Building a dockerfile for a Python application is easy to get started with, but usually they are not optimized for faster build times and smaller image sizes. This is especially true for applications which require non-Python dependencies, e.g. C/C++/Rust extensions.

This repository has a sample app using FastAPI, which serves a machine learning model trained on MNIST digits. In addition, it has a small C++ extension in the extension folder. The has a dockerfile which has a base setup, and then other optimizations, numbered 1 through 6 that we can do on top of it to improve the resultant container.

Note that this app will only run on Python 3.12.

1. Install poetry

2. Install dependencies

make setup

3. Run server at localhost:8000

uvicorn src.server:app

4. Check https://localhost:8000/docs/