Scaling Laws Predict Particle Physics Foundation Model Performance

A joint model-and-data scaling law for transformer models trained on collider jets can predict loss within one percent across three orders of magnitude of compute. This allows researchers to forecast downstream physics performance, such as background rejection, before committing to expensive training runs.