Just provide an easy way to export trained models in pure PyTorch format.
Then we can directly use Pytorch mobile.
This should already be doable. learn.save('model.pth') will save the model state_dict.
image classifiers can be exported easily, just follow the steps from the Pytorch mobile website. I am having trouble exporting detectors like Retinanet.
Have a look at the docs of https://course.fast.ai/deployment_seeme_ai deploying your fastai model on the web and mobile.
TLDR: Once uploaded, your model is automatically converted to the different formats and you use your model on iOS and Android apps without any further development.
Disclaimer: I’m the creator of SeeMe.ai, but very happy to assist or answer questions. 
I wrote a post with a step by step guide for begginer of how to train using FastAI + run on a device natively using Pytorch Mobile:
It’s pretty straightforward, as some users already said, but maybe it helps someone without a very deep understanding (like me).
Essentially you need to save the learn.model (pytorch model) and import and use with Pytorch Mobile