Initial uTVM doc generation

microTVM
#1

Couple of things to bring up.

  1. The python docs for python/tvm/micro aren’t being generated near as I can tell. Seems like that would be worth it to enable.

  2. We don’t as of yet have any tutorials for micro. While it might be early in the eyes of some, I kinda like small focused bits of code to drive use cases, fuel test cases and tutorials. I’ve got an initial one pulling in a tflite model mostly written following the style used in the tutorial/frontend directory. I’m not exactly clear how the web pages from <some_random_tutorial.py> are generated tho. If it actually runs the python within the file that’ll crash and burn without OpenOCD, a board etc.

Thoughts?

#2

Given the silence I went ahead and got ahold of Zhao Wu who has written a number of tutorials for TVM. He gave me some suggestions which I’ll utilize and then push my tutorial up for consideration.

#3

some initial thoughts here.

Our python docs are generated with sphinx and code blocks are executed as part of doc generation, so we’ll need to be able to run the example code on x86. Right now we’ve been focused on running on the STM-Nucleo board that we currently support, so while there is a way to run code on x86, it might not be especially robust right now. Moving to the RPC-based on-device runtime would make it more robust, so we’re hoping to release some tutorials after that effort’s finished. That said, we’re happy to have more tutorials if you’d like to contribute one!

I’ll see if I can get the python docs to generate for python/tvm/micro in the meanwhile.