arXiv · cs.GR· Zixuan Huang (School of Computer Science, Peking University, Beijing, China, School of Physics, Peking University, Beijing, China), Tianhao You (School of Computer Science, Peking University, Beijing, China), Guozhang Chen (School of Computer Science, Peking University, Beijing, China, National Engineering Research Center of Visual Technology, Peking University, Beijing, China)·· 7 小时前
北大团队提出 SFA：受统计物理启发的扩散 Transformer 注意力机制，262K token 下前向加速 357 倍
SFA: A Statistical Physics-Inspired Diffusion Transformer for Efficient and High-Quality Image Generation
AI 导读
北京大学团队提出 Scale-Free Attention（SFA），通过均值池化、token 计数校正和无标度空间层级上的共享 softmax，将注意力复杂度降至 O(N log N)。
来源：arXiv · cs.GR · arxiv.org