Course Wiki

Model Registry

The promotion path between training and serving: registered versions, stages, and a single source of the production model.

The model registry sits between experiment tracking and deployment. Trained models are registered as named models with versioned versions; a version moves through stages — staging, then production — and the deployment code loads from the registry by name and stage instead of by hard-coded run id.

Module 4 completes the loop: the web service queries the registry for the production version of the NY Taxi model and downloads it at startup. Moving a new model to production becomes a registry operation (promote the version), not a redeploy — the serving code always pulls whatever production points to.

Taught in

• MLOps Zoomcamp

  • Module 2: Experiment Tracking & Model Management
    • Model registry
  • Module 4: Model Deployment
    • Web-services: Getting the models from the model registry (MLflow)
  • Module 07: Course Project
    • Course Project ## Related concepts

• MLflow

• Experiment Tracking

• Model Deployment

• Model Monitoring

• MLOps Maturity Model