I’ve found that autoTVM uses the predefined template in the source code, and my goal is to deploy my model to my specific board.
It is possible and already successful to use autoTVM(autotuning) to tune the template to my specific board. But it seems to have no support for using autoScheduling with microTVM now.
My questions are those:
- Is it necessary to use autoScheduling for my specific board? Or using the template in TVM is enough?
- Is it possible to call the TVM operator to generate the new schedule to my board by autoScheduling and replace the predefined one in the source code, then use autoTVM to tune it?
- Or the worst one, if there is no way to use autoScheduling with microTVM and I have to use the original TVM way to tune it, is there any recommended OS that I can use on the STM32 board? I’m trying to use uCLinux but I’m afraid it can not support Python 3.x cause it seems been deprecated for 10 years.
Thanks a lot.