shell and move around and then edit files.
BYOB: complexity without payoff
，详情可参考51吃瓜
Factorized embed, rotation Q (2 angles), tied embed+V dir, rank-1 MLP, parabolic head, sinusoidal PE (period 11)
You can start your design by choosing from a variety of ready-made templates, searching for a template matching your needs, or working with a blank template.