Skip to content

How to Run MLflow Without Docker

By default, mlflow-experience is configured to connect to an MLflow server running in Docker at http://localhost:5000. However, you can track experiments completely offline on your local machine without running Docker.

Using a local SQLite database allows you to log parameters, metrics, and models to a single local database file (mlflow.db).

Step 1: Update Your YAML Configuration

Set tracking_uri to point to a local SQLite database file:

project:
  name: offline_exploration
  owner: data_scientist
  environment: local

mlflow:
  experiment_name: local_experiments
  # SQLite tracking URI (stores data in mlflow.db in your project root)
  tracking_uri: sqlite:///mlflow.db
  track_params: true
  track_metrics: true
  track_artifacts:
    - classification_report
    - confusion_matrix

Step 2: Run Your Experiment

Execute the training pipeline using uv:

uv run python -m main --config configs/my_experiment.yaml --logging-level INFO

MLflow will automatically initialize mlflow.db and log all run parameters, metrics, and artifacts locally.

Step 3: Launch the MLflow UI

View your runs, graphs, and metric comparisons by starting the MLflow UI pointed at the SQLite file:

uv run mlflow ui --backend-store-uri sqlite:///mlflow.db

Open http://127.0.0.1:5000 in your web browser.

Option 2: File-Based Tracking (mlruns Directory)

If you prefer not to use SQLite, you can write experiment runs directly to local JSON and YAML files inside an mlruns/ directory:

Step 1: Configure YAML

mlflow:
  experiment_name: file_based_experiments
  tracking_uri: file:./mlruns
  track_params: true
  track_metrics: true

Step 2: Run Your Pipeline

uv run python -m main --config configs/my_experiment.yaml

Step 3: Launch the MLflow UI

uv run mlflow ui

Open http://127.0.0.1:5000 to inspect your runs.

Comparison: When to Use What

Mode Tracking URI Prerequisites Best For
Docker Compose http://localhost:5000 Docker Desktop / Engine running Multi-user development, persistent PostgreSQL backend, production staging.
Local SQLite sqlite:///mlflow.db None (uv run) Single-developer laptop, quick experiments, offline working, no Docker overhead.
Local Files file:./mlruns None (uv run) Ephemeral prototyping, git-ignored local run folders.

Switching Back to Docker

When you are ready to switch back to the shared Docker stack:

  1. Start the Docker containers: 
    docker compose up -d
  2. Set tracking_uri: http://localhost:5000 in your YAML config.
  3. Run your pipeline!