安装PyTorch-Geometric时导入torch-sparse触发段错误求助

问题现象 #

导入torch-sparse时直接触发段错误，导入pytorch_geometric也因依赖该库出现同样问题：

尝试导入依赖库的报错：

Python 3.9.12 (main, Apr  5 2022, 06:56:58) 
[GCC 7.5.0] :: Anaconda, Inc. on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import torch
>>> import torch_scatter
>>> import torch_cluster
>>> import torch_sparse
Segmentation fault (core dumped)

导入PyTorch-Geometric的报错：

Python 3.9.12 (main, Apr  5 2022, 06:56:58) 
[GCC 7.5.0] :: Anaconda, Inc. on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import torch_geometric
Segmentation fault (core dumped)

已尝试操作 #

• 最初怀疑GPU兼容性问题，更换过多个版本的依赖库
• 切换为纯CPU版本的PyTorch及PyTorch-Geometric相关库，问题未解决

环境信息 #

系统版本 #

Distributor ID:    Ubuntu
Description:    Ubuntu 22.04.1 LTS
Release:    22.04
Codename:   jammy

Conda环境依赖 #

# Name                    Version                   Build  Channel
_libgcc_mutex             0.1                        main  
_openmp_mutex             5.1                       1_gnu  
blas                      1.0                         mkl  
brotlipy                  0.7.0           py310h7f8727e_1002  
bzip2                     1.0.8                h7b6447c_0  
ca-certificates           2022.07.19           h06a4308_0  
certifi                   2022.9.24       py310h06a4308_0  
cffi                      1.15.1          py310h74dc2b5_0  
charset-normalizer        2.0.4              pyhd3eb1b0_0  
cpuonly                   2.0                           0    pytorch
cryptography              37.0.1          py310h9ce1e76_0  
fftw                      3.3.9                h27cfd23_1  
idna                      3.4             py310h06a4308_0  
intel-openmp              2021.4.0          h06a4308_3561  
jinja2                    3.0.3              pyhd3eb1b0_0  
joblib                    1.1.1           py310h06a4308_0  
ld_impl_linux-64          2.38                 h1181459_1  
libffi                    3.3                  he6710b0_2  
libgcc-ng                 11.2.0               h1234567_1  
libgfortran-ng            11.2.0               h00389a5_1  
libgfortran5              11.2.0               h1234567_1  
libgomp                   11.2.0               h1234567_1  
libstdcxx-ng              11.2.0               h1234567_1  
libuuid                   1.0.3                h7f8727e_2  
markupsafe                2.1.1           py310h7f8727e_0  
mkl                       2021.4.0           h06a4308_640  
mkl-service               2.4.0           py310h7f8727e_0  
mkl_fft                   1.3.1           py310hd6ae3a3_0  
mkl_random                1.2.2           py310h00e6091_0  
ncurses                   6.3                  h5eee18b_3  
numpy                     1.23.3          py310hd5efca6_0  
numpy-base                1.23.3          py310h8e6c178_0  
openssl                   1.1.1q               h7f8727e_0  
pip                       22.2.2          py310h06a4308_0  
pycparser                 2.21               pyhd3eb1b0_0  
pyg                       2.1.0           py310_torch_1.12.0_cpu    pyg
pyopenssl                 22.0.0             pyhd3eb1b0_0  
pyparsing                 3.0.9           py310h06a4308_0  
pysocks                   1.7.1           py310h06a4308_0  
python                    3.10.6               haa1d7c7_0  
pytorch                   1.12.1             py3.10_cpu_0    pytorch
pytorch-cluster           1.6.0           py310_torch_1.12.0_cpu    pyg
pytorch-mutex             1.0                         cpu    pytorch
pytorch-scatter           2.0.9           py310_torch_1.12.0_cpu    pyg
pytorch-sparse            0.6.15          py310_torch_1.12.0_cpu    pyg
readline                  8.1.2                h7f8727e_1  
requests                  2.28.1          py310h06a4308_0  
scikit-learn              1.1.2           py310h6a678d5_0  
scipy                     1.9.1           py310hd5efca6_0  
setuptools                63.4.1          py310h06a4308_0  
six                       1.16.0             pyhd3eb1b0_1  
sqlite                    3.39.3               h5082296_0  
threadpoolctl             2.2.0              pyh0d69192_0  
tk                        8.6.12               h1ccaba5_0  
tqdm                      4.64.1          py310h06a4308_0  
typing_extensions         4.3.0           py310h06a4308_0  
tzdata                    2022e                h04d1e81_0  
urllib3                   1.26.12         py310h06a4308_0  
wheel                     0.37.1             pyhd3eb1b0_0  
xz                        5.2.6                h5eee18b_0  
zlib                      1.2.13               h5eee18b_0  

解决建议 #

1. 严格匹配PyTorch与PyG依赖版本 #

你的环境中PyTorch是1.12.1，但PyG相关库是基于torch_1.12.0编译的，小版本差异可能导致兼容性问题。重新安装匹配版本：

先卸载现有PyG相关库：

conda uninstall pyg pytorch-cluster pytorch-scatter pytorch-sparse

再用pip精准安装对应版本：

pip install torch-scatter torch-sparse torch-cluster torch-spline-conv torch-geometric -f https://data.pyg.org/whl/torch-1.12.1+cpu.html

2. 降级GCC版本 #

Ubuntu 22.04默认GCC版本较高，部分PyG编译库对高版本GCC兼容性不佳。安装GCC 9并临时切换：

安装GCC 9：

sudo apt install gcc-9 g++-9

临时切换环境变量：

export CC=/usr/bin/gcc-9
export CXX=/usr/bin/g++-9

之后重新执行上述pip安装命令。

3. 清理conda缓存并重建环境 #

conda缓存可能存在损坏，清理后重建环境：

conda clean --all
conda create -n pyg_env python=3.10
conda activate pyg_env
conda install pytorch==1.12.1 cpuonly -c pytorch
pip install torch-scatter torch-sparse torch-cluster torch-spline-conv torch-geometric -f https://data.pyg.org/whl/torch-1.12.1+cpu.html

4. 补充系统底层依赖 #

段错误可能源于缺少系统依赖，安装以下库：

sudo apt install libopenblas-dev liblapack-dev libeigen3-dev

内容的提问来源于stack exchange，提问作者Alex Davies