Simplified, reproducible reimplementation of DeepMind's GraphCast weather model using graph neural networks (JAX · Haiku · Jraph). Runs on a single GPU, with honest negative-results reporting and detailed failure-mode analysis of a heavily scaled-down forecasting model.
machine-learningdeep-learningreproducibilitymessage-passinghaikujaxgraph-neural-networksweather-forecastinggnnautoregressive-modelsweatherbenchjraphgraphcasticosahedral-mesh
-
Updated
Jul 28, 2026 - Python