dgNN is currently used for fast GAT training with much less GPU memory.
ActNN can reduce the training memory footprint by compressing the saved activations.
FastMoE can be used for GNN models wtih the Mixture of Experts (MoE).
For dgNN,
cd dgNN
python setup.py installFor ActNN,
cd actnn/actnn
pip install -v -e .For FastMoE,
cd fastmoe
python setup.py install