What is the difference between dgl 0.6.0.post1 and 0.6.0?

[LukeLIN-web]

I am trying to install dgl 0.6.0, but I have some problems.

1. What is the difference between dgl 0.6.0.post1 and 0.6.0?
2. How to install the cuda version from pip or conda? pip install dgl==0.6.0 is the CPU version.
3. Does it support cuda11.3?
4. apt-get install -y build-essential python3-dev will install python3.6, should his version be the same as the original python 3.8 ?

[BarclayII]

1. There is no difference other than supporting PyTorch 1.8 and CUDA 11.1.
2. Since 0.5 DGL is hosting pip wheels by its own due to size limitations: pip install dgl==0.6.0 -f https://data.dgl.ai/wheels/repo.html. If you want to install CUDA version, you can specify pip install dgl-cu111==0.6.0post1 -f https://data.dgl.ai/wheels/repo.html for CUDA 11.1 for instance.
3. I don’t think it supports CUDA 11.3; by the time 0.6.0post1 is released PyTorch doesn’t support CUDA 11.3 yet.
4. I think 0.6.0 supports Python 3.6 so it should be OK.