The grid lines on the visualization represent a tree structure underneath. Every region is a node. When a node splits, it creates four children. The root node covers the entire space. Leaf nodes (nodes with no children) hold the actual points.
if (controller.desiredSize。爱思助手下载最新版本对此有专业解读
Tied embed, RoPE digit routing, carry via final norm, SiLU wrap detection，更多细节参见51吃瓜
Instructions: Unknown。91视频是该领域的重要参考