Installation of DGl with Pytorch compatibility

Questions
#1

System Info:

  • Ubuntu 22.04.5, x86_64, jammy_jellyfish
  • CUDA GPU GA106 GeForce RTX 3060
  • gcc 11.4.0
  • NVIDIA driver version 560.35.03
  • CUDA version 12.6

Steps:

  1. new conda environment
  2. use previous Pytorch version:

conda install pytorch==2.2.1 torchvision==0.17.1 pytorch-cuda=12.1 -c pytorch -c nvidia — to install pytorch 2.2.1 with cuda 12.1

  1. had problem with numpy, install numpy again,

pip install "“numpy<2”

  1. install dgl 2.4.0,

pip install dgl -f https://data.dgl.ai/wheels/torch-2.2/cu121/repo.html

This is working for the time being. Not the latest version of numpy, pytorch and dgl.

#2

Could you provide any detailed error messages that you encountered during installation? Also, we listed system requirements here (including pytorch and cuda versions) Deep Graph Library. Please make sure the corresponding environment is compatible as indicated in the web page.

#3

Hi @dyru , thanks for asking. I was having difficulty installing latest DGL version with Pytorch compatibility. With the given system specs, I was eventually able solve the issue with the stated versions, but there was still Numpy incompatibility, that was solved by downgrading Numpy.
I wanted to write this for future reference for myself and if anyone else finds useful. These steps worked for the given specs.