Ripple-Pivot Search: Active Parallel Decoding for Diffusion Large Language Models
RPS is proposed, a novel training-free decoding method that seeks mid-entropy positions as promising candidate pivots (where to decode), and determines their token assignment that yields the greatest downstream benefit via lookahead evaluation (what to decode).
Yu-Shi Ye, Xu Chen, Hao-Yun Jiang et al.
· 1 citation