apex安装

zslefour2025-01-08 14:36

安装过程复杂曲折，网上说的很多办法，貌似成功了，实际还是没起作用。

先说成功过程，执行下面命令，安装成功（当然，前提是你要先配置好编译环境）：

（我的环境：windows10 、embeded python3.11.9 、pytorch version: 2.5.1+cu124、RTX2080ti）

复制代码 
git clone https://github.com/NVIDIA/apex.git
 
cd apex

curl -L "https://github.com/NVIDIA/apex/files/14844602/windows_support.patch" | git apply

pip install -v --disable-pip-version-check --no-cache-dir --no-build-isolation --config-settings "--build-option=--cpp_ext" --config-settings "--build-option=--cuda_ext" .

这个信息，不知道是否说明我的显卡是2系（TORCH_CUDA_ARCH_LIST=7.5）？

成功安装的apex，在卸载apex，你会发现安装的apex含有下面部件：

---------------下面只是曲折过程，如无必要就不要浪费时间往下看了--------------

1、在comfyUI中，我安装了54个节点，只在安装了节点ComfyUI-PuLID-Flux-Enhanced后，在没有安装apex时，会有如下提示：

于是便想着要安装APEX，看看会不会速度快点，经过了漫长而曲折的过程，终于安装成功，但效果好像没什么变化。

从apex上述介绍中，是否表示，大多时候在Pytorch中已含相关代码？所以，究竟是否值得再花大力气去安装apex呢？还有待进一步探究。

我的环境是python3.11.9 、pytorch version: 2.5.1+cu124

2、首先是报如下错误：

复制代码 
  building 'amp_C' extension
  H:\ComfyUI_cp3.11.9\python_embeded\Lib\site-packages\torch\utils\cpp_extension.py:1964: UserWarning: TORCH_CUDA_ARCH_LIST is not set, all archs for visible cards are included for compilation.
  If this is not desired, please set os.environ['TORCH_CUDA_ARCH_LIST'].
    warnings.warn(
  Emitting ninja build file H:\ComfyUI_cp3.11.9\apex\build\temp.win-amd64-cpython-311\Release\build.ninja...
  Compiling objects...
  Allowing ninja to set a default number of workers... (overridable by setting the environment variable MAX_JOBS=N)
  [1/7] C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.4\bin\nvcc --generate-dependencies-with-compile --dependency-output H:\ComfyUI_cp3.11.9\apex\build\temp.win-amd64-cpython-311\Release\csrc/multi_tensor_scale_kernel.obj.d -std=c++17 --use-local-env -Xcompiler /MD -Xcompiler /wd4819 -Xcompiler /wd4251 -Xcompiler /wd4244 -Xcompiler /wd4267 -Xcompiler /wd4275 -Xcompiler /wd4018 -Xcompiler /wd4190 -Xcompiler /wd4624 -Xcompiler /wd4067 -Xcompiler /wd4068 -Xcompiler /EHsc -Xcudafe --diag_suppress=base_class_has_different_dll_interface -Xcudafe --diag_suppress=field_without_dll_interface -Xcudafe --diag_suppress=dll_interface_conflict_none_assumed -Xcudafe --diag_suppress=dll_interface_conflict_dllexport_assumed -IH:\ComfyUI_cp3.11.9\python_embeded\Lib\site-packages\torch\include -IH:\ComfyUI_cp3.11.9\python_embeded\Lib\site-packages\torch\include\torch\csrc\api\include -IH:\ComfyUI_cp3.11.9\python_embeded\Lib\site-packages\torch\include\TH -IH:\ComfyUI_cp3.11.9\python_embeded\Lib\site-packages\torch\include\THC "-IC:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.4\include" -IH:\ComfyUI_cp3.11.9\python_embeded\include -IH:\ComfyUI_cp3.11.9\python_embeded\Include "-IC:\Program Files (x86)\Microsoft Visual Studio\2022\BuildTools\VC\Tools\MSVC\14.40.33807\include" "-IC:\Program Files (x86)\Microsoft Visual Studio\2022\BuildTools\VC\Auxiliary\VS\include" "-IC:\Program Files (x86)\Windows Kits\10\include\10.0.22621.0\ucrt" "-IC:\Program Files (x86)\Windows Kits\10\\include\10.0.22621.0\\um" "-IC:\Program Files (x86)\Windows Kits\10\\include\10.0.22621.0\\shared" "-IC:\Program Files (x86)\Windows Kits\10\\include\10.0.22621.0\\winrt" "-IC:\Program Files (x86)\Windows Kits\10\\include\10.0.22621.0\\cppwinrt" -c H:\ComfyUI_cp3.11.9\apex\csrc\multi_tensor_scale_kernel.cu -o H:\ComfyUI_cp3.11.9\apex\build\temp.win-amd64-cpython-311\Release\csrc/multi_tensor_scale_kernel.obj -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -lineinfo -O3 --use_fast_math -DVERSION_GE_1_1 -DVERSION_GE_1_3 -DVERSION_GE_1_5 -DTORCH_API_INCLUDE_EXTENSION_H -DTORCH_EXTENSION_NAME=amp_C -D_GLIBCXX_USE_CXX11_ABI=0 -gencode=arch=compute_75,code=compute_75 -gencode=arch=compute_75,code=sm_75
  FAILED: H:/ComfyUI_cp3.11.9/apex/build/temp.win-amd64-cpython-311/Release/csrc/multi_tensor_scale_kernel.obj
  C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.4\bin\nvcc --generate-dependencies-with-compile --dependency-output H:\ComfyUI_cp3.11.9\apex\build\temp.win-amd64-cpython-311\Release\csrc/multi_tensor_scale_kernel.obj.d -std=c++17 --use-local-env -Xcompiler /MD -Xcompiler /wd4819 -Xcompiler /wd4251 -Xcompiler /wd4244 -Xcompiler /wd4267 -Xcompiler /wd4275 -Xcompiler /wd4018 -Xcompiler /wd4190 -Xcompiler /wd4624 -Xcompiler /wd4067 -Xcompiler /wd4068 -Xcompiler /EHsc -Xcudafe --diag_suppress=base_class_has_different_dll_interface -Xcudafe --diag_suppress=field_without_dll_interface -Xcudafe --diag_suppress=dll_interface_conflict_none_assumed -Xcudafe --diag_suppress=dll_interface_conflict_dllexport_assumed -IH:\ComfyUI_cp3.11.9\python_embeded\Lib\site-packages\torch\include -IH:\ComfyUI_cp3.11.9\python_embeded\Lib\site-packages\torch\include\torch\csrc\api\include -IH:\ComfyUI_cp3.11.9\python_embeded\Lib\site-packages\torch\include\TH -IH:\ComfyUI_cp3.11.9\python_embeded\Lib\site-packages\torch\include\THC "-IC:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.4\include" -IH:\ComfyUI_cp3.11.9\python_embeded\include -IH:\ComfyUI_cp3.11.9\python_embeded\Include "-IC:\Program Files (x86)\Microsoft Visual Studio\2022\BuildTools\VC\Tools\MSVC\14.40.33807\include" "-IC:\Program Files (x86)\Microsoft Visual Studio\2022\BuildTools\VC\Auxiliary\VS\include" "-IC:\Program Files (x86)\Windows Kits\10\include\10.0.22621.0\ucrt" "-IC:\Program Files (x86)\Windows Kits\10\\include\10.0.22621.0\\um" "-IC:\Program Files (x86)\Windows Kits\10\\include\10.0.22621.0\\shared" "-IC:\Program Files (x86)\Windows Kits\10\\include\10.0.22621.0\\winrt" "-IC:\Program Files (x86)\Windows Kits\10\\include\10.0.22621.0\\cppwinrt" -c H:\ComfyUI_cp3.11.9\apex\csrc\multi_tensor_scale_kernel.cu -o H:\ComfyUI_cp3.11.9\apex\build\temp.win-amd64-cpython-311\Release\csrc/multi_tensor_scale_kernel.obj -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -lineinfo -O3 --use_fast_math -DVERSION_GE_1_1 -DVERSION_GE_1_3 -DVERSION_GE_1_5 -DTORCH_API_INCLUDE_EXTENSION_H -DTORCH_EXTENSION_NAME=amp_C -D_GLIBCXX_USE_CXX11_ABI=0 -gencode=arch=compute_75,code=compute_75 -gencode=arch=compute_75,code=sm_75
  C:/Program Files (x86)/Microsoft Visual Studio/2022/BuildTools/VC/Tools/MSVC/14.40.33807/include\type_traits(1167): error: static assertion failed with "You've instantiated std::aligned_storage<Len, Align> with an extended alignment (in other words, Align > alignof(max_align_t)). Before VS 2017 15.8, the member "type" would non-conformingly have an alignment of only alignof(max_align_t). VS 2017 15.8 was fixed to handle this correctly, but the fix inherently changes layout and breaks binary compatibility (*only* for uses of aligned_storage with extended alignments). To suppress this error, please define either (1) _ENABLE_EXTENDED_ALIGNED_STORAGE to confirm that you want a type with an extended alignment, or (2) _DISABLE_EXTENDED_ALIGNED_STORAGE to get the old non-conforming behavior."
        static_assert(_Always_false<_Aligned>,
        ^
            detected during:
              instantiation of class "std::_Aligned<_Len, _Align, double, false> [with _Len=16ULL, _Align=16ULL]" at line 1185
              instantiation of class "std::_Aligned<_Len, _Align, int, false> [with _Len=16ULL, _Align=16ULL]" at line 1192
              instantiation of class "std::_Aligned<_Len, _Align, short, false> [with _Len=16ULL, _Align=16ULL]" at line 1199
              instantiation of class "std::_Aligned<_Len, _Align, char, false> [with _Len=16ULL, _Align=16ULL]" at line 1207
              instantiation of class "std::_Aligned_storage<_Len, _Align> [with _Len=16ULL, _Align=16ULL]" at line 1211
              instantiation of type "std::_Aligned_storage_t<16ULL, 16ULL>" at line 1215
              instantiation of class "std::aligned_storage<_Len, _Align> [with _Len=16ULL, _Align=16ULL]" at line 25 of H:\ComfyUI_cp3.11.9\apex\csrc\multi_tensor_scale_kernel.cu
              instantiation of "void load_store(T *, T *, int, int) [with T=float]" at line 64 of H:\ComfyUI_cp3.11.9\apex\csrc\multi_tensor_scale_kernel.cu
              instantiation of "void ScaleFunctor<in_t, out_t>::operator()(int, volatile int *, TensorListMetadata<2> &, float) [with in_t=float, out_t=float]" at line 38 of H:\ComfyUI_cp3.11.9\apex\csrc\multi_tensor_apply.cuh
              instantiation of "void multi_tensor_apply_kernel(int64_t, volatile int *, T, U, ArgTypes...) [with T=TensorListMetadata<2>, U=ScaleFunctor<float, float>, ArgTypes=<float>]" at line 109 of H:\ComfyUI_cp3.11.9\apex\csrc\multi_tensor_apply.cuh
              instantiation of "void multi_tensor_apply<depth,T,ArgTypes...>(int64_t, int64_t, const at::Tensor &, const std::vector<std::vector<at::Tensor, std::allocator<at::Tensor>>, std::allocator<std::vector<at::Tensor, std::allocator<at::Tensor>>>> &, T, ArgTypes...) [with depth=2, T=ScaleFunctor<float, float>, ArgTypes=<float>]" at line 124 of H:\ComfyUI_cp3.11.9\apex\csrc\multi_tensor_scale_kernel.cu



  1 error detected in the compilation of "H:/ComfyUI_cp3.11.9/apex/csrc/multi_tensor_scale_kernel.cu".

  multi_tensor_scale_kernel.cu

针对出现的第一个错误：

复制代码 
H:\ComfyUI_cp3.11.9\python_embeded\Lib\site-packages\torch\utils\cpp_extension.py:1964: UserWarning: TORCH_CUDA_ARCH_LIST is not set, all archs for visible cards are included for compilation.
  If this is not desired, please set os.environ['TORCH_CUDA_ARCH_LIST'].

参考文章《如何设置 TORCH_CUDA_ARCH_LIST 环境变量以优化 PyTorch 性能》，我的显卡是RTX2080Ti，所以设置：

复制代码 
set TORCH_CUDA_ARCH_LIST=7.5

再次编译，错误信息如下：

复制代码 
  已完成代码的生成
  building 'amp_C' extension
  Emitting ninja build file H:\ComfyUI_cp3.11.9\apex\build\temp.win-amd64-cpython-311\Release\build.ninja...
  Compiling objects...
  Allowing ninja to set a default number of workers... (overridable by setting the environment variable MAX_JOBS=N)
  [1/14] C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.4\bin\nvcc --generate-dependencies-with-compile --dependency-output H:\ComfyUI_cp3.11.9\apex\build\temp.win-amd64-cpython-311\Release\csrc/multi_tensor_l2norm_scale_kernel.obj.d -std=c++17 --use-local-env -Xcompiler /MD -Xcompiler /wd4819 -Xcompiler /wd4251 -Xcompiler /wd4244 -Xcompiler /wd4267 -Xcompiler /wd4275 -Xcompiler /wd4018 -Xcompiler /wd4190 -Xcompiler /wd4624 -Xcompiler /wd4067 -Xcompiler /wd4068 -Xcompiler /EHsc -Xcudafe --diag_suppress=base_class_has_different_dll_interface -Xcudafe --diag_suppress=field_without_dll_interface -Xcudafe --diag_suppress=dll_interface_conflict_none_assumed -Xcudafe --diag_suppress=dll_interface_conflict_dllexport_assumed -IH:\ComfyUI_cp3.11.9\python_embeded\Lib\site-packages\torch\include -IH:\ComfyUI_cp3.11.9\python_embeded\Lib\site-packages\torch\include\torch\csrc\api\include -IH:\ComfyUI_cp3.11.9\python_embeded\Lib\site-packages\torch\include\TH -IH:\ComfyUI_cp3.11.9\python_embeded\Lib\site-packages\torch\include\THC "-IC:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.4\include" -IH:\ComfyUI_cp3.11.9\python_embeded\include -IH:\ComfyUI_cp3.11.9\python_embeded\Include "-IC:\Program Files (x86)\Microsoft Visual Studio\2022\BuildTools\VC\Tools\MSVC\14.40.33807\include" "-IC:\Program Files (x86)\Microsoft Visual Studio\2022\BuildTools\VC\Auxiliary\VS\include" "-IC:\Program Files (x86)\Windows Kits\10\include\10.0.22621.0\ucrt" "-IC:\Program Files (x86)\Windows Kits\10\\include\10.0.22621.0\\um" "-IC:\Program Files (x86)\Windows Kits\10\\include\10.0.22621.0\\shared" "-IC:\Program Files (x86)\Windows Kits\10\\include\10.0.22621.0\\winrt" "-IC:\Program Files (x86)\Windows Kits\10\\include\10.0.22621.0\\cppwinrt" -c H:\ComfyUI_cp3.11.9\apex\csrc\multi_tensor_l2norm_scale_kernel.cu -o H:\ComfyUI_cp3.11.9\apex\build\temp.win-amd64-cpython-311\Release\csrc/multi_tensor_l2norm_scale_kernel.obj -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -lineinfo -O3 --use_fast_math -DVERSION_GE_1_1 -DVERSION_GE_1_3 -DVERSION_GE_1_5 -DTORCH_API_INCLUDE_EXTENSION_H -DTORCH_EXTENSION_NAME=amp_C -D_GLIBCXX_USE_CXX11_ABI=0 -gencode=arch=compute_75,code=sm_75
  FAILED: H:/ComfyUI_cp3.11.9/apex/build/temp.win-amd64-cpython-311/Release/csrc/multi_tensor_l2norm_scale_kernel.obj
  C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.4\bin\nvcc --generate-dependencies-with-compile --dependency-output H:\ComfyUI_cp3.11.9\apex\build\temp.win-amd64-cpython-311\Release\csrc/multi_tensor_l2norm_scale_kernel.obj.d -std=c++17 --use-local-env -Xcompiler /MD -Xcompiler /wd4819 -Xcompiler /wd4251 -Xcompiler /wd4244 -Xcompiler /wd4267 -Xcompiler /wd4275 -Xcompiler /wd4018 -Xcompiler /wd4190 -Xcompiler /wd4624 -Xcompiler /wd4067 -Xcompiler /wd4068 -Xcompiler /EHsc -Xcudafe --diag_suppress=base_class_has_different_dll_interface -Xcudafe --diag_suppress=field_without_dll_interface -Xcudafe --diag_suppress=dll_interface_conflict_none_assumed -Xcudafe --diag_suppress=dll_interface_conflict_dllexport_assumed -IH:\ComfyUI_cp3.11.9\python_embeded\Lib\site-packages\torch\include -IH:\ComfyUI_cp3.11.9\python_embeded\Lib\site-packages\torch\include\torch\csrc\api\include -IH:\ComfyUI_cp3.11.9\python_embeded\Lib\site-packages\torch\include\TH -IH:\ComfyUI_cp3.11.9\python_embeded\Lib\site-packages\torch\include\THC "-IC:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.4\include" -IH:\ComfyUI_cp3.11.9\python_embeded\include -IH:\ComfyUI_cp3.11.9\python_embeded\Include "-IC:\Program Files (x86)\Microsoft Visual Studio\2022\BuildTools\VC\Tools\MSVC\14.40.33807\include" "-IC:\Program Files (x86)\Microsoft Visual Studio\2022\BuildTools\VC\Auxiliary\VS\include" "-IC:\Program Files (x86)\Windows Kits\10\include\10.0.22621.0\ucrt" "-IC:\Program Files (x86)\Windows Kits\10\\include\10.0.22621.0\\um" "-IC:\Program Files (x86)\Windows Kits\10\\include\10.0.22621.0\\shared" "-IC:\Program Files (x86)\Windows Kits\10\\include\10.0.22621.0\\winrt" "-IC:\Program Files (x86)\Windows Kits\10\\include\10.0.22621.0\\cppwinrt" -c H:\ComfyUI_cp3.11.9\apex\csrc\multi_tensor_l2norm_scale_kernel.cu -o H:\ComfyUI_cp3.11.9\apex\build\temp.win-amd64-cpython-311\Release\csrc/multi_tensor_l2norm_scale_kernel.obj -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -lineinfo -O3 --use_fast_math -DVERSION_GE_1_1 -DVERSION_GE_1_3 -DVERSION_GE_1_5 -DTORCH_API_INCLUDE_EXTENSION_H -DTORCH_EXTENSION_NAME=amp_C -D_GLIBCXX_USE_CXX11_ABI=0 -gencode=arch=compute_75,code=sm_75
  C:/Program Files (x86)/Microsoft Visual Studio/2022/BuildTools/VC/Tools/MSVC/14.40.33807/include\type_traits(1167): error: static assertion failed with "You've instantiated std::aligned_storage<Len, Align> with an extended alignment (in other words, Align > alignof(max_align_t)). Before VS 2017 15.8, the member "type" would non-conformingly have an alignment of only alignof(max_align_t). VS 2017 15.8 was fixed to handle this correctly, but the fix inherently changes layout and breaks binary compatibility (*only* for uses of aligned_storage with extended alignments). To suppress this error, please define either (1) _ENABLE_EXTENDED_ALIGNED_STORAGE to confirm that you want a type with an extended alignment, or (2) _DISABLE_EXTENDED_ALIGNED_STORAGE to get the old non-conforming behavior."
        static_assert(_Always_false<_Aligned>,
        ^
            detected during:
              instantiation of class "std::_Aligned<_Len, _Align, double, false> [with _Len=16ULL, _Align=16ULL]" at line 1185
              instantiation of class "std::_Aligned<_Len, _Align, int, false> [with _Len=16ULL, _Align=16ULL]" at line 1192
              instantiation of class "std::_Aligned<_Len, _Align, short, false> [with _Len=16ULL, _Align=16ULL]" at line 1199
              instantiation of class "std::_Aligned<_Len, _Align, char, false> [with _Len=16ULL, _Align=16ULL]" at line 1207
              instantiation of class "std::_Aligned_storage<_Len, _Align> [with _Len=16ULL, _Align=16ULL]" at line 1211
              instantiation of type "std::_Aligned_storage_t<16ULL, 16ULL>" at line 1215
              instantiation of class "std::aligned_storage<_Len, _Align> [with _Len=16ULL, _Align=16ULL]" at line 24 of H:\ComfyUI_cp3.11.9\apex\csrc\multi_tensor_l2norm_scale_kernel.cu
              instantiation of "void load_store(T *, T *, int, int) [with T=float]" at line 75 of H:\ComfyUI_cp3.11.9\apex\csrc\multi_tensor_l2norm_scale_kernel.cu
              instantiation of "void L2NormScaleFunctor<in_t, out_t>::operator()(int, volatile int *, TensorListMetadata<2> &, float *, float *, float, __nv_bool, int) [with in_t=float, out_t=float]" at line 38 of H:\ComfyUI_cp3.11.9\apex\csrc\multi_tensor_apply.cuh
              instantiation of "void multi_tensor_apply_kernel(int64_t, volatile int *, T, U, ArgTypes...) [with T=TensorListMetadata<2>, U=L2NormScaleFunctor<float, float>, ArgTypes=<float *, float *, float, __nv_bool, int>]" at line 109 of H:\ComfyUI_cp3.11.9\apex\csrc\multi_tensor_apply.cuh
              instantiation of "void multi_tensor_apply<depth,T,ArgTypes...>(int64_t, int64_t, const at::Tensor &, const std::vector<std::vector<at::Tensor, std::allocator<at::Tensor>>, std::allocator<std::vector<at::Tensor, std::allocator<at::Tensor>>>> &, T, ArgTypes...) [with depth=2, T=L2NormScaleFunctor<float, float>, ArgTypes=<float *, float *, float, __nv_bool, int>]" at line 292 of H:\ComfyUI_cp3.11.9\apex\csrc\multi_tensor_l2norm_scale_kernel.cu



  1 error detected in the compilation of "H:/ComfyUI_cp3.11.9/apex/csrc/multi_tensor_l2norm_scale_kernel.cu".

  multi_tensor_l2norm_scale_kernel.cu

3、无奈之下，想着试试 torch.version = 2.5.1+cpu时，错误信息提示如下：

复制代码 
H:\ComfyUI_cp3.11.9\apex>..\python_embeded\python -m pip install -v --disable-pip-version-check --no-cache-dir --no-build-isolation --config-settings "--build-option=--cpp_ext" --config-settings "--build-option=--cuda_ext" .
Using pip 24.3.1 from H:\ComfyUI_cp3.11.9\python_embeded\Lib\site-packages\pip (python 3.11)
Processing h:\comfyui_cp3.11.9\apex
  Running command Preparing metadata (pyproject.toml)

  Warning: Torch did not find available GPUs on this system.
   If your intention is to cross-compile, this is not an error.
  By default, Apex will cross-compile for Pascal (compute capabilities 6.0, 6.1, 6.2),
  Volta (compute capability 7.0), Turing (compute capability 7.5),
  and, if the CUDA version is >= 11.0, Ampere (compute capability 8.0).
  If you wish to cross-compile for a single specific architecture,
  export TORCH_CUDA_ARCH_LIST="compute capability" before running setup.py.

  torch.__version__  = 2.5.1+cpu

4、文章《NVIDIA APEX安装完全指南及Megatron-LM/Pytorch运行问题解决》

切换到不同的分支：从 GitHub - NVIDIA/apex: A PyTorch Extension: Tools for easy mixed precision and distributed training in Pytorch 的标签页中，可以知道发布的版本，不至于在众多的分支中无所适从。

切换到分支：24.04.01-devel

不带参数，显示目前所选分支，git clone的初始分支为master：

复制代码 
git checkout 24.04.01-devel

依然报错。

5、参考文章《使用 Visual Studio 2022、CUDA 12.1 和 PyTorch 2.2.2 进行 Windows 编译的最新补丁 #1792》

以下只是想完全照抄原文，使用虚拟环境编译，最终因为我是使用的embeddable python，需要搭建太多东西，放弃，以下只是记录尝试过程中的一些知识。

文章里的路径，与我的环境并不一样，所以使用dir /x 查看短路径，并且只能一级一级查看：

最终，我的路径如下：

复制代码 
cmd /c 'C:\PROGRA~2\MICROS~3\2022\BUILDT~1\VC\AUXILI~1\Build\VCVARS~1.BAT amd64 & busybox64u sh -ilX'

将环境中文件复制到windows文件夹，文件夹需要双 \\ ：

建立的虚拟环境，以及git clone 的路径，在"C:\Users\你的用户名" 路径下。

受文章启发，直接在我的环境中再次安装（执行下面命令生成的whl要比文章头的命令较小，不知道是否因为设置了TORCH_CUDA_ARCH_LIST=7.5只针对我的RTX2080ti）：

复制代码 
git clone https://github.com/NVIDIA/apex.git

cd apex

curl -L "https://github.com/NVIDIA/apex/files/14844602/windows_support.patch" | git apply

set DISTUTILS_USE_SDK=1

set TORCH_CUDA_ARCH_LIST=7.5

pip install -v --disable-pip-version-check --no-cache-dir --no-build-isolation --config-settings "--build-option=--cpp_ext" --config-settings "--build-option=--cuda_ext" .

经过十几分钟，安装成功（如欲将编译成功的whl文件保存下来，需要在编译成功，要进行安装时，按下暂停键Pause，从缓存路径将文件复制，否则，装完后，文件会被删除）。

成功后，再次启动comfyUI，没有了开始提到的信息。

编译成功后，缓存里面有个文件origin.json，有如下内容，不知道与文章中提到的url有关联？：

{"dir_info": {}, "url": "file:///H:/ComfyUI_cp3.11.9/apex"}

相关推荐
数据智能老司机2 小时前
精通 Python 设计模式——分布式系统模式
python·设计模式·架构
数据智能老司机4 小时前
精通 Python 设计模式——并发与异步模式
python·设计模式·编程语言
数据智能老司机4 小时前
精通 Python 设计模式——测试模式
python·设计模式·架构
数据智能老司机4 小时前
精通 Python 设计模式——性能模式
python·设计模式·架构
c8i4 小时前
drf初步梳理
python·django
每日AI新事件4 小时前
python的异步函数
python
这里有鱼汤5 小时前
miniQMT下载历史行情数据太慢怎么办？一招提速10倍!
前端·python
databook14 小时前
Manim实现脉冲闪烁特效
后端·python·动效
程序设计实验室14 小时前
2025年了，在 Django 之外，Python Web 框架还能怎么选？
python
倔强青铜三16 小时前
苦练Python第46天：文件写入与上下文管理器
人工智能·python·面试
热门推荐
01GitHub 镜像站点02UV 工具安装与国内镜像源配置指南03Claude Code 平替：OpenAI发布 Codex CLI ，GPT-5 国内直接使用0446个Nano-banana 精选提示词，持续更新中05UV安装并设置国内源06保姆级教程：手把手教你用Dify实现完美多轮对话（附Chatflow和提示词）07A股预测还能更准？开源大模型Kronos带你跑通预测+回测全流程08Spec-Kit 使用指南09智能库存管理的需求预测模型：从业务痛点到落地代码的完整实践10KGG转MP3工具|非KGM文件|解密音频