Distributed GraphFlow (DGF)?

 Distributed GraphFlow (DGF) is an open-source Python library developed by Google CoreML and Google Research to train, evaluate, and deploy Graph Neural Networks (GNNs) at scale, succeeding TensorFlow GNN. [1, 2, 3]

Key Features & Architecture

Capabilities: Covers the full GNN lifecycle across tabular, relational, and temporal data using a Scikit-learn-style simple API or an advanced JAX/Flax layer-based API. [1, 2]
Scalability & Interoperability: Scales from single-machine execution (up to 1 billion edges) to trillion-edge graphs via Apache Beam distributed sampling, while integrating with PyTorch Geometric, NetworkX, and TensorFlow. [1]
Setup: Installs via PyPI for Python 3.11–3.13 on Linux and macOS, with full details available on the GitHub Repository and Google Cloud Blog. [1, 2]