ImportError: Cannot load DGL C++ sparse library

Questions
#1

Traceback (most recent call last):
File “dgl_batch_c.py”, line 2, in
import dgl.sparse as dglsp
File “/home/lixintao/anaconda3/envs/dgl/lib/python3.8/site-packages/dgl/sparse/init.py”, line 43, in
load_dgl_sparse()
File “/home/lixintao/anaconda3/envs/dgl/lib/python3.8/site-packages/dgl/sparse/init.py”, line 40, in load_dgl_sparse
raise ImportError(“Cannot load DGL C++ sparse library”)
ImportError: Cannot load DGL C++ sparse library

dgl version:1.1.0+cu102
torch version:1.12.1+cu102
cuda version:11.1

I want to run HGNN using dgl.sparse,I have intsalled all the packages ,but when I run the code ,an issue above occured,can anyone tell me how to solve it?

#2

DGL Sparse requires torch >=1.13.0. Could you try with latest DGL with torch 1.13.x or 2.0.x?

#3

Okay,I 'll have a try

#4

As my CUDA version is 11.1,the highest pytorch version I can install is 1.12.1,so I can’t install torch 1.13.x or 2.0.x.
And I can’t update my CUDA version,because I’m using server and I don’t have the right to update. Is there any other solution ?

#5

you could try to build DGL from source with latest torch you have. Please refer to Install and Setup — DGL 1.1.2post1 documentation for the instructions. But I’m afraid build would fail.

Another option is docker. You could update the cuda driver, torch, DGL in the docker as you want.

#6

Okay,I’ll try to use docker.Thanks

#7

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.