Can we use trained models with Googles Coral TPU

fastai
1

Hi just started looking into the fast.ai library and course. However, as I have the specific need to run any models on an edge device I wanted to know a few things before investing time into learning fast.ai.

  • Can you run a trained model using the Google Coral TPU?
  • Do you need to install PyTorch etc on the device to run the models?
  • Will the newer Swift version of the API support this?

Thanks

2

fastaiv1 has some support for TPU, but fastai2 does not.

3

Thanks, though I find it odd that the next version wouldn’t have this, given a lot of solutions need to have an aspect that runs on the edge. So the need to reuse models with TF can be quite important.

Though I am assuming the Swift version will support Corals TPU given its build on top of Tensorflow?