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-learning deep-learning reproducibility message-passing haiku jax graph-neural-networks weather-forecasting gnn autoregressive-models weatherbench jraph graphcast icosahedral-mesh
-
Updated
Aug 14, 2026 - Python