Protecting Pytoch .pt model from decompilation in mobile app

I wonder if there is a good practice available on how to protect trained Pytorch models compiled to .pt files and deployed to an Android app from being readable to anyone decompiling the app?
Seems to be an important IP-protection issue.

I am writing an app that needs to be very performant and thus I move parts of the app logic into a C++ library that I include as an external package to my Android project.
Would it be theoretically possible to include my Pytorch models directly as a dependency to the C++ library and include it together with the C++ library as a compiled artifact in the Android project?

Anyone has any experience with something like this?