本研究提出广义邻域注意力（GNA）模型，以提高稀疏注意力机制的速度。通过在NVIDIA Blackwell架构上实现，GNA在多个生成模型中验证了28%至46%的速度提升，有效解决了注意力机制的O(n^2)复杂性问题。

A ?green sticky substance? discovered in the Logan Square neighborhood was determined to be guacamole, while a suspicious powder in the shoe department of the Marshall Field?s store at Water Tower...