This container image includes the complete source of the NVIDIA version of PyG in /opt/pyg/pytorch_geometric. It is prebuilt and installed as a system Python module. The /workspace/examples folder is copied from /opt/pyg/pytorch_geometric/examples for users starting to run PyG. For an introductory example about training a GCN, try python /workspace/examples/gcn.py
See /workspace/README.md for details
The container also uses torch-geometric 2.4.0 & pyg-lib 0.2.0. This container also contains the GNN Platform (/opt/pyg/gnn-platform), an NVIDIA project that provides a low-code API for rapid GNN experimentation and training/deploying end-to-end GNN pipelines. Examples can be found at /workspace/gnn-platform-examples. For more details about the GNN Platform, see /opt/pyg/gnn-platform/README.md.
This container is built on NVIDIA PyTorch container (see contents of PyTorch container)
Key Features in this release:
NVIDIAs syngen tool for synthetic graph data generation. See README.md for details
- We recommend using torch.compile on your GNN models for accelerating any example effortlessly.
- Example: model = torch.compile(model)
There are two main prerequisites for this container:
- NVIDIA Drivers NVIDIA Drivers 515.48.07+ is recommended. For a complete list of supported drivers (older versions), see the CUDA Application Compatibility topic. For more information, see CUDA Compatibility and Upgrades.
- Docker (19.03+)
- (Optional) NGC API Key for logging in to NVIDIA's registry. Details are available here.
Running the container
Use the following commands to run the container, where <xx.xx> is the container version.
docker run --gpus all -it --rm nvcr.io/nvidia/pyg:xx.xx-py3
For example, 23.11 for November 2023 release:
docker run --gpus all -it --rm nvcr.io/nvidia/pyg:23.11-py3