{"id":784,"date":"2024-06-06T20:26:30","date_gmt":"2024-06-06T12:26:30","guid":{"rendered":"http:\/\/madapapa.com\/wordpress\/?p=784"},"modified":"2024-06-06T20:27:20","modified_gmt":"2024-06-06T12:27:20","slug":"hua-weiascend-npu-shi-pei","status":"publish","type":"post","link":"http:\/\/madapapa.com\/wordpress\/?p=784","title":{"rendered":"\u534e\u4e3aAscend NPU\u9002\u914d"},"content":{"rendered":"

<\/span><\/a>Ascend Extension for PyTorch\u63d2\u4ef6<\/h1>\n

<\/span><\/a>\u7b80\u4ecb<\/h2>\n

\u672c\u9879\u76ee\u5f00\u53d1\u4e86\u540d\u4e3atorch_npu<\/strong>\u7684Ascend Extension for PyTorch<\/strong>\u63d2\u4ef6\uff0c\u4f7f\u6607\u817eNPU\u53ef\u4ee5\u9002\u914dPyTorch\u6846\u67b6\uff0c\u4e3a\u4f7f\u7528PyTorch\u6846\u67b6\u7684\u5f00\u53d1\u8005\u63d0\u4f9b\u6607\u817eAI\u5904\u7406\u5668\u7684\u8d85\u5f3a\u7b97\u529b\u3002<\/p>\n

\u6607\u817e\u4e3a\u57fa\u4e8e\u534e\u4e3a\u6607\u817e\u5904\u7406\u5668\u548c\u8f6f\u4ef6\u7684\u884c\u4e1a\u5e94\u7528\u53ca\u670d\u52a1\u63d0\u4f9b\u5168\u6808AI\u8ba1\u7b97\u57fa\u7840\u8bbe\u65bd\u3002\u60a8\u53ef\u4ee5\u901a\u8fc7\u8bbf\u95ee\u6607\u817e\u793e\u533a<\/a>\uff0c\u4e86\u89e3\u5173\u4e8e\u6607\u817e\u7684\u66f4\u591a\u4fe1\u606f\u3002<\/p>\n

<\/span><\/a>\u5b89\u88c5<\/h2>\n

<\/span><\/a>\u4f7f\u7528\u4e8c\u8fdb\u5236\u6587\u4ef6\u8fdb\u884c\u5b89\u88c5<\/h3>\n

\u6211\u4eec\u4e3a\u7528\u6237\u63d0\u4f9b\u53ef\u4ee5\u5feb\u901f\u5b89\u88c5torch_npu<\/strong>\u7684whl\u5b89\u88c5\u5305\u3002\u5728\u5b89\u88c5torch_npu<\/strong>\u4e4b\u524d\uff0c\u60a8\u9700\u8981\u5148\u5b89\u88c5CANN<\/strong>\u8f6f\u4ef6\u3002\u6607\u817e\u8f85\u52a9\u8f6f\u4ef6<\/a>\u4e2d\u6709\u66f4\u591a\u5173\u4e8eCANN\u7684\u7248\u672c\u4fe1\u606f\u3002\u8bf7\u53c2\u8003CANN\u5b89\u88c5\u6307\u5357<\/a>\u83b7\u53d6CANN<\/strong>\u5b89\u88c5\u5305\u3002<\/p>\n

    \n
  1. \u5b89\u88c5PyTorch<\/strong><\/li>\n<\/ol>\n

    \u901a\u8fc7 pip \u5b89\u88c5 PyTorch\u3002<\/p>\n

    aarch64:<\/strong><\/p>\n

    pip3 install torch==2.1.0\n<\/code><\/pre>\n
    x86:<\/strong><\/p>\n
    pip3 install torch==2.1.0+cpu  --index-url https:\/\/download.pytorch.org\/whl\/cpu\n<\/code><\/pre>\n
    \u82e5\u4f7f\u7528pip\u547d\u4ee4\u5b89\u88c5\u5931\u8d25\uff0c\u8bf7\u4f7f\u7528\u4e0b\u8f7d\u94fe\u63a5\u6216\u8fdb\u5165PyTorch\u5b98\u65b9\u7f51\u7ad9<\/a>\u8fdb\u884c\u67e5\u8be2\u4e0b\u8f7d\u5bf9\u5e94\u7248\u672c\u3002<\/p>\n\n\n\n\n\n\n\n\n\n\n
    \u67b6\u6784<\/th>\nPython\u7248\u672c<\/th>\n\u4e0b\u8f7d\u94fe\u63a5<\/th>\n<\/tr>\n<\/thead>\n
    x86<\/td>\nPython3.8<\/td>\n\u4e0b\u8f7d\u94fe\u63a5<\/a><\/td>\n<\/tr>\n
    x86<\/td>\nPython3.9<\/td>\n\u4e0b\u8f7d\u94fe\u63a5<\/a><\/td>\n<\/tr>\n
    x86<\/td>\nPython3.10<\/td>\n\u4e0b\u8f7d\u94fe\u63a5<\/a><\/td>\n<\/tr>\n
    aarch64<\/td>\nPython3.8<\/td>\n\u4e0b\u8f7d\u94fe\u63a5<\/a><\/td>\n<\/tr>\n
    aarch64<\/td>\nPython3.9<\/td>\n\u4e0b\u8f7d\u94fe\u63a5<\/a><\/td>\n<\/tr>\n
    aarch64<\/td>\nPython3.10<\/td>\n\u4e0b\u8f7d\u94fe\u63a5<\/a><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n
      \n
    1. \u5b89\u88c5torch_npu\u4f9d\u8d56<\/strong><\/li>\n<\/ol>\n
      \u8fd0\u884c\u4ee5\u4e0b\u547d\u4ee4\u5b89\u88c5\u4f9d\u8d56\u3002<\/p>\n
      pip3 install pyyaml\npip3 install setuptools\n<\/code><\/pre>\n
        \n
      1. \u5b89\u88c5torch_npu<\/strong><\/li>\n<\/ol>\n
        pip3 install torch-npu==2.1.0.post3\n<\/code><\/pre>\n
        \u5982\u9700\u8981\u4fdd\u5b58\u5b89\u88c5\u65e5\u5fd7\uff0c\u53ef\u5728pip3 install\u547d\u4ee4\u540e\u9762\u52a0\u4e0a\u53c2\u6570 --log <PATH><\/code>\uff0c\u5e76\u5bf9\u60a8\u6307\u5b9a\u7684\u76ee\u5f55<PATH><\/code>\u505a\u597d\u6743\u9650\u7ba1\u63a7\u3002<\/p>\n
        <\/span><\/a>\u4f7f\u7528\u6e90\u4ee3\u7801\u8fdb\u884c\u5b89\u88c5<\/h3>\n
        \u67d0\u4e9b\u7279\u6b8a\u573a\u666f\u4e0b\uff0c\u7528\u6237\u53ef\u80fd\u9700\u8981\u81ea\u884c\u7f16\u8bd1torch_npu<\/strong>\u3002\u53ef\u4ee5\u6839\u636e\u6607\u817e\u8f85\u52a9\u8f6f\u4ef6\u8868<\/a>\u548cPyTorch\u4e0ePython\u7248\u672c\u914d\u5957\u8868<\/a>\u9009\u62e9\u5408\u9002\u7684\u5206\u652f\u3002\u63a8\u8350\u4f7f\u7528Docker\u955c\u50cf\u7f16\u8bd1torch_npu<\/strong>\uff0c\u53ef\u4ee5\u901a\u8fc7\u4ee5\u4e0b\u6b65\u9aa4\u83b7\u53d6(\u5efa\u8bae\u53ea\u6302\u8f7d\u5de5\u4f5c\u8def\u5f84\uff0c\u5e76\u907f\u5f00\u7cfb\u7edf\u8def\u5f84\uff0c\u4ee5\u964d\u4f4e\u5b89\u5168\u98ce\u9669), \u751f\u6210\u7684.whl\u6587\u4ef6\u8def\u5f84\u4e3a.\/dist\/\uff1a<\/p>\n
          \n
        1. \n
          \u514b\u9686torch_npu\u4ee3\u7801\u4ed3<\/strong><\/p>\n
          git clone https:\/\/gitee.com\/ascend\/pytorch.git -b v2.1.0-6.0.rc1 --depth 1\n<\/code><\/pre>\n<\/li>\n
        2. \n
          \u6784\u5efa\u955c\u50cf<\/strong><\/p>\n
          cd pytorch\/ci\/docker\/{arch} # {arch} for X86 or ARM\ndocker build -t manylinux-builder:v1 .\n<\/code><\/pre>\n<\/li>\n
        3. \n
          \u8fdb\u5165Docker\u5bb9\u5668<\/strong><\/p>\n
          docker run -it -v \/{code_path}\/pytorch:\/home\/pytorch manylinux-builder:v1 bash\n# {code_path} is the torch_npu source code path\n<\/code><\/pre>\n<\/li>\n
        4. \n
          \u7f16\u8bd1torch_npu<\/strong><\/p>\n
          \u4ee5Python 3.8<\/strong> \u4e3a\u4f8b\u3002<\/p>\n
          cd \/home\/pytorch\nbash ci\/build.sh --python=3.8\n<\/code><\/pre>\n<\/li>\n<\/ol>\n
          <\/span><\/a>\u5378\u8f7d<\/h2>\n
          Pytorch\u6846\u67b6\u8bad\u7ec3\u73af\u5883\u7684\u5378\u8f7d\u53ef\u4ee5\u53c2\u8003\u6607\u817e\u5b98\u65b9\u6587\u6863<\/a>\u3002<\/p>\n
          torch_npu\u7684\u5378\u8f7d\u53ea\u9700\u6267\u884c\u547d\u4ee4\uff1a<\/p>\n
          pip3 uninstall torch_npu\n<\/code><\/pre>\n
          \u5982\u9700\u8981\u4fdd\u5b58\u5378\u8f7d\u65e5\u5fd7\uff0c\u53ef\u5728pip3 uninstall\u547d\u4ee4\u540e\u9762\u52a0\u4e0a\u53c2\u6570 --log <PATH><\/code>\uff0c\u5e76\u5bf9\u60a8\u6307\u5b9a\u7684\u76ee\u5f55<PATH><\/code>\u505a\u597d\u6743\u9650\u7ba1\u63a7\u3002<\/p>\n
          <\/span><\/a>\u5165\u95e8<\/h2>\n
          <\/span><\/a>\u524d\u63d0<\/h3>\n
          \u8fd0\u884c\u4ee5\u4e0b\u547d\u4ee4\u521d\u59cb\u5316CANN<\/strong>\u73af\u5883\u53d8\u91cf\u3002<\/p>\n
          # Default path, change it if needed.\nsource \/usr\/local\/Ascend\/ascend-toolkit\/set_env.sh\n<\/code><\/pre>\n
          <\/span><\/a>\u5feb\u901f\u9a8c\u8bc1<\/h3>\n
          \u53ef\u4ee5\u901a\u8fc7\u4ee5\u4e0b\u6837\u4f8b\u5feb\u901f\u4f53\u9a8c\u6607\u817eNPU<\/strong>\u3002<\/p>\n
          import torch\nimport torch_npu\n\nx = torch.randn(2, 2).npu()\ny = torch.randn(2, 2).npu()\nz = x.mm(y)\n\nprint(z)\n<\/code><\/pre>\n
          <\/span><\/a>PyTorch\u4e0ePython\u7248\u672c\u914d\u5957\u8868<\/h2>\n\n\n\n\n\n\n\n
          PyTorch\u7248\u672c<\/th>\nPython\u7248\u672c<\/th>\n<\/tr>\n<\/thead>\n
          PyTorch1.11.0<\/td>\nPython3.7.x(>=3.7.5), Python3.8.x, Python3.9.x, Python3.10.x<\/td>\n<\/tr>\n
          PyTorch2.1.0<\/td>\nPython3.8.x, Python3.9.x, Python3.10.x<\/td>\n<\/tr>\n
          PyTorch2.2.0<\/td>\nPython3.8.x, Python3.9.x, Python3.10.x<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n
          <\/span><\/a>\u6607\u817e\u8f85\u52a9\u8f6f\u4ef6<\/h2>\n
          PyTorch Extension<\/strong>\u7248\u672c\u53f7\u91c7\u7528{PyTorch\u7248\u672c}-{\u6607\u817e\u7248\u672c}<\/code>\u547d\u540d\u89c4\u5219\uff0c\u524d\u8005\u4e3aPyTorch Extension<\/strong>\u5339\u914d\u7684PyTorch\u7248\u672c\uff0c\u540e\u8005\u7528\u4e8e\u5339\u914dCANN\u7248\u672c\uff0c\u8be6\u7ec6\u5339\u914d\u5982\u4e0b\uff1a<\/p>\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
          CANN\u7248\u672c<\/th>\n\u652f\u6301\u7684PyTorch\u7248\u672c<\/th>\n\u652f\u6301\u7684Extension\u7248\u672c<\/th>\nGitee\u5206\u652f<\/th>\nAscendHub\u955c\u50cf\u7248\u672c\/\u540d\u79f0(\u94fe\u63a5<\/a>)<\/th>\n<\/tr>\n<\/thead>\n
          CANN 8.0.RC1<\/td>\n2.2.0<\/td>\n2.2.0<\/td>\nv2.2.0-6.0.rc1<\/td>\n–<\/td>\n<\/tr>\n
          <\/td>\n2.1.0<\/td>\n2.1.0.post3<\/td>\nv2.1.0-6.0.rc1<\/td>\n–<\/td>\n<\/tr>\n
          <\/td>\n1.11.0<\/td>\n1.11.0.post11<\/td>\nv1.11.0-6.0.rc1<\/td>\n–<\/td>\n<\/tr>\n
          CANN 7.0.0<\/td>\n2.1.0<\/td>\n2.1.0<\/td>\nv2.1.0-5.0.0<\/td>\n–<\/td>\n<\/tr>\n
          <\/td>\n2.0.1<\/td>\n2.0.1.post1<\/td>\nv2.0.1-5.0.0<\/td>\n–<\/td>\n<\/tr>\n
          <\/td>\n1.11.0<\/td>\n1.11.0.post8<\/td>\nv1.11.0-5.0.0<\/td>\n–<\/td>\n<\/tr>\n
          CANN 7.0.RC1<\/td>\n2.1.0<\/td>\n2.1.0.rc1<\/td>\nv2.1.0-5.0.rc3<\/td>\n–<\/td>\n<\/tr>\n
          <\/td>\n2.0.1<\/td>\n2.0.1<\/td>\nv2.0.1-5.0.rc3<\/td>\n–<\/td>\n<\/tr>\n
          <\/td>\n1.11.0<\/td>\n1.11.0.post4<\/td>\nv1.11.0-5.0.rc3<\/td>\n–<\/td>\n<\/tr>\n
          CANN 6.3.RC3.1<\/td>\n1.11.0<\/td>\n1.11.0.post3<\/td>\nv1.11.0-5.0.rc2.2<\/td>\n–<\/td>\n<\/tr>\n
          CANN 6.3.RC3<\/td>\n1.11.0<\/td>\n1.11.0.post2<\/td>\nv1.11.0-5.0.rc2.1<\/td>\n–<\/td>\n<\/tr>\n
          CANN 6.3.RC2<\/td>\n2.0.1<\/td>\n2.0.1.rc1<\/td>\nv2.0.1-5.0.rc2<\/td>\n–<\/td>\n<\/tr>\n
          <\/td>\n1.11.0<\/td>\n1.11.0.post1<\/td>\nv1.11.0-5.0.rc2<\/td>\n23.0.RC1-1.11.0<\/td>\n<\/tr>\n
          <\/td>\n1.8.1<\/td>\n1.8.1.post2<\/td>\nv1.8.1-5.0.rc2<\/td>\n23.0.RC1-1.8.1<\/td>\n<\/tr>\n
          CANN 6.3.RC1<\/td>\n1.11.0<\/td>\n1.11.0<\/td>\nv1.11.0-5.0.rc1<\/td>\n–<\/td>\n<\/tr>\n
          <\/td>\n1.8.1<\/td>\n1.8.1.post1<\/td>\nv1.8.1-5.0.rc1<\/td>\n–<\/td>\n<\/tr>\n
          CANN 6.0.1<\/td>\n1.5.0<\/td>\n1.5.0.post8<\/td>\nv1.5.0-3.0.0<\/td>\n22.0.0<\/td>\n<\/tr>\n
          <\/td>\n1.8.1<\/td>\n1.8.1<\/td>\nv1.8.1-3.0.0<\/td>\n22.0.0-1.8.1<\/td>\n<\/tr>\n
          <\/td>\n1.11.0<\/td>\n1.11.0.rc2\uff08beta)<\/td>\nv1.11.0-3.0.0<\/td>\n–<\/td>\n<\/tr>\n
          CANN 6.0.RC1<\/td>\n1.5.0<\/td>\n1.5.0.post7<\/td>\nv1.5.0-3.0.rc3<\/td>\n22.0.RC3<\/td>\n<\/tr>\n
          <\/td>\n1.8.1<\/td>\n1.8.1.rc3<\/td>\nv1.8.1-3.0.rc3<\/td>\n22.0.RC3-1.8.1<\/td>\n<\/tr>\n
          <\/td>\n1.11.0<\/td>\n1.11.0.rc1\uff08beta)<\/td>\nv1.11.0-3.0.rc3<\/td>\n–<\/td>\n<\/tr>\n
          CANN 5.1.RC2<\/td>\n1.5.0<\/td>\n1.5.0.post6<\/td>\nv1.5.0-3.0.rc2<\/td>\n22.0.RC2<\/td>\n<\/tr>\n
          <\/td>\n1.8.1<\/td>\n1.8.1.rc2<\/td>\nv1.8.1-3.0.rc2<\/td>\n22.0.RC2-1.8.1<\/td>\n<\/tr>\n
          CANN 5.1.RC1<\/td>\n1.5.0<\/td>\n1.5.0.post5<\/td>\nv1.5.0-3.0.rc1<\/td>\n22.0.RC1<\/td>\n<\/tr>\n
          <\/td>\n1.8.1<\/td>\n1.8.1.rc1<\/td>\nv1.8.1-3.0.rc1<\/td>\n–<\/td>\n<\/tr>\n
          CANN 5.0.4<\/td>\n1.5.0<\/td>\n1.5.0.post4<\/td>\n2.0.4.tr5<\/td>\n21.0.4<\/td>\n<\/tr>\n
          CANN 5.0.3<\/td>\n1.8.1<\/td>\n1.5.0.post3<\/td>\n2.0.3.tr5<\/td>\n21.0.3<\/td>\n<\/tr>\n
          CANN 5.0.2<\/td>\n1.5.0<\/td>\n1.5.0.post2<\/td>\n2.0.2.tr5<\/td>\n21.0.2<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n
          <\/span><\/a>\u5efa\u8bae\u4e0e\u4ea4\u6d41<\/h2>\n
          \u6b22\u8fce\u5927\u5bb6\u4e3a\u793e\u533a\u505a\u8d21\u732e\u3002\u5982\u679c\u6709\u4efb\u4f55\u7591\u95ee\u6216\u5efa\u8bae\uff0c\u8bf7\u63d0\u4ea4gitee Issues<\/a>\uff0c\u6211\u4eec\u4f1a\u5c3d\u5feb\u56de\u590d\u3002\u611f\u8c22\u60a8\u7684\u652f\u6301\u3002<\/p>\n
          <\/span><\/a>\u5206\u652f\u7ef4\u62a4\u7b56\u7565<\/h2>\n
          AscendPyTorch\u7248\u672c\u5206\u652f\u7684\u7ef4\u62a4\u9636\u6bb5\u5982\u4e0b\uff1a<\/p>\n\n\n\n\n\n\n\n\n\n
          \u72b6\u6001<\/strong><\/th>\n\u65f6\u95f4<\/strong><\/th>\n\u8bf4\u660e<\/strong><\/th>\n<\/tr>\n<\/thead>\n
          \u8ba1\u5212<\/td>\n1\u20143 \u4e2a\u6708<\/td>\n\u8ba1\u5212\u7279\u6027<\/td>\n<\/tr>\n
          \u5f00\u53d1<\/td>\n3 \u4e2a\u6708<\/td>\n\u5f00\u53d1\u7279\u6027<\/td>\n<\/tr>\n
          \u7ef4\u62a4<\/td>\n6-12 \u4e2a\u6708<\/td>\n\u5408\u5165\u6240\u6709\u5df2\u89e3\u51b3\u7684\u95ee\u9898\u5e76\u53d1\u5e03\u7248\u672c\uff0c\u9488\u5bf9\u4e0d\u540c\u7684PyTorch\u7248\u672c\u91c7\u53d6\u4e0d\u540c\u7684\u7ef4\u62a4\u7b56\u7565\uff0c\u5e38\u89c4\u7248\u672c\u548c\u957f\u671f\u652f\u6301\u7248\u672c\u7ef4\u62a4\u5468\u671f\u5206\u522b\u4e3a6\u4e2a\u6708\u548c12\u4e2a\u6708<\/td>\n<\/tr>\n
          \u65e0\u7ef4\u62a4<\/td>\n0\u20143 \u4e2a\u6708<\/td>\n\u5408\u5165\u6240\u6709\u5df2\u89e3\u51b3\u7684\u95ee\u9898\uff0c\u65e0\u4e13\u804c\u7ef4\u62a4\u4eba\u5458\uff0c\u65e0\u7248\u672c\u53d1\u5e03<\/td>\n<\/tr>\n
          \u751f\u547d\u5468\u671f\u7ec8\u6b62\uff08EOL\uff09<\/td>\nN\/A<\/td>\n\u5206\u652f\u4e0d\u518d\u63a5\u53d7\u4efb\u4f55\u4fee\u6539<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n
          <\/span><\/a>PyTorch\u7248\u672c\u7ef4\u62a4\u7b56\u7565<\/h2>\n\n\n\n\n\n\n\n\n\n\n
          PyTorch\u7248\u672c<\/strong><\/th>\n\u7ef4\u62a4\u7b56\u7565<\/strong><\/th>\n\u5f53\u524d\u72b6\u6001<\/strong><\/th>\n\u53d1\u5e03\u65f6\u95f4<\/strong><\/th>\n\u540e\u7eed\u72b6\u6001<\/strong><\/th>\nEOL\u65e5\u671f<\/strong><\/th>\n<\/tr>\n<\/thead>\n
          2.2.0<\/td>\n\u5e38\u89c4\u7248\u672c<\/td>\n\u7ef4\u62a4<\/td>\n2024\/04\/01<\/td>\n\u9884\u8ba12024\/10\/15\u8d77\u65e0\u7ef4\u62a4<\/td>\n<\/td>\n<\/tr>\n
          2.1.0<\/td>\n\u957f\u671f\u652f\u6301<\/td>\n\u7ef4\u62a4<\/td>\n2023\/10\/15<\/td>\n\u9884\u8ba12024\/10\/15\u8d77\u65e0\u7ef4\u62a4<\/td>\n<\/td>\n<\/tr>\n
          2.0.1<\/td>\n\u5e38\u89c4\u7248\u672c<\/td>\nEOL<\/td>\n2023\/7\/19<\/td>\n<\/td>\n2024\/3\/14<\/td>\n<\/tr>\n
          1.11.0<\/td>\n\u957f\u671f\u652f\u6301<\/td>\n\u7ef4\u62a4<\/td>\n2023\/4\/19<\/td>\n\u9884\u8ba12024\/4\/19\u8d77\u65e0\u7ef4\u62a4<\/td>\n<\/td>\n<\/tr>\n
          1.8.1<\/td>\n\u957f\u671f\u652f\u6301<\/td>\nEOL<\/td>\n2022\/4\/10<\/td>\n<\/td>\n2023\/4\/10<\/td>\n<\/tr>\n
          1.5.0<\/td>\n\u957f\u671f\u652f\u6301<\/td>\nEOL<\/td>\n2021\/7\/29<\/td>\n<\/td>\n2022\/7\/29<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n
          <\/span><\/a>\u5b89\u5168\u58f0\u660e<\/h2>\n
          Ascend Extension for PyTorch\u63d2\u4ef6 \u5b89\u5168\u58f0\u660e<\/a><\/p>\n
          <\/span><\/a>\u53c2\u8003\u6587\u6863<\/h2>\n
          \u6709\u5173\u5b89\u88c5\u6307\u5357\u3001\u6a21\u578b\u8fc1\u79fb\u548c\u8bad\u7ec3\/\u63a8\u7406\u6559\u7a0b\u548cAPI\u5217\u8868\u7b49\u66f4\u591a\u8be6\u7ec6\u4fe1\u606f\uff0c\u8bf7\u53c2\u8003\u6607\u817e\u793e\u533aAscend Extension for PyTorch<\/a>\u3002<\/p>\n\n\n\n\n\n\n\n\n
          \u6587\u6863\u540d\u79f0<\/th>\n\u6587\u6863\u94fe\u63a5<\/th>\n<\/tr>\n<\/thead>\n
          \u5b89\u88c5\u6307\u5357<\/td>\n\u53c2\u8003\u94fe\u63a5<\/a><\/td>\n<\/tr>\n
          \u7f51\u7edc\u6a21\u578b\u8fc1\u79fb\u548c\u8bad\u7ec3<\/td>\n\u53c2\u8003\u94fe\u63a5<\/a><\/td>\n<\/tr>\n
          \u7b97\u5b50\u9002\u914d<\/td>\n\u53c2\u8003\u94fe\u63a5<\/a><\/td>\n<\/tr>\n
          API\u6e05\u5355\uff08PyTorch\u539f\u751f\u63a5\u53e3\u4e0e\u81ea\u5b9a\u4e49\u63a5\u53e3\uff09<\/td>\n\u53c2\u8003\u94fe\u63a5<\/a><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n
          <\/span><\/a>\u8bb8\u53ef\u8bc1<\/h2>\n
          Ascend Extension for PyTorch\u63d2\u4ef6\u4f7f\u7528BSD\u8bb8\u53ef\u8bc1\u3002\u8be6\u89c1LICENSE<\/a>\u6587\u4ef6\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"
          Ascend Extension for PyTorch\u63d2\u4ef6 \u7b80\u4ecb \u672c\u9879\u76ee\u5f00\u53d1\u4e86\u540d\u4e3atorch_npu\u7684Ascend Extension for PyTorch\u63d2\u4ef6\uff0c\u4f7f\u6607\u817eNPU\u53ef\u4ee5\u9002\u914dPyTorch\u6846\u67b6\uff0c\u4e3a\u4f7f\u7528PyTorch\u6846\u67b6\u7684\u5f00\u53d1\u8005\u63d0\u4f9b\u6607\u817eAI\u5904\u7406\u5668\u7684\u8d85\u5f3a\u7b97\u529b\u3002 \u6607\u817e\u4e3a\u57fa\u4e8e\u534e\u4e3a\u6607\u817e\u5904\u7406\u5668\u548c\u8f6f\u4ef6\u7684\u884c\u4e1a\u5e94\u7528\u53ca\u670d\u52a1\u63d0\u4f9b\u5168\u6808AI\u8ba1\u7b97\u57fa\u7840\u8bbe\u65bd\u3002\u60a8\u53ef\u4ee5\u901a\u8fc7\u8bbf\u95ee\u6607\u817e\u793e\u533a\uff0c\u4e86\u89e3\u5173\u4e8e\u6607\u817e\u7684\u66f4\u591a\u4fe1\u606f\u3002 \u5b89\u88c5 \u4f7f\u7528\u4e8c\u8fdb\u5236\u6587\u4ef6\u8fdb\u884c\u5b89\u88c5 \u6211\u4eec\u4e3a\u7528\u6237\u63d0\u4f9b\u53ef\u4ee5\u5feb\u901f\u5b89\u88c5torch_npu\u7684whl\u5b89\u88c5\u5305\u3002\u5728\u5b89\u88c5torch_npu\u4e4b\u524d\uff0c\u60a8\u9700\u8981\u5148\u5b89\u88c5CANN\u8f6f\u4ef6\u3002\u6607\u817e\u8f85\u52a9\u8f6f\u4ef6\u4e2d\u6709\u66f4\u591a\u5173\u4e8eCANN\u7684\u7248\u672c\u4fe1\u606f\u3002\u8bf7\u53c2\u8003CANN\u5b89\u88c5\u6307\u5357\u83b7\u53d6CANN\u5b89\u88c5\u5305\u3002 \u5b89\u88c5PyTorch \u901a\u8fc7 pip \u5b89\u88c5 PyTorch\u3002 aarch64: pip3 install torch==2.1.0 x86: pip3 install torch==2.1.0+cpu –index-url https:\/\/download.pytorch.org\/whl\/cpu \u82e5\u4f7f\u7528pip\u547d\u4ee4\u5b89\u88c5\u5931\u8d25\uff0c\u8bf7\u4f7f\u7528\u4e0b\u8f7d\u94fe\u63a5\u6216\u8fdb\u5165PyTorch\u5b98\u65b9\u7f51\u7ad9\u8fdb\u884c\u67e5\u8be2\u4e0b\u8f7d\u5bf9\u5e94\u7248\u672c\u3002 \u67b6\u6784 Python\u7248\u672c \u4e0b\u8f7d\u94fe\u63a5 x86 Python3.8 \u4e0b\u8f7d\u94fe\u63a5 x86 Python3.9 \u4e0b\u8f7d\u94fe\u63a5 x86 Python3.10 \u4e0b\u8f7d\u94fe\u63a5 aarch64 Python3.8 \u4e0b\u8f7d\u94fe\u63a5 aarch64 Python3.9 \u4e0b\u8f7d\u94fe\u63a5 aarch64 Python3.10 \u4e0b\u8f7d\u94fe\u63a5 \u5b89\u88c5torch_npu\u4f9d\u8d56 \u8fd0\u884c\u4ee5\u4e0b\u547d\u4ee4\u5b89\u88c5\u4f9d\u8d56\u3002 pip3 install pyyaml … Continue reading \u534e\u4e3aAscend NPU\u9002\u914d<\/span><\/a><\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_bbp_topic_count":0,"_bbp_reply_count":0,"_bbp_total_topic_count":0,"_bbp_total_reply_count":0,"_bbp_voice_count":0,"_bbp_anonymous_reply_count":0,"_bbp_topic_count_hidden":0,"_bbp_reply_count_hidden":0,"_bbp_forum_subforum_count":0,"footnotes":""},"categories":[48,5],"tags":[],"class_list":["post-784","post","type-post","status-publish","format-standard","hentry","category-artificial-intelligence","category-linux"],"_links":{"self":[{"href":"http:\/\/madapapa.com\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/784","targetHints":{"allow":["GET"]}}],"collection":[{"href":"http:\/\/madapapa.com\/wordpress\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/madapapa.com\/wordpress\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/madapapa.com\/wordpress\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"http:\/\/madapapa.com\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=784"}],"version-history":[{"count":1,"href":"http:\/\/madapapa.com\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/784\/revisions"}],"predecessor-version":[{"id":785,"href":"http:\/\/madapapa.com\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/784\/revisions\/785"}],"wp:attachment":[{"href":"http:\/\/madapapa.com\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=784"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/madapapa.com\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=784"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/madapapa.com\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=784"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}