<\/p>\n

\u6700\u540e\u4e0a\u53f0\u7684\u963f\u91cc\u4e91\u9a6c\u817e\uff0c\u5e26\u6765\u4e86Mooncake\u9879\u76ee\u7684\u6700\u65b0\u8fdb\u5c55\u3002<\/p>\n

\u8fd9\u4e2a\u4e13\u6ce8\u4e8eKVCache\u89e3\u8026\u7684\u9879\u76ee\uff0c\u73b0\u5728\u5df2\u7ecf\u662fSGLang\u3001vLLM\u7b49\u4e3b\u6d41\u63a8\u7406\u6846\u67b6\u7684\u6807\u914d\u3002<\/p>\n

\u4ed6\u5c55\u793a\u7684\u4e00\u7ec4\u8f83\u4e3a\u5438\u775b\u7684\u6570\u636e\uff1a<\/p>\n

\u901a\u8fc7RDMA P2P\u6743\u91cd\u66f4\u65b0\uff0cKimi K2 1T\u6a21\u578b\u7684\u6743\u91cd\u540c\u6b65\u65f6\u95f4\u4ece53\u79d2\u964d\u5230\u4e867.2\u79d2\uff0c\u52a0\u901f\u4e867.37\u500d\uff1bEPD\u4e09\u7ea7\u89e3\u8026\u67b6\u6784\u8ba9\u591a\u6a21\u6001\u6a21\u578b\u7684\u9996token\u5ef6\u8fdf\u964d\u4f4e\u4e866-8\u500d\uff1bHiCache + Mooncake\u540e\u7aef\u8ba9\u591a\u8f6e\u5bf9\u8bdd\u7684\u7f13\u5b58\u547d\u4e2d\u7387\u8d85\u8fc790%\u3002<\/p>\n

\u81f3\u6b64\uff0c\u8fd9\u573aMeetup\u7684\u62fc\u56fe\u57fa\u672c\u5b8c\u6574\u2014\u2014<\/p>\n

SGLang\u662f\u63a8\u7406\u6846\u67b6\u4e3b\u94fe\u8def\uff0cMUSA\u662f\u56fd\u4ea7GPU\u5e95\u5c42\u5e73\u53f0\uff0cFlagOS\/Triton\u89e3\u51b3\u5173\u952e\u7b97\u5b50\u4f18\u5316\uff0cTileLang\u964d\u4f4e\u9ad8\u6027\u80fdkernel\u7f16\u7a0b\u95e8\u69db\uff0cMooncake\u8865\u4e0aKVCache\u548c\u751f\u4ea7\u90e8\u7f72\u3002<\/p>\n

\u8fd9\uff0c\u4fbf\u662f\u4e00\u6761\u8f83\u4e3a\u5b8c\u6574\u7684\u5de5\u7a0b\u94fe\u8def\u3002<\/p>\n

\u8fd9\u4e2a\u95ee\u9898\u7684\u7b54\u6848\u4e0d\u80fd\u53ea\u5f52\u7ed3\u4e3a\u529e\u4e86\u4e00\u573a\u6d3b\u52a8\u3002<\/p>\n

\u5f00\u6e90\u5708\u5f88\u73b0\u5b9e\uff0c\u5927\u5bb6\u613f\u610f\u6765\uff0c\u6838\u5fc3\u539f\u56e0\u4e0d\u662f\u8c01\u4f1a\u8bb2\u6545\u4e8b\uff0c\u662f\u8fd9\u4ef6\u4e8b\u771f\u7684\u548c\u4ed6\u4eec\u6b63\u5728\u505a\u7684\u5de5\u7a0b\u95ee\u9898\u6709\u5173\u3002<\/p>\n

\u9996\u5148\u770bMUSA\u672c\u8eab\u7684\u8bbe\u8ba1\u521d\u5fc3\u3002<\/p>\n

\u6469\u5c14\u7ebf\u7a0bCTO\u5f20\u94b0\u52c3\u5728\u5f00\u573a\u4e2d\u89e3\u91ca\uff0cMUSA\u662fMeta-computing Unified System Architecture\u3002<\/p>\n

Meta-computing\u6307\u5411\u901a\u7528\u8ba1\u7b97\uff0c\u6469\u5c14\u7ebf\u7a0b\u5e0c\u671bGPU\u5c3d\u91cf\u62e5\u62b1\u901a\u7528\u8ba1\u7b97\uff0c\u800c\u4e0d\u662f\u7ed9\u672a\u6765\u53ef\u8ba1\u7b97\u7684\u9886\u57df\u8bbe\u9650\uff1bUnified\u5219\u610f\u5473\u7740\u6469\u5c14\u7ebf\u7a0b\u4ea7\u54c1\u5e0c\u671b\u9075\u5faa\u540c\u4e00\u5957\u7edf\u4e00\u6807\u51c6\uff0c\u907f\u514d\u4e0d\u540c\u4ea7\u54c1\u7ebf\u4f7f\u7528\u4e0d\u540c\u6307\u4ee4\u96c6\u548c\u67b6\u6784\uff0c\u5bfc\u81f4\u8f6f\u4ef6\u751f\u6001\u65e0\u6cd5\u79ef\u7d2f\u3002<\/p>\n

\u66f4\u5173\u952e\u7684\u4e00\u53e5\u8bdd\u662f\uff0cMUSA\u4e0d\u5e0c\u671b\u5f00\u53d1\u8005\u4e3a\u4e86\u4f7f\u7528MUSA\u800c\u91cd\u65b0\u5b66\u4e60\u4e00\u5957\u4e1c\u897f\u3002<\/p>\n

\u8fd9\u53e5\u8bdd\u770b\u4f3c\u6734\u7d20\uff0c\u5176\u5b9e\u76f4\u6307\u56fd\u4ea7GPU\u751f\u6001\u7684\u75db\u70b9\u3002<\/p>\n

\u5f00\u53d1\u8005\u6700\u6015\u4ec0\u4e48\uff1f<\/p>\n

\u4e0d\u662f\u65b0\u786c\u4ef6\u672c\u8eab\uff0c\u662f\u4e3a\u4e86\u65b0\u786c\u4ef6\uff0c\u5b66\u4e60\u4e00\u6574\u5957\u65b0API\uff0c\u91cd\u5199\u4e00\u5806\u4ee3\u7801\uff0c\u6539\u5b8c\u8fd8\u8fdb\u4e0d\u4e86\u4e0a\u6e38\uff0c\u793e\u533a\u4e00\u66f4\u65b0\u53c8\u8981\u91cd\u65b0\u8865\u4e01\u3002<\/p>\n

\u5982\u679c\u4e00\u4e2a\u56fd\u4ea7GPU\u751f\u6001\u8981\u6c42\u5f00\u53d1\u8005\u4ece\u5934\u5b66\u4e00\u904d\uff0c\u90a3\u5b83\u9762\u5bf9\u7684\u4fbf\u662f\u5de8\u5927\u7684\u8fc1\u79fb\u963b\u529b\u3002<\/p>\n

\u6240\u4ee5MUSA\u7684\u8def\u7ebf\uff0c\u662f\u5c3d\u91cf\u8d34\u8fd1\u5f00\u53d1\u8005\u5df2\u7ecf\u719f\u6089\u7684GPU\u7f16\u7a0b\u65b9\u5f0f\u3001API\u63a5\u53e3\u548c\u4f7f\u7528\u4e60\u60ef\u3002\u5e95\u5c42\u5b9e\u73b0\u53ef\u4ee5\u4e0d\u540c\uff0c\u4f46\u4e0a\u5c42\u4f53\u9a8c\u5c3d\u53ef\u80fd\u4e00\u81f4\u3002<\/p>\n

\u4e09\u5c42CUDA\u517c\u5bb9\u6808\u7684\u610f\u4e49\u5c31\u5728\u8fd9\u91cc\u3002<\/p>\n

torch_musa\u8d1f\u8d23\u628aPyTorch\u548cMUSA\u7684\u57fa\u7840\u80fd\u529b\u63a5\u8d77\u6765\uff1btorchada\u8d1f\u8d23\u8ba9CUDA-first\u751f\u6001\u7ee7\u7eed\u5de5\u4f5c\uff1bmthreads-ml-py\u8d1f\u8d23\u628a\u8bbe\u5907\u7ba1\u7406\u3001\u62d3\u6251\u3001\u663e\u5b58\u3001MTLink\u3001P2P\u7b49\u4fe1\u606f\u66b4\u9732\u7ed9\u4e0a\u5c42\u6846\u67b6\u3002<\/p>\n

\u7528\u4e00\u53e5\u66f4\u901a\u4fd7\u7684\u8bdd\u8bf4\uff0c\u6469\u5c14\u7ebf\u7a0b\u5728\u5c3d\u91cf\u628a\u539f\u6765\u7684\u8def\u4fee\u5230\u81ea\u5df1\u95e8\u53e3\u3002<\/p>\n

\u8fd9\u76f4\u63a5\u5f71\u54cd\u5230\u5f00\u6e90\u793e\u533a\u534f\u4f5c\u7684\u53ef\u884c\u6027\u3002<\/p>\n

\u56e0\u4e3a\u4e0a\u6e38\u9879\u76ee\u6700\u770b\u91cd\u7684\u662f\u4f4e\u4fb5\u5165\u3001\u53ef\u7ef4\u62a4\u3001\u53ef\u590d\u7528\u3002\u5982\u679c\u4e00\u4e2a\u9002\u914d\u65b9\u6848\u9700\u8981\u5927\u9762\u79ef\u6539\u52a8\u4e3b\u7ebf\u4ee3\u7801\uff0c\u540e\u7eed\u6bcf\u6b21rebase\u90fd\u75db\u82e6\uff0c\u4e0a\u6e38\u5f88\u96be\u63a5\u53d7\u3002<\/p>\n

\u53cd\u8fc7\u6765\uff0c\u5982\u679c\u9002\u914d\u53ef\u4ee5\u901a\u8fc7\u66f4\u900f\u660e\u7684\u65b9\u5f0f\u5b8c\u6210\uff0cPR\u5c31\u66f4\u5bb9\u6613\u88abreview\uff0c\u4e5f\u66f4\u5bb9\u6613\u6301\u7eed\u8ddf\u968f\u793e\u533a\u8fed\u4ee3\u3002<\/p>\n

\u8fd9\u5c31\u662f\u4ece\u201c\u6211\u81ea\u5df1\u7ef4\u62a4\u4e00\u4e2a\u5206\u652f\u201d\u5230\u201c\u6211\u8fdb\u5165\u4e3b\u7ebf\u201d\u7684\u533a\u522b\u3002<\/p>\n

\u518d\u770b\u751f\u6001\u7ed3\u5408\u3002<\/p>\n

SGLang \u00d7 MUSA\uff0c\u662f\u63a8\u7406\u4e3b\u94fe\u8def\u6253\u901a\u3002<\/p>\n

\u6469\u5c14\u7ebf\u7a0b\u4ece\u53bb\u5e74\u5f00\u59cb\u628aSGLang\u4f5c\u4e3a\u91cd\u70b9\u63a5\u5165\u548c\u8d21\u732e\u7684\u5f00\u6e90\u9879\u76ee\uff0c\u7ecf\u8fc7\u5927\u534a\u5e74\u52aa\u529b\uff0cMUSA\u540e\u7aef\u8fd1\u671f\u5df2\u7ecf\u5408\u5165SGLang\u4e3b\u7ebf\u3002\u540e\u7eed\u4e0d\u53ea\u662f\u8ddf\u968ffeature\uff0c\u4e5f\u5e0c\u671b\u5728\u6846\u67b6\u5c42\u9762\u8d21\u732e\u66f4\u591a\u80fd\u529b\u3002<\/p>\n

\u8fd9\u4ef6\u4e8b\u7684\u610f\u4e49\u5728\u4e8e\uff0c\u56fd\u4ea7GPU\u4e0d\u518d\u53ea\u662f\u67d0\u4e2a\u6846\u67b6\u7684\u5916\u90e8\u9002\u914d\u5bf9\u8c61\uff0c\u5df2\u7ecf\u5f00\u59cb\u6210\u4e3a\u4e3b\u7ebf\u751f\u6001\u7684\u4e00\u90e8\u5206\u3002<\/p>\n

FlagOS \u00d7 MUSA\uff0c\u662f\u5173\u952e\u7b97\u5b50\u548c\u65b0\u6a21\u578b\u9002\u914d\u3002<\/p>\n

\u5927\u6a21\u578b\u63a8\u7406\u7684\u6027\u80fd\u7ade\u4e89\uff0c\u8d8a\u6765\u8d8a\u591a\u53d1\u751f\u5728kernel\u3001\u7f16\u8bd1\u5668\u3001\u8c03\u5ea6\u3001\u4f4e\u7cbe\u5ea6\u548c\u901a\u4fe1\u5c42\u3002DeepSeek V4 day-0\u9002\u914d\u8fd9\u6837\u7684\u5de5\u4f5c\uff0c\u672c\u8d28\u4e0a\u8003\u9a8c\u7684\u662f\u4ece\u6a21\u578b\u53d1\u5e03\u5230\u5de5\u7a0b\u843d\u5730\u4e4b\u95f4\u7684\u53cd\u5e94\u901f\u5ea6\u3002\u80fd\u4e0d\u80fd\u7b2c\u4e00\u65f6\u95f4\u8dd1\u901a\uff0c\u80fd\u4e0d\u80fd\u5feb\u901f\u8c03\u4f18\uff0c\u80fd\u4e0d\u80fd\u5728\u771f\u5b9eshape\u4e0a\u627e\u5230\u66f4\u597d\u7684\u914d\u7f6e\uff0c\u51b3\u5b9a\u4e86\u751f\u6001\u8ddf\u4e0d\u8ddf\u5f97\u4e0a\u3002<\/p>\n

Mooncake \u00d7 MUSA\uff0c\u662f\u63a8\u7406\u89e3\u8026\u548c\u751f\u4ea7\u90e8\u7f72\u3002<\/p>\n

KVCache\u7684\u4ef7\u503c\u5728Agent\u3001\u591a\u8f6e\u5bf9\u8bdd\u3001\u957f\u4e0a\u4e0b\u6587\u65f6\u4ee3\u88ab\u8fdb\u4e00\u6b65\u653e\u5927\u3002Mooncake\u4e0eMUSA\u7684\u7ed3\u5408\uff0c\u4e0d\u53ea\u662f\u8ba9\u67d0\u4e2a\u7f13\u5b58\u540e\u7aef\u80fd\u8dd1\u5728\u56fd\u4ea7GPU\u4e0a\uff0c\u66f4\u662f\u5728\u63a2\u7d22\u8de8\u5b9e\u4f8bKVCache\u5171\u4eab\u3001\u5f39\u6027\u6269\u7f29\u5bb9\u3001\u7f13\u5b58\u590d\u7528\u3001\u539f\u5730\u5347\u7ea7\u8fd9\u7c7b\u751f\u4ea7\u7ea7\u95ee\u9898\u3002<\/p>\n

TileLang \u00d7 MUSA\uff0c\u5219\u662f\u4e0b\u4e00\u4ee3\u7b97\u5b50\u751f\u6001\u7684\u63d0\u524d\u5e03\u5c40\u3002<\/p>\n

\u5982\u679c\u672a\u6765\u66f4\u591a\u6a21\u578b\u548c\u786c\u4ef6\u90fd\u9700\u8981\u5b9a\u5236kernel\uff0c\u7b97\u5b50\u7f16\u7a0b\u4e0d\u80fd\u6c38\u8fdc\u505c\u7559\u5728\u5c11\u6570\u4e13\u5bb6\u624b\u91cc\u3002TileLang\u8fd9\u7c7bDSL\u7684\u4ef7\u503c\uff0c\u662f\u628a\u9ad8\u6027\u80fdkernel\u7f16\u7a0b\u53d8\u6210\u66f4\u591a\u5f00\u53d1\u8005\u80fd\u4e0a\u624b\u7684\u5de5\u7a0b\u5de5\u5177\u3002<\/p>\n

\u8fd9\u56db\u6761\u7ebf\u5408\u5728\u4e00\u8d77\uff0c\u624d\u662f\u6469\u5c14\u7ebf\u7a0b\u80fd\u7ec4\u5c40\u7684\u5e95\u6c14\u3002<\/p>\n

\u5b83\u628a\u81ea\u5df1\u653e\u8fdb\u4e86\u5927\u6a21\u578b\u63a8\u7406\u7684\u771f\u5b9e\u5de5\u7a0b\u7f51\u7edc\u91cc\uff0c\u5305\u62ec\u6846\u67b6\u3001\u7b97\u5b50\u3001\u7f13\u5b58\u3001\u901a\u4fe1\u3001\u90e8\u7f72\u3001CI\/CD\u3001upstream\u7b49\u7b49\u3002<\/p>\n

\u800c\u8fd9\uff0c\u4e5f\u662f\u56fd\u4ea7 GPU \u751f\u6001\u771f\u6b63\u8981\u8865\u7684\u8bfe\u3002<\/p>\n

\u5982\u679c\u628a\u8fd9\u573aMeetup\u4ece\u66f4\u5b8f\u5927\u7684\u7b97\u529b\u53d1\u5c55\u89d2\u5ea6\u6765\u770b\uff0c\u5b83\u7684\u4ef7\u503c\u6216\u8bb8\u8fdc\u8d85\u6280\u672f\u5206\u4eab\u672c\u8eab\u3002<\/p>\n

\u8fc7\u53bb\u51e0\u5e74\uff0c\u56fd\u4ea7GPU\u7684\u751f\u6001\u56f0\u5883\u662f\u6bd4\u8f83\u660e\u663e\u7684\u3002<\/p>\n

\u8bb8\u591a\u5382\u5546\u4e60\u60ef\u4e86\u95ed\u95e8\u9020\u8f66\uff0c\u81ea\u5df1\u4ece\u5934\u5199\u4e00\u5957\u6df1\u5ea6\u5b66\u4e60\u6846\u67b6\uff0c\u81ea\u5df1\u6512\u4e00\u5957\u7b97\u5b50\u5e93\uff0c\u7ed3\u679c\u56e0\u4e3a\u4e0d\u7b26\u5408\u4e3b\u6d41\u5f00\u53d1\u8005\u7684\u4e60\u60ef\uff0c\u9c9c\u6709\u4eba\u95ee\u6d25\u3002<\/p>\n

\u53c8\u6216\u8005\uff0c\u6709\u7684\u5382\u5546\u53ea\u662f\u62c9\u4e00\u4e2a\u79c1\u6709Fork\u505a\u9002\u914d\uff0c\u4ece\u6765\u4e0d\u5411\u5f00\u6e90\u4e0a\u6e38\u63d0\u4ea4\u4ee3\u7801\uff0c\u5bfc\u81f4\u4e3b\u6d41\u6846\u67b6\u4e00\u66f4\u65b0\uff0c\u81ea\u5df1\u7684\u9002\u914d\u7248\u672c\u5c31\u6210\u4e86\u65e0\u4eba\u7ef4\u62a4\u7684\u5b64\u54c1\u3002<\/p>\n

\u800c\u73b0\u5728\uff0c\u6469\u5c14\u7ebf\u7a0b\u7ed9\u51fa\u4e86\u4e00\u4e2a\u5b8c\u5168\u4e0d\u540c\u7684\u7b54\u6848\uff1a<\/p>\n

\u5168\u9762\u878d\u5165\u5168\u7403\u5f00\u6e90\u751f\u6001\uff0c\u53bb\u548c\u4e16\u754c\u4e0a\u6700\u806a\u660e\u7684\u4e00\u6279\u4eba\u4e00\u8d77\u505a\u4e8b\u3002<\/p>\n

\u5728\u8fd9\u573a\u6d3b\u52a8\u4e2d\uff0c\u6211\u4eec\u9891\u7e41\u542c\u5230\u51e0\u4e2a\u8bcd\uff1aDay-0 Support\u3001Upstream PR\u3001CI\/CD\u3002<\/p>\n

\u8fd9\u8bf4\u660e\u56fd\u4ea7GPU\u7684\u751f\u6001\u4f4d\u6b63\u5728\u53d1\u751f\u8d28\u53d8\u3002\u6469\u5c14\u7ebf\u7a0b\u4e0d\u518d\u53ea\u6ee1\u8db3\u4e8e\u505a\u4e00\u4e2a\u88ab\u52a8\u7684\u9002\u914d\u8005\uff0c\u5b83\u8981\u7684\u662f\u4e3b\u52a8\u51fa\u51fb\uff0c\u6210\u4e3a\u6838\u5fc3\u4ee3\u7801\u7684\u201c\u8d21\u732e\u8005\u201d\uff0c\u751a\u81f3\u662f\u672a\u6765\u67b6\u6784\u7684\u201c\u5171\u5efa\u8005\u201d\u3002<\/p>\n

\u4ed6\u4eec\u4e0d\u4ec5\u4ec5\u662f\u4e22\u4e00\u4e2a\u5355\u70b9\u7684Patch\u8fc7\u53bb\uff0c\u800c\u662f\u628a\u4e00\u6574\u5957\u5305\u542b\u73af\u5883\u6784\u5efa\u3001PR\u63d0\u4ea4\u3001CI\u81ea\u52a8\u5316\u6d4b\u8bd5\u3001Release\u53d1\u5e03\u3001\u6587\u6863\u7ef4\u62a4\u5728\u5185\u7684\u5de5\u7a0b\u95ed\u73af\uff0c\u6df1\u6df1\u5730\u5d4c\u5165\u5230\u4e86SGLang\u7b49\u9876\u7ea7\u9879\u76ee\u7684\u8840\u8109\u4e2d\u3002<\/p>\n

\u8fd9\u79cd\u53ef\u6301\u7eed\u7684Upstream\u6a21\u5f0f\uff0c\u624d\u662f\u771f\u6b63\u638c\u63e1\u751f\u6001\u8bdd\u8bed\u6743\u7684\u65b9\u5f0f\u3002<\/p>\n

\u8fd9\u573a\u5f00\u6e90\u5c40\u8fd8\u8bc1\u660e\u4e86\u4e00\u4ef6\u4e8b\uff0c\u56fd\u4ea7GPU\u5df2\u7ecf\u8d70\u4e0a\u4e86\u5927\u6a21\u578b\u63a8\u7406\u5f00\u6e90\u751f\u6001\u7684\u516c\u5171\u724c\u684c\u3002<\/p>\n

\u5728\u8fd9\u4e2a\u724c\u684c\u4e0a\uff0c\u5df2\u7ecf\u5750\u7740\u98ce\u5934\u6b63\u52b2\u7684SGLang\uff0c\u5750\u7740\u6b7b\u78d5\u5e95\u5c42\u7f16\u8bd1\u7684Triton\/FlagOS\uff0c\u5750\u7740\u91cd\u5851\u7b97\u5b50\u751f\u6001\u7684TileLang\uff0c\u5750\u7740\u4e3b\u5bfc\u89e3\u8026\u67b6\u6784\u7684Mooncake\u3002<\/p>\n

\u800c\u73b0\u5728\uff0c\u56fd\u4ea7GPU\uff0c\u4e5f\u53ef\u4ee5\u62c9\u5f00\u6905\u5b50\uff0c\u4ece\u5bb9\u5730\u5750\u4e0b\u6765\uff0c\u548c\u8fd9\u7fa4\u660e\u661f\u73a9\u5bb6\u4eec\u4e00\u8d77\u6253\u597d\u5927\u6a21\u578b\u65f6\u4ee3\u6700\u5173\u952e\u7684\u8fd9\u628a\u724c\u3002<\/p>\n

\n

\u7248\u6743\u58f0\u660e\uff1a<\/strong><\/p>\n

\u672c\u535a\u5ba2\u90e8\u5206\u5185\u5bb9\u4e3a\u8f6c\u8f7d\u6587\u7ae0\uff0c\u65e8\u5728\u5206\u4eab\u6709\u4ef7\u503c\u7684\u4fe1\u606f\uff0c\u7248\u6743\u5f52\u539f\u4f5c\u8005\u6240\u6709\u3002<\/p>\n

\u8f6c\u8f7d\u4ec5\u4e3a\u4e2a\u4eba\u5b66\u4e60\u4e0e\u4ea4\u6d41\u76ee\u7684\uff0c\u4e0d\u5bf9\u6587\u7ae0\u89c2\u70b9\u8d1f\u8d23\uff0c\u4ea6\u4e0d\u7528\u4e8e\u4efb\u4f55\u5546\u4e1a\u7528\u9014\u3002<\/p>\n

\u5982\u6d89\u53ca\u7248\u6743\u95ee\u9898\uff0c\u8bf7\u8054\u7cfb\u672c\u4eba\u5220\u9664\u3002<\/p>\n<\/blockquote>\n","protected":false},"excerpt":{"rendered":"

\u91d1\u78ca \u53d1\u81ea \u51f9\u975e\u5bfa<\/p>\n

\u6ca1\u6709\u5927\u5382\u9ad8\u7ba1\u7ad9\u53f0\uff0c\u4e00\u5c4b\u5b50\u5374\u6324\u6ee1\u4e86\u5f00\u6e90\u5708\u7684\u719f\u9762\u5b54\u3002<\/p>\n

\u968f\u4fbf\u5f80\u53f0\u4e0b\u626b\u4e00\u773c\uff0c\u5c31\u80fd\u5bf9\u4e0a\u597d\u51e0\u4e2aGitHub\u4e0a\u7684\u660e\u661fID\uff1a<\/p>\n

\u8fd9\u573a\u770b\u4f3c\u662f\u5f00\u6e90\u5708\u6781\u5ba2\u4eec\u7684\u9762\u57fa\u4f1a\uff0c\u5374\u7740\u5b9e\u662f\u6709\u70b9\u53cd\u5dee\u5728\u8eab\u4e0a\u7684\u2014\u2014<\/p>\n

\u6d3b\u52a8\u7684\u6512\u5c40\u8005\uff0c\u662f\u56fd\u4ea7GPU\u73a9\u5bb6\uff0c\u6469\u5c14\u7ebf\u7a0b\u3002<\/p>\n

\u8fd9\u4e8b\u786e\u5b9e\u6709\u70b9\u610f\u601d\u3002<\/p>\n

\u56e0\u4e3a\u8fc7\u53bb\u63d0\u5230\u56fd\u4ea7GPU\uff0c\u5916\u754c\u6700\u5bb9\u6613\u60f3\u5230\u7684\u5173\u952e\u8bcd\uff0c\u5f80\u5f80\u8fd8\u662f\u786c\u4ef6\u53c2\u6570\u3001\u663e\u5b58\u5bb9\u91cf\u3001\u7b97\u529b\u6307\u6807\u3001\u751f\u6001\u66ff\u4ee3\u3001\u6a21\u578b\u80fd\u4e0d\u80fd\u8dd1\u8d77\u6765\u3002<\/p>\n

\u4f46\u8fd9\u573aSGLa<\/p>\n","protected":false},"author":1,"featured_media":5003,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[846,35,847,36],"class_list":["post-5007","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-xwzx","tag-sglang","tag-gpu","tag-847","tag-36"],"_links":{"self":[{"href":"https:\/\/1ow.top\/index.php\/wp-json\/wp\/v2\/posts\/5007","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/1ow.top\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/1ow.top\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/1ow.top\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/1ow.top\/index.php\/wp-json\/wp\/v2\/comments?post=5007"}],"version-history":[{"count":0,"href":"https:\/\/1ow.top\/index.php\/wp-json\/wp\/v2\/posts\/5007\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/1ow.top\/index.php\/wp-json\/wp\/v2\/media\/5003"}],"wp:attachment":[{"href":"https:\/\/1ow.top\/index.php\/wp-json\/wp\/v2\/media?parent=5007"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/1ow.top\/index.php\/wp-json\/wp\/v2\/categories?post=5007"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/1ow.top\/index.php\/wp-json\/wp\/v2\/tags?post=5007"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}