We propose Hierarchical Diffusion Language Models, which is a discrete diffusion with a general time-varying next semantic scale prediction process for language modeling.