#machine learning
Jan 2026
Window-Diffusion: Accelerating Diffusion Language Model Inference with Windowed Token Pruning and Caching
The token-level analysis reveals pronounced structural locality in DLM inference, and proposes a window-based token pruning and caching method for inference that achieves up to $99\times inference speedup while largely preserving generation performance.
Feng-Rui Zuo, Zhiwei Ke, Yi-Ming Liu et al.
· Advanced Parallel Programmin... · 3 citations