Part 3. Deployment

 

In part 3, we’ll look at how to get our models to the point where they can be used. We saw how to build models in the previous parts: part 1 introduced the building and training of models, and part 2 thoroughly covered an example from start to finish, so the hard work is done.

But no model is useful until you can actually use it. So, now we need to put the models out there and apply them to the tasks they are designed to solve. This part is closer to part 1 in spirit, because it introduces a lot of PyTorch components. As before, we’ll focus on applications and tasks we wish to solve rather than just looking at PyTorch for its own sake.

In part 3’s single chapter, we’ll take a tour of the PyTorch deployment landscape as of early 2020. We’ll get to know and use the PyTorch just-in-time compiler (JIT) to export models for use in third-party applications to the C++ API for mobile support.