How to use Triton compiler to compile the entire model structure instead of the kernel function?

[Mark3123]

Hello everyone, I found that the Triton compiler only compiles kernel functions, not the overall structure of the algorithm model. If I want to compile the overall structure of the algorithm model, is there any recommended method? Is it feasible for me to use pytorch+triton? If I use pytorch+triton, how can I save the algorithm model structure and weights into a data format corresponding to ttir?