ImportError: libcusparse.so.10.0: cannot open shared object file: No such file or directory about pytorch_sparse HOT 8 CLOSED

If your CUDA version is 10.0 and you use PyTorch 1.5, you may need to install from source using pip install torch-sparse command. We provide 9.2, 10.1 and 10.2 wheels for PyTorch 1.5, and 9.2, 10.0 and 10.1 wheels for PyTorch 1.4. The easier way is to just upgrade your CUDA though.

from pytorch_sparse.

You may need to add the lib directory of miniconda to LD_LIBRARY_PATH. This directory should contain libcusparse.so.10.

from pytorch_sparse.

Hi, can you follow this guide and see if that fixes your issue? Editing LD_LIBRARY_PATH may be crucial here.

from pytorch_sparse.

This issue was solved !
It caused by my conda virtual env .
I used /usr/lobal/cuda to compiler the torch-sparse .
However , my codes depend on the virtual env when it runs . . .
So I just moved the libcusparse.so.10.0 files from /usr/lobal/cuda/lib64 to my virtual env , then it was solved!
Thanks so much !! !

from pytorch_sparse.

I run into the same problem. My Cuda version is 10.0. Is there any way that I could install it with this version or should I upgrade?

from pytorch_sparse.

If your CUDA version is 10.0 and you use PyTorch 1.5, you may need to install from source using pip install torch-sparse command. We provide 9.2, 10.1 and 10.2 wheels for PyTorch 1.5, and 9.2, 10.0 and 10.1 wheels for PyTorch 1.4. The easier way is to just upgrade your CUDA though.

Yes, I upgraded my cuda version and that saved me a lot of time playing around with library versions

from pytorch_sparse.

This issue was solved !
It caused by my conda virtual env .
I used /usr/lobal/cuda to compiler the torch-sparse .
However , my codes depend on the virtual env when it runs . . .
So I just moved the libcusparse.so.10.0 files from /usr/lobal/cuda/lib64 to my virtual env , then it was solved!
Thanks so much !! !

I am using virtual environment too. But there is not any cuda directory in local directory. my torch cuda version however is 10.1. How can I fix this?

from pytorch_sparse.

Appending $CUDA_PATH/lib64 (which has libcusparse.so.10.0) to $LD_LIBRARY_PATH helped me resolve this issue.

export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$CUDA_PATH/lib64

from pytorch_sparse.