VoidPadding: Let [VOID] Handle Padding in Masked Diffusion Language Models so that [EOS] Can Focus on Semantic Termination
Published in COLM 2026 Workshop on Efficient Reasoning (Spotlight), 2026
VoidPadding separates padding from semantic termination in masked diffusion language models with a dedicated [VOID] token and adaptive response-canvas expansion.
Recommended citation: Liu et al., 2026.
Download Paper