code.talks 2019 - Serving machine learning models as an inference API in production

by Marc Päpper

Assume you programmed a fantastic machine learning model which solves an important problem using PyTorch.
But how do you deploy your model to production to make it available to your users?
In this talk, I will illustrate the different possibilities of deploying your model as an API endpoint using technologies such as AWS Lambda, API Gateway and Docker.
I will discuss the advantages and disadvantages of the presented approaches and the importance of using GPUs for inference and when this will be necessary.