1
SAID: Accelerating Diffusion-Based Language Models via Scaffold-Aware Iterative Decoding
基于扩散的LLM解码太慢？SAID用“脚手架感知”迭代策略，大幅提速且不损质量。
arXiv:2606.04974v1 Announce Type: new Abstract: Diffusion large language models (DLLMs) enable non-autoregressive generation by iteratively denoising …