モデルAPI用のDockerfileを書く

Your model API runs on your laptop but breaks elsewhere. A Dockerfile is a recipe that builds the exact same environment everywhere.

Every Dockerfile starts FROM a base image. A slim Python image gives you the interpreter without a bloated operating system.

The WORKDIR instruction picks the folder your commands run in. It is created if missing, so later paths stay clean and predictable.

Copy requirements.txt alone before your code. Docker caches this layer, so installs are skipped when only your app changes.

Run pip install inside the image. The --no-cache-dir flag skips the pip cache and keeps the final image a little smaller.

Now COPY the rest of your project in. Doing this after the install means code edits do not bust the dependency cache layer.

The EXPOSE instruction declares which port your API listens on. It is documentation for humans and tools, not an actual open.

The CMD sets what runs when the container starts. Here it launches Uvicorn to serve your FastAPI model app.

Inside a container you must serve on 0.0.0.0, not 127.0.0.1. Otherwise the host cannot reach your model through the mapped port.

Turn the recipe into an image with docker build. The -t flag tags it with a name you can run and push later.

Each instruction becomes a cached layer. Order them from least to most frequently changed so rebuilds stay fast.

Let us check the most cache-friendly build order.

You wrote a Dockerfile: base image, workdir, cached install, copy, expose, and a start command. Your model API now builds the same anywhere. 🐳