Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/characat0/mlops-fundamentals-homework/llms.txt

Use this file to discover all available pages before exploring further.

The model_serving Dockerfile produces a self-contained container image that includes the champion model artifact baked directly into the image layers — no MLflow server is required at runtime. The build-time download strategy means the resulting image can be shipped to any environment and will serve predictions immediately, without any dependency on the MLflow tracking server that registered the model.

​
Current Dockerfile

The skeleton Dockerfile sets up the Python environment and copies the application code, but leaves the model download step as a TODO for students to implement: 
FROM python:3.11-slim

WORKDIR /app

COPY requirements.txt .
RUN pip install -r requirements.txt

# TODO: Pull the @champion model from MLflow and bake it into the image.
#
# The model was registered in evaluate.py with the alias "champion".
# Use the mlflow CLI to download it here so the container is self-contained
# and doesn't need a live MLflow server at runtime.
#
# Hint:
#   ARG MLFLOW_TRACKING_URI=http://localhost:5000
#   RUN mlflow models download -m "models:/spotify-genre-classifier@champion" \
#       -d ./models --no-directory
#
# After this step, load the model in predict_genre() using:
#   mlflow.sklearn.load_model("./models")

COPY ./app ./app
RUN touch /app/app/__init__.py

CMD ["uvicorn", "app.main:app", "--host", "0.0.0.0", "--port", "8000"]

​
What to Implement (TODO)

To make the container self-contained, replace the TODO comment block with these two lines: 
ARG MLFLOW_TRACKING_URI=http://localhost:5000
RUN mlflow models download -m "models:/spotify-genre-classifier@champion" \
    -d ./models --no-directory
What each line does:
  • ARG MLFLOW_TRACKING_URI — Declares a build-time variable that is automatically picked up by the MLflow CLI as the MLFLOW_TRACKING_URI environment variable during the RUN step. The default value (http://localhost:5000) can be overridden with --build-arg at build time. This variable is not present in the final image, keeping credentials out of the runtime environment.
  • RUN mlflow models download — Invokes the MLflow CLI during the image build to fetch the model registered under the @champion alias. The -d ./models flag writes the artifact into /app/models/ inside the container. The --no-directory flag prevents MLflow from creating an extra subdirectory, so predict_genre() can load the model directly from ./models.
The MLflow tracking server must be running and the champion model must be registered with the @champion alias before running docker build. If MLflow is unreachable or the alias does not exist, the RUN mlflow models download step will fail and the build will abort.

​
Complete Dockerfile

After implementing the TODO, the full Dockerfile should look like this: 
FROM python:3.11-slim

WORKDIR /app

COPY requirements.txt .
RUN pip install -r requirements.txt

ARG MLFLOW_TRACKING_URI=http://localhost:5000
RUN mlflow models download -m "models:/spotify-genre-classifier@champion" \
    -d ./models --no-directory

COPY ./app ./app
RUN touch /app/app/__init__.py

CMD ["uvicorn", "app.main:app", "--host", "0.0.0.0", "--port", "8000"]

​
Build Commands

# Build — MLflow must be running and champion model must be registered
docker build \
  --build-arg MLFLOW_TRACKING_URI=http://localhost:5000 \
  -t genre-classifier .

# Run the container
docker run -p 8000:8000 genre-classifier

# Test the health endpoint
curl http://localhost:8000/health
The --build-arg MLFLOW_TRACKING_URI value must point to your actual MLflow tracking server. If MLflow is running on a different host or port, substitute that address here. The value is used only at build time and is not embedded in the final image.

​
Loading the Model in predict_genre()

Once the build step has downloaded the artifact to ./models/, the predict_genre() function in app/main.py loads it at inference time using the MLflow sklearn loader: 
import mlflow

model = mlflow.sklearn.load_model("./models")
The path ./models resolves to /app/models/ inside the container, which is exactly where the RUN mlflow models download -d ./models step placed the artifact. Because the model is already on disk, this call is fast — there is no network round-trip to MLflow at prediction time.
See Evaluate & Register Champion Model for the steps that register the trained model with the @champion alias in the MLflow Model Registry — a prerequisite before docker build will succeed.

Build docs developers (and LLMs) love

Get started for freeTalk to us