Run WholeGraph in a container

Use a RAPIDS container when you want a preconfigured Linux, CUDA, and RAPIDS environment. Select an image whose RAPIDS and CUDA versions match the desired WholeGraph packages.

The current image names and launch commands are maintained in the RAPIDS installation guide. Follow that guide instead of pinning an old PyTorch or CUDA base image in application documentation.

WholeGraph requires:

• access to NVIDIA GPUs through the NVIDIA Container Toolkit;

• shared memory sized for the workload; and

• the networking and IPC access required by NCCL for multi-GPU or multi-node execution.

After starting the container, verify the installed version:

python -c "import pylibwholegraph; print(pylibwholegraph.__version__)"

If the selected RAPIDS image does not include WholeGraph, install the matching packages using the WholeGraph installation instructions.