作者位次 #4/7ACL 2026Findings· Findings of ACL
Improving the Throughput of Diffusion-based Large Language Models via a Training-Free Confidence-Aware Calibration
Jucheng Shen · Gaurav Sarkar · Yeonju Ro · Sharath Nittur Sridhar · Zhangyang Wang · Aditya Akella · Souvik Kundu
摘要
We present CadLLM, a training-free method to accelerate the inference throughput of diffusion-based LLMs (dLLMs). We first investigate on the dynamic nature of the token unmasking confidence across blocks and steps. Based on this observation, we then present a lightweight adaptive approach that c...
效率与系统OpenReview 项目页