1 paper
Xuejie Liu, Yap Vit Chun, Yitao Liang +1
Diffusion Large Language Models (dLLMs) support arbitrary-order generation, yet their inference performance critically depends on the unmasking order. Existing strategies rely on h…