1
Efficient Diffusion LLMs via Temporal-Spatial Parallel Decoding and Confidence Extrapolation
时空并行解码+置信外推，大幅提升扩散语言模型推理效率，架构创新值得深挖。
arXiv:2605.30753v1 Announce Type: new Abstract: Diffusion-based large language models (dLLMs) support parallel text generation via iterative denoising…