Apple researchers scale Categorical Flow Maps for faster discrete-data generation

Apple ML research shows scaled-up Categorical Flow Maps can generate discrete data (like text) via flow matching, enabling faster sampling than typical diffusion approaches.