euler

Why we included this project

Teams building graph neural networks at scale will find Euler a genuinely different tool from the usual TensorFlow or PyTorch pipelines. Instead of loading the whole graph into memory, it runs a distributed graph engine on the data side and exposes graph operators that GNN layers call, which matters when the adjacency structure is too large for one machine. You get ready-made GCN and GraphSAGE implementations and random-walk based embedding methods, plus a message passing interface for writing custom models, so it is easy to start from known recipes before adapting them. Attribute indexing and a GQL-style query layer make the framework practical on real graphs that carry rich node and edge features. One honest caveat: Euler is an older, largely dormant project from Alibaba, so treat it as a reference design or a foundation rather than something you will get active maintenance on.