Alchemist: Unlocking Efficiency in Text-to-Image Model Training via Meta-Gradient Data Selection Paper • 2512.16905 • Published Dec 18, 2025 • 31