Barrel files are convenient, but they often come with trade-offs including: Performance and memory: they artificially inflate the module graph and slow down startup times, HMR, and CI pipelines.

Timber takes a trained ML model — XGBoost, LightGBM, scikit-learn, CatBoost, ONNX (tree ensembles, linear models, SVMs), or a URDF robot description — runs it through a multi-pass optimizing compiler, ...