google-deepmind/jraph

Health v2 · maintenance only

Flags: archived

Adoption not part of the score

1474 stars · 106 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

Jraph is a lightweight Graph Neural Network library built on JAX, providing a GraphsTuple data structure, graph utilities, and a zoo of forkable GNN models. It integrates with Haiku or Flax for parameter management and supports distributed training of large graphs.

Use cases

• train graph neural networks in jax
• batch and pad variable-sized graphs for jit compilation
• run message passing neural networks on molecules
• distribute large graph networks across multiple devices
• learn graph neural networks with tutorials

When to choose

• you use JAX and need GNN primitives you can fork and adapt
• you want lightweight graph data structures and utilities without a heavy framework
• you need to scale large MPNNs across devices with pmap

When to avoid

• you want a framework that manages model parameters for you
• you need PyTorch or TensorFlow GNN support
• you need actively developed features with frequent releases

Facets

library · maturity maintenance

machine-learning deep-learning machine-learning deep-learning graph-processing python cross-platform graph-neural-networks jax message-passing graphs gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="google-deepmind/jraph")