Failed to install Horovod

Environment:

1. Framework: TensorFlow, PyTorch
2. Framework version:
   • TensorFlow: 2.18.0
   • PyTorch: 2.4.1
3. Horovod version: Attempting to install latest via pip
4. MPI version: Microsoft MPI (attempted with version 10)
5. CUDA version: 11.8
6. NCCL version: None
7. Python version: 3.11.9
8. Spark / PySpark version: N/A
9. Ray version: N/A
10. OS and version: Windows 10
11. GCC version: Not installed (Windows environment)
12. CMake version: 3.30

Checklist:

1. Did you search issues to find if somebody asked this question before? Yes
2. If your question is about hang, did you read this doc? Yes
3. If your question is about docker, did you read this doc? N/A
4. Did you check if your question is answered in the troubleshooting guide? Yes

Bug report: I'm encountering issues installing Horovod on Windows 10 with TensorFlow and PyTorch frameworks. Here’s a summary of the setup and error details:

Steps Taken:

1. Set environment variables:
   cmd复制

   set HOROVOD_WITH_MPI=1
   set HOROVOD_WITH_CUDA=1
   set HOROVOD_CUDA_HOME=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8
   set MPI_HOME="E:/Program Files/Microsoft MPI"
   set MPIEXEC_EXECUTABLE="E:\Program Files\Microsoft MPI\Bin\mpiexec.exe"

2. Attempted installation command:
   bash复制

   pip install horovod --no-cache-dir

Error:

Could NOT find MPI (missing: MPI_CXX_FOUND)
CMake Error: The source directory "E:/Projects" does not appear to contain CMakeLists.txt.

Additional Details:

• Running mpiexec --version did not provide the expected version output.
• Verified CUDA 11.8 installation via nvcc --version.
• Using Microsoft MPI, but suspect compatibility issues with Horovod on Windows.