Hi,
Our team has been trying to run your codes for our experiments. We have issues following the versions given for pytorch, pyg and other pytorch related installations. Our installation details are as follows:
# Install CUDA toolkits, including nvcc, other packages similar to the one we have in the /usr/local/cuda/
conda install -c conda-forge cudatoolkit-dev=10.1.243
# Add the path to the cudatoolkit installed by conda (if needed)
# export PATH=~/.conda/envs/<env_name>/bin:$PATH
# export CPATH=~/.conda/envs/<env_name>/include:$CPATH
# export LD_LIBRARY_PATH=~/.conda/envs/<env_name>/lib64:$LD_LIBRARY_PATH
# export DYLD_LIBRARY_PATH=~/.conda/envs/<env_name>/lib$DYLD_LIBRARY_PATH
# Downgrade pip to get older versions of torch-scatter, otherwise, version not found
python -m pip install pip==20.0.2
pip install torch==1.4.0 torchvision==0.5.0
export TORCH=1.4.0
export CUDA=cu101
# add --no-index option if not found wheels
pip install torch-scatter==2.0.4 -f https://pytorch-geometric.com/whl/torch-${TORCH}+${CUDA}.html
pip install torch-sparse==0.6.0 -f https://pytorch-geometric.com/whl/torch-${TORCH}+${CUDA}.html
pip install torch-cluster==1.5.2 -f https://pytorch-geometric.com/whl/torch-${TORCH}+${CUDA}.html
# pip install torch-spline-conv -f https://pytorch-geometric.com/whl/torch-${TORCH}+${CUDA}.html
pip install torch-geometric==1.6.3
# OR just
pip install torch-geometric==1.6.3 torch-sparse==0.6.1 torch-scatter==2.0.4 torch-cluster==1.5.4 -f https://pytorch-geometric.com/whl/torch-${TORCH}+${CUDA}.html
The error we get seems to originate from CUDA:
Can you help us figure out what we may be doing wrong or what can we do to mitigate this issue? Do you have implementations with higher versions of torch and pyg?
Hi, Our team has been trying to run your codes for our experiments. We have issues following the versions given for pytorch, pyg and other pytorch related installations. Our installation details are as follows:
The error we get seems to originate from CUDA:
Can you help us figure out what we may be doing wrong or what can we do to mitigate this issue? Do you have implementations with higher versions of torch and pyg?