WholeGraph

WholeGraph provides distributed memory, tensor, embedding, and graph operations for large-scale GNN workflows. It is maintained in the cuGraph-GNN repository and is split into two packages:

• pylibwholegraph is the Python and PyTorch-facing library.

• libwholegraph is the native C/CUDA implementation.

cuGraph-PyG can use WholeGraph to store features and graph structures across multiple GPUs and nodes.

WholeGraph documentation

• Concepts and architecture
  • WholeGraph overview
  • WholeMemory concepts
  • WholeMemory implementation details
• Installation
  • Install WholeGraph
  • Run WholeGraph in a container
  • Build WholeGraph from source
• WholeGraph API
  • pylibwholegraph API
  • libwholegraph API