Reproduce “Scaling Laws of Global Weather Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yuejiang Yu, Langwen Huang, Alexandru Calotoiu, Torsten Hoefler
The authors test how model width, depth, training data, and parameter budgets affect six-hour weather-model validation loss across five model families. At roughly matched sizes, tested wider configurations generally outperform deeper ones; Aurora has a data-scaling exponent of about 0.51, with tenfold more data reducing validation loss by up to 3.2-fold. These are fitted, configuration-dependent findings, not causal or operational guarantees.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.