pytorch_geometric: Please help me with OSError: libcusparse.so.10: cannot open shared object file: No such file or directory

Ubuntu 18.04 This is my procedure to fix this bug.

1. cd to /usr/local/cuda
2. run find -name libcus*

if you see “libcusparse.so.11”, continue following steps:

remove current cuda

1. sudo apt-get --purge remove "*cublas*" "cuda*" "nsight*"
2. sudo apt-get --purge remove "*nvidia*"

install new cuda-10-2 5. wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/cuda-ubuntu1804.pin 6. sudo mv cuda-ubuntu1804.pin /etc/apt/preferences.d/cuda-repository-pin-600 7. wget http://developer.download.nvidia.com/compute/cuda/10.2/Prod/local_installers/cuda-repo-ubuntu1804-10-2-local-10.2.89-440.33.01_1.0-1_amd64.deb 8. sudo dpkg -i cuda-repo-ubuntu1804-10-2-local-10.2.89-440.33.01_1.0-1_amd64.deb 9. sudo apt-key add /var/cuda-repo-10-2-local-10.2.89-440.33.01/7fa2af80.pub 10. sudo apt-get update 11. sudo apt-get -y install cuda-10-2

add CUDA to PATH

$ export PATH=/usr/local/cuda/bin:$PATH
$ echo $PATH
>>> /usr/local/cuda/bin:...

$ export CPATH=/usr/local/cuda/include:$CPATH
$ echo $CPATH
>>> /usr/local/cuda/include:...

$ export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH
$ echo $LD_LIBRARY_PATH
>>> /usr/local/cuda/lib64:...

$ export DYLD_LIBRARY_PATH=/usr/local/cuda/lib:$DYLD_LIBRARY_PATH
$ echo $DYLD_LIBRARY_PATH
>>> /usr/local/cuda/lib:...

I had this problem when using conda. In my conda environment I had installed pytorch and torchvision with pip, but pip uninstall pytorch and torchvision and then install them through conda instead solved the issue

Dear author, I made it. Thank you for your help. I downgrade CUDA to version 10.0, pytorch to version 1.4.0+cu100, torchvision to 0.5.0+cu100, and install torch-scatter torch-sparse torch-cluster torch-spline-conv from source. I tried to use version cu100 .whl to install them, but it doesn’t work. The commands I used are as follows:

pip3 install torch-scatter pip3 install torch-sparse pip3 install torch-cluster pip3 install torch-spline-conv pip3 install torch-geometric torch==1.4.0+cu100 torchvision==0.5.0+cu100 -f https://download.pytorch.org/whl/torch_stable.html

If someone need help, please contact me.

I think i found the answer… the reason is when you are inside the virtual environment, you can not open shared directory of cudatoolkit which is standing outside the virtual environment.

you can use sudo like

–> sudo python main.py

or just install cudatoolkit inside the virtual environment like

–> conda activate [your env] –> conda install -c anaconda cudatoolkit=10.1 –> python main.py

I used second way (installing cudatoolkit inside the environment) because “sudo python” uses “python” outside the virtual environment. So it’s not using the right version I want to use inside the virtual environment.

Had this problem on a cluster (cuda11, pytorch1.7). export LD_LIBRARY_PATH=somewhere/cudacore/11.0.2/lib64 solved the problem.