pytorch fail to install with glibc version conflicts

NVIDIA / MinkowskiEngine

Minkowski Engine is an auto-diff neural network library for high-dimensional sparse tensors

https://nvidia.github.io/MinkowskiEngine

Other

2.48k stars 367 forks source link

pytorch fail to install with glibc version conflicts #354

Closed TerenceCYJ closed 3 years ago

TerenceCYJ commented 3 years ago

Error Package wheel conflicts for: python=3.8 -> pip -> wheel pip -> wheelThe following specifications were found to be incompatible with your system:

feature:/linux-64::__glibc==2.31=0
feature:|@/linux-64::__glibc==2.31=0
pytorch=1.7.1 -> cudatoolkit[version='>=11.0,<11.1'] -> __glibc[version='>=2.17|>=2.17,<3.0.a0']
torchvision -> __glibc[version='>=2.17|>=2.17,<3.0.a0']

Your installed version is: 2.31

To Reproduce

Just follow https://github.com/NVIDIA/MinkowskiEngine#cuda-102

conda create -n py3-mink python=3.8
conda activate py3-mink

conda install openblas-devel -c anaconda
conda install pytorch=1.7.1 torchvision cudatoolkit=10.2 -c pytorch -c conda-forge

OS: Ubuntu 20.04
Python version: 3.8.8
CUDA version: 10.2
NVIDIA Driver version: 460.67
GCC version: 7.5.0

TerenceCYJ commented 3 years ago

The same error when using CUDA 11.2 and gcc 9.3

abahnasy commented 3 years ago

I'm also facing the same problem with CUDA 11.2 and gcc 9.3. Have you found any solution?

TerenceCYJ commented 3 years ago

I'm also facing the same problem with CUDA 11.2 and gcc 9.3. Have you found any solution?

Hi, you can try v0.4.3.

lilanxiao commented 3 years ago

Hi, I just solved this issue. I think the reason is this command installed OpenBLAS in the conda environment, but it's not compatible with Pytorch: conda install openblas-devel -c anaconda My workaround is just to install OpenBLAS with a system package manager (like apt in Ubuntu) instead of using conda: sudo apt install libopenblas-dev Also, since the OpenBLAS is not installed by conda, you don't need this flag anymore. --blas_include_dirs=${CONDA_PREFIX}/include So just remove this flag and follow the instruction. Then the code should be compiled without any issue.

TerenceCYJ commented 3 years ago

With cuda11.0 gcc9.3

conda create -n py3-cu11 python=3.8
conda activate py3-cu11
conda install -c anaconda libopenblas
conda install pytorch=1.7.1 torchvision cudatoolkit=11.0 -c pytorch -c conda-forge
git clone https://github.com/NVIDIA/MinkowskiEngine.git
cd MinkowskiEngine
TORCH_CUDA_ARCH_LIST="5.2 6.0 6.1 7.0 7.5 8.0" python setup.py install