search

Xtra-Computing / thundersvm

ThunderSVM: A Fast SVM Library on GPUs and CPUs
Apache License 2.0
1.57k stars 218 forks source link

trouble installing #256

Open pritamqu opened 2 years ago

pritamqu commented 2 years ago

i am trying to install thundersvm using pip in linux system with cuda 10.2, it gives me this error while trying to use. /lib/python3.7/site-packages/thundersvm/libthundersvm.so: cannot open shared object file: No such file or directory

however, i do have that file in that location, can you please help me with this? __init__.py libthundersvm.so __pycache__ thundersvm.py

thanks!

ericlaycock commented 1 year ago

Unfortunately you'll have to go back to Cuda version 9.0 for ThunderSVM to work - that's what ended up working for me.

Specifically, I executed this code in Google Colab:

!wget https://developer.nvidia.com/compute/cuda/9.0/Prod/local_installers/cuda-repo-ubuntu1704-9-0-local_9.0.176-1_amd64-deb

!dpkg -i cuda-repo-ubuntu1704-9-0-local_9.0.176-1_amd64-deb

!ls /var/cuda-repo-9-0-local | grep .pub

!apt-key add /var/cuda-repo-9-0-local/7fa2af80.pub

!apt-get update

!sudo apt-get install cuda-9.0

!pip install thundersvm #!!nvcc --version should return cuda9.0 by this point

from thundersvm import SVC

You should be able to run this without exclamation marks in your code.

Kurt-Liuhf commented 1 year ago

Hi @ericlaycock, I have tried building ThunderSVM (on Linux) using CUDA 11.8 and it succeeded. The wheel currently provided was built upon CUDA 9.0. So I suggest you build the wheel on your own, if necessary. Thanks.