arXiv · cs.AI · 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 小时前模型评分46
北大团队提出 SFA：受统计物理启发的扩散 Transformer 注意力机制
SFA: A Statistical Physics-Inspired Diffusion Transformer for Efficient and High-Quality Image Generation
导读
北京大学团队提出受统计物理启发的无标度注意力（SFA，Scale-Free Attention），在 262,144 个 token 下相比 FlashAttention 实现 357.3 倍前向和 265.6 倍反向加速。
分类置信度
中 · 待人工复核
- 只有单一媒体来源，需要人工复核。
来源级别：媒体与个人
事件类型：研究发布
证据摘录
- “Inspired by statistical physics and the widespread occurrence of scale-free laws in nature, we introduce Scale-Free Attention (SFA), which preserves fine local interactions while summarizing distant regions at progressively coarser scales.”
来源：arXiv · cs.AI · arxiv.org