Worked on the jax-ml/jax repository to implement cluster reference transformation commutation within the Mosaic GPU framework, targeting improved GPU computation efficiency. The feature enabled specific cluster_ref transformations to commute, optimizing performance for Mosaic GPU workloads. The development process involved extensive use of Python and GPU programming, with a strong emphasis on unit testing to ensure correctness. Tests were updated and expanded to validate new output shapes and transformation behaviors, covering a range of edge cases. This work demonstrated a thorough approach to both feature implementation and validation, contributing to the robustness and performance of the Mosaic GPU workflow in jax.