New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Please help me with OSError: libcusparse.so.10: cannot open shared object file: No such file or directory #1125
Comments
libcusparse.so and libcusparse.so.10 is already included in usr/local/cuda/lib64 |
Is this path added to |
thank you ,yes ,I check it,this is the result: /usr/lcoal/cuda-10.1/lib64: |
And what does |
Do you mean torch.version.cuda? the result of ' torch.version.cuda' is : 10.1 |
Can you do me a favor and see if you can install from |
Yes, I am glad to do it. what should I do?And how can I install from torch-scatter from source? |
where can I find the way to install torch-scatter from source codes? |
See here. |
I follow your instruction to install torch-scatter from source,.the process and result is as following, but it still has the problem mentioned above.What should I do? Your PyG is really important for me,thank you very much. ~$ python3 -c "import torch; print(torch.version)" |
Mh, this is super weird :( Do you have multiple CUDA versions installed on your system? There must be a reason why it tries to look in the wrong folder. |
no,I only have one CUDA version installed on my system. |
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 If someone need help, please contact me. |
Glad that you made it, but why did downgrading help? |
I want to ask the problem. In the usr/local have the CUDE and CUDE-10.0. Does the meaning represent the one code or multiply CUDE |
|
Could you please show your versions of torch-scatter, torch-sparse, torch-cluster, and torch-split-conv respectively? |
CUDA 10.0 init(torch.torch_sparse.storage.SparseStorage self, Tensor? row, Tensor? rowptr, Tensor? col, Tensor? value, (int, int)? sparse_sizes, Tensor? rowcount, Tensor? colptr, Tensor? colcount, Tensor? csr2csc, Tensor? csc2csr, bool is_sorted) -> (None):
|
Can you hack
|
OK! It works! Thank you! |
I've encountered a similar problem ( |
torch 1.5.0+cu101 I'm having a similar but slightly different issue: File "/home/aqd215/pyenv/py3.7/lib/python3.7/site-packages/torch_geometric/init.py", line 2, in |
Hey, I followed your installation, but the problem is still here. torch 1.4.0+cu100 |
Ubuntu 18.04
if you see "libcusparse.so.11", continue following steps: remove current cuda
install new cuda-10-2 add CUDA to PATH
|
First check this
I install |
I followed your steps now I am getting this error: Can you help me to fix it? |
Using PyTorch 1.4.0 is no longer supported and it is recommended to update your PyTorch version. I suggest to use PyTorch 1.6.0 (since wheels are not yet ready for PyTorch 1.7.0). You can then install PyG as described here: |
I followed your steps as describe in: https://github.com/rusty1s/pytorch_geometric#pytorch-160 My setting is as follows: Error: OSError: libcusparse.so.10: cannot open shared object file: No such file or directory Trace: I tried every other method that worked for others listed in this thread but still this issue persists for me. @rusty1s Could you please help me resolve this? This is key to proceed with my work. Thanks in advance! |
|
I wasn't able to figure that out but I managed to get it running with the +cpu version instead. |
Just an update: |
I had this problem when using conda. In my conda environment I had installed pytorch and torchvision with pip, but |
Had this problem on a cluster (cuda11, pytorch1.7). |
I am having a similar issue. My system has CUDA 11.2 installed:
But Torch uses 10.2:
I learned that torch only works with CUDA 10.2 and is installed with its own CUDA, so that makes sense. I install everything in a virtualenv. If I install torch geometric like this:
I get the error
So it appears like torch geometric is searching for the library in the system's cuda installation and not in the cuda installation that is delivered with pytorch. I can't find the library in the virtualenv folder though and I do not want to downgrade the CUDA on my system... Is there a way to point torch geometric to the cuda supplied with torch installed in a virtualenv? |
This should be possible by setting |
@rusty1s Tried to go through the answers, but couldn't find any that answer my question, so I would like to ask you here: Is it not possible to install pytorch-geometric without a system-wide install of cuda-toolkit (nvcc)? I tried to avoid installing cuda systemwide and use cuda-toolkit in conda environment only, but from the installation guide it seems a path to /usr/local/cuda is needed. I'd like to be able to install torch wheel built on specific cuda and do the same for pytorch-geometric. Is that possible? |
You can either use a system-wide CUDA installation or via If you want to build PyTorch from source using a specific CUDA installation, you can also do that for our extension packages (e.g., by cloning and running |
I think i found the answer.. the reason is you can use sudo like --> sudo python main.py or just install cudatoolkit inside the virtual environment like --> conda activate [your env] I used second way (installing cudatoolkit inside the environment) |
❓ Questions & Help
this is the traceback
`Traceback (most recent call last):
File "/home/yrwang/.local/lib/python3.6/site-packages/torch_sparse/init.py", line 15, in
library, [osp.dirname(file)]).origin)
File "/home/yrwang/.local/lib/python3.6/site-packages/torch/_ops.py", line 106, in load_library
ctypes.CDLL(path)
File "/usr/lib/python3.6/ctypes/init.py", line 348, in init
self._handle = _dlopen(self._name, mode)
OSError: libcusparse.so.10: cannot open shared object file: No such file or directory
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "", line 1, in
File "/home/yrwang/.local/lib/python3.6/site-packages/torch_geometric/init.py", line 2, in
import torch_geometric.nn
File "/home/yrwang/.local/lib/python3.6/site-packages/torch_geometric/nn/init.py", line 2, in
from .data_parallel import DataParallel
File "/home/yrwang/.local/lib/python3.6/site-packages/torch_geometric/nn/data_parallel.py", line 5, in
from torch_geometric.data import Batch
File "/home/yrwang/.local/lib/python3.6/site-packages/torch_geometric/data/init.py", line 1, in
from .data import Data
File "/home/yrwang/.local/lib/python3.6/site-packages/torch_geometric/data/data.py", line 7, in
from torch_sparse import coalesce
File "/home/yrwang/.local/lib/python3.6/site-packages/torch_sparse/init.py", line 23, in
raise OSError(e)
OSError: libcusparse.so.10: cannot open shared object file: No such file or directory
`
my cuda,cudnn is well installed :
nvcc -V nvcc: NVIDIA (R) Cuda compiler driver Copyright (c) 2005-2019 NVIDIA Corporation Built on Sun_Jul_28_19:07:16_PDT_2019 Cuda compilation tools, release 10.1, V10.1.243
my torch version:
>>> print(torch.__version__) 1.4.0
I use
`pip3 install torch-scatter==2.0.4+cu101 -f https://pytorch-geometric.com/whl/torch-1.4.0.html
pip3 install torch-sparse==0.6.1+cu101 -f https://pytorch-geometric.com/whl/torch-1.4.0.html
pip3 install torch-cluster==1.5.4+cu101 -f https://pytorch-geometric.com/whl/torch-1.4.0.html
pip3 install torch-spline-conv==1.2.0+cu101 -f https://pytorch-geometric.com/whl/torch-1.4.0.html
pip3 install torch-geometric`
to install torch-geometric, but the problem occur, thanks for helping me
The text was updated successfully, but these errors were encountered: