A comprehensive framework for training graph neural networks on dynamic graphs.
NB: this is an ongoing work.
Our development environment:
- Ubuntu 20.04LTS
- g++ 9.4
- CUDA 11.3 / 11.6
- cmake 3.23
Dependencies:
- torch >= 1.10
- dgl (CUDA version)
Compile and install:
python setup.py install
For debug mode,
DEBUG=1 pip install -v -e .
cd scripts/ && ./download_data.sh
Multi-GPU single machine
Training TGN model on the REDDIT dataset with LRU feature cache (cache ratio=0.2) on four GPUs.
./scripts/run_offline.sh TGN REDDIT LRUCache 0.2 4
Distributed training
Training TGN model on the REDDIT dataset with LRU feature cache (cache ratio=0.2) and hash-based graph partitioning strategy.
./scripts/run_offline.sh TGN REDDIT LRUCache 0.2 hash