作者位次 #3/8ACL 2026Poster· Natural Language Processing
C redit D ecoding: Accelerating Parallel Decoding in Diffusion Large Language Models with Trace Credit
Kangyu Wang · Zhiyun Jiang · Haibo Feng · Weijia Zhao · Lin Liu · Jianguo Li · Zhenzhong Lan · Weiyao Lin
摘要
Diffusion large language models (dLLMs) generate text through iterative denoising. In commonly adopted parallel decoding schemes, each step confirms only high-confidence positions while remasking the others. By analyzing dLLM denoising traces, we uncover a key inefficiency: models often predict t...