The process of restoring a compressed model's weights to higher numerical precision, improving quality but requiring more memory.