TypeError: unsupported operand type(s) for +: 'NoneType' and 'str'

[KrisWongz]

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.

Traceback (most recent call last):
File "setup.py", line 35, in
_, bare_metal_major, _ = get_cuda_bare_metal_version(cpp_extension.CUDA_HOME)
File "setup.py", line 14, in get_cuda_bare_metal_version
raw_output = subprocess.check_output([cuda_dir + "/bin/nvcc", "-V"], universal_newlines=True)
TypeError: unsupported operand type(s) for +: 'NoneType' and 'str'

[6cigarette]

I have the same questions.Now do you find the solution?

[Symbolk]

This occurs when you try to install apex on a non-GPU machine.

Run the following command would work like a charm:

cd apex
export TORCH_CUDA_ARCH_LIST="compute capability" 
python3 setup.py install

[BUAADreamer]

Hello, did you solve this problem? I tried export TORCH_CUDA_ARCH_LIST="compute capability" but it can't work either.

[tzcskys]

Hello, did you solve this problem? I tried export TORCH_CUDA_ARCH_LIST="compute capability" but it can't work either.

Hi, I also met the problem, did you find the solution? thanks

[tzcskys]

Hello, did you solve this problem? I tried export TORCH_CUDA_ARCH_LIST="compute capability" but it can't work either.

Hi, I also met the problem, did you find the solution? thanks

Hi, I found out the problem with my environment is that the cuda installed with conda is not the full version, which means the nvcc is not installed. So use "conda install -c nvidia cuda-nvcc" first and try again, works for me. Put my solution here in case someone meet the same problem.

[BUAADreamer]

Hello, did you solve this problem? I tried export TORCH_CUDA_ARCH_LIST="compute capability" but it can't work either.

Hi, I also met the problem, did you find the solution? thanks

Hi, I found out the problem with my environment is that the cuda installed with conda is not the full version, which means the nvcc is not installed. So use "conda install -c nvidia cuda-nvcc" first and try again, works for me. Put my solution here in case someone meet the same problem.

Yeah, I find the reason is I didn't install the whole package, I finally use conda env and pip3 install from source

[xywhat]

Hello, did you solve this problem? I tried export TORCH_CUDA_ARCH_LIST="compute capability" but it can't work either.

Hi, I also met the problem, did you find the solution? thanks

Hi, I found out the problem with my environment is that the cuda installed with conda is not the full version, which means the nvcc is not installed. So use "conda install -c nvidia cuda-nvcc" first and try again, works for me. Put my solution here in case someone meet the same problem.

Yeah, I find the reason is I didn't install the whole package, I finally use conda env and pip3 install from source

Hello, did you solve this problem? I tried export TORCH_CUDA_ARCH_LIST="compute capability" but it can't work either.

Hi, I also met the problem, did you find the solution? thanks

Hi, I found out the problem with my environment is that the cuda installed with conda is not the full version, which means the nvcc is not installed. So use "conda install -c nvidia cuda-nvcc" first and try again, works for me. Put my solution here in case someone meet the same problem.

you saved me!

[Fanstering]

Hello, did you solve this problem? I tried export TORCH_CUDA_ARCH_LIST="compute capability" but it can't work either.

Hi, I also met the problem, did you find the solution? thanks

Hi, I found out the problem with my environment is that the cuda installed with conda is not the full version, which means the nvcc is not installed. So use "conda install -c nvidia cuda-nvcc" first and try again, works for me. Put my solution here in case someone meet the same problem.

Thanks, I encountered the same problem, and successfully solved it with your method

[olayinkaajayi]

Hello, did you solve this problem? I tried export TORCH_CUDA_ARCH_LIST="compute capability" but it can't work either.

Hi, I also met the problem, did you find the solution? thanks

Hi, I found out the problem with my environment is that the cuda installed with conda is not the full version, which means the nvcc is not installed. So use "conda install -c nvidia cuda-nvcc" first and try again, works for me. Put my solution here in case someone meet the same problem.

Thank you so much! This appears to work for me.

[WilliamTambellini]

python could nt find nvcc. Could need:

PATH=/usr/local/cuda/bin/:$PATH pip ...

[MSaqibShah]

Got it sorted after trying out different things and digging into the web for answers. Ended up fixing it by installing the whole CUDA toolkit manually from this link. Figured I'd drop it here in case anyone else is pulling their hair out over the same issue. Cheers!

[gndps]

For some reason the CUDA_HOME is none check is not working leading to this misleading error statement.
It was happening because CUDA_HOME was not found in setup.py
i.e from torch.utils.cpp_extension import CUDA_HOME was None
To solve it, I installed CUDA from here https://developer.nvidia.com/cuda-downloads
Then tried the apex installation command in a new shell

[PussyCat0700]

Hello, did you solve this problem? I tried export TORCH_CUDA_ARCH_LIST="compute capability" but it can't work either.

Hi, I also met the problem, did you find the solution? thanks

Hi, I found out the problem with my environment is that the cuda installed with conda is not the full version, which means the nvcc is not installed. So use "conda install -c nvidia cuda-nvcc" first and try again, works for me. Put my solution here in case someone meet the same problem.

Hi, How can I specify version of nvcc installed with this command?

[SuperIRabbit]

In my case running module load cuda/{YOUR_CUDA_VERSION} solves the problem (inspired by the answers above about nvcc)