Serving PyTorch models in the cloud

Deep learning is computationally expensive and therefore demands powerful and sophisticated computational hardware. Not everyone has access to a local machine that has enough CPUs and GPUs to train gigantic deep learning models in a reasonable amount of time. Furthermore, we cannot guarantee 100 percent availability for a local machine that is serving a trained model for inference.

For reasons such as these, cloud computing platforms are a vital alternative for both training and serving deep learning models.

In this section, we will discuss how to use PyTorch with some of the most popular cloud platforms – AWS, Google Cloud, and Microsoft Azure. We will explore the different ways of serving a trained PyTorch model on each of these platforms. The model-serving exercises we discussed in earlier sections of this chapter were executed on a local machine. The goal of this section is to enable you to perform similar exercises using virtual...