作者位次 #4/13ACL 2026Findings· Findings of ACL
On-Policy Self-Distillation for Efficient Diffusion Language Models with Early-Stage Calibration
Huaisheng Zhu · MingYu Liu · Junze Liu · Zhen Ge · Tian Wang · Jiri Gesi · Dakuo Wang · Weiqi Zhang · Houyu Zhang · Yufan Guo · Xian Li · Bing Yin · sujay sanghavi
摘要
Diffusion Large Language Models (DLLMs) have recently achieved strong performance, e.g., masked diffusion models (MDMs) can surpass autoregressive models (ARMs) in various tasks. However, DLLMs often struggle with inaccurate early-stage predictions due to limited context, which hinders both the m...