Why mask diffusion does not work (2510.03289v1)

Abstract: The main advantages of diffusion LLMs over autoregressive (AR) models lie in their ability to support parallel generation and bidirectional attention, enabling a more controllable generation process. In recent years, open-source mask diffusion LLMs have emerged, most of which are based on a variant known as absorbing diffusion. However, this paper demonstrates why mask diffusion faces inherent difficulties in achieving parallel generation and bidirectional attention. We also propose the most effective training and inference strategies for mask diffusion.