CUDA_HOME environment variable is not set

[zzd1992]

I work on ubuntu16.04, cuda9.0 and Pytorch1.0.
When I run your example code cuda/setup.py:

Traceback (most recent call last):
  File "setup.py", line 9, in <module>
    'lltm_cuda_kernel.cu',
  File "/usr/local/lib/python2.7/dist-packages/torch/utils/cpp_extension.py", line 476, in CUDAExtension
    library_dirs += library_paths(cuda=True)
  File "/usr/local/lib/python2.7/dist-packages/torch/utils/cpp_extension.py", line 549, in library_paths
    paths.append(_join_cuda_home(lib_dir))
  File "/usr/local/lib/python2.7/dist-packages/torch/utils/cpp_extension.py", line 1121, in _join_cuda_home
    raise EnvironmentError('CUDA_HOME environment variable is not set. '
EnvironmentError: CUDA_HOME environment variable is not set. Please set it to your CUDA install root.

However, I am sure cuda9.0 in my computer is installed correctly.
How to fix this problem?

[goldsborough]

That's odd. Maybe you have an unusual install location for CUDA. Either way, just setting CUDA_HOME to your cuda install path before running python setup.py should work: CUDA_HOME=/path/to/your/cuda/home python setup.py install

[zzd1992]

Thanks. My problem is solved.

[DaLei001]

@zzd1992 Could you tell how to solve the problem about "the CUDA_HOME environment variable is not set"?

[zzd1992]

@DaLei001

CUDA_HOME=/path/to/your/cuda/home python setup.py install

This is worked in my situation.

[mmahdavian]

I am getting this error in a conda env on a server and I have cudatoolkit installed on the conda env. What should the CUDA_HOME be in my case.

[parasdegregore]

@mmahdavian cudatoolkit probably won't work for you, it doesn't provide access to low level c++ apis. You'd need to install CUDA using the official method.

[ShoufaChen]

I got a similar error when using pycharm, with unusual cuda install location.

I just add the CUDA_HOME env and solve this problem.

[fengyuan1993012]

The problem could be solved by installing the whole cuda through the nvida website. and when installing it, you may come across some problem. then https://askubuntu.com/questions/1280205/problem-while-installing-cuda-toolkit-in-ubuntu-18-04/1315116#1315116?newreg=ec85792ef03b446297a665e21fff5735 the answer may be to help you. GOOD LUCK

[saadam1n]

I am getting this issue on Windows. The weird thing is, I have set CUDA_HOME and when I run the code from pytorch to find CUDA_HOME outside of the pip installer, it works perfectly. Does pip hide environment variables?

[saadam1n]

Upon further debugging, it seems that install of torch that pip installs in its isolated build environment cannot detect CUDA. However, the install of torch outside it can. Unfortunately, building with --no-build-isolation leads to a ton of other errors.