Closed DaRinker closed 1 year ago
Since I also run into this exact issue myself today, I wanted to leave a solution for posterity. If you're using conda/mamba, then you need to install cudatoolkit-dev
from the conda-forge channel, because the standard cudatoolkit
(listed in the yml file in the repo) is runtime-only and doesn't come with nvcc
which is needed here.
Hi @DaRinker I am currently trying to install OpenFold in my virtualVM and stumbled upon exactly the same error as you. I wanted to know what did you exactly mean by loading CUDA? I have already installed nvcc and the cudatoolkit-dev @lm-jkominek mentioned but it has not worked UnU].
I'm following the installation instructions (installing all this in a custom conda environment on a shared computing resource) Everything has gone well up until the OpenFold installation step.
Bug description After I run the command:
(esm_gpu)$ pip install 'openfold @ git+https://github.com/aqlaboratory/openfold.git@4b41059694619831a7db195b7e0988fc4ff3a307'
I get the error:Not sure what to do here? Anybody have any suggestions of how to fix? Thank you
EDIT: Looks like it has to do with my CUDA installation (or lack of one!) Loading in CUDA fixed the error.