Install MLflow with pip install mlflow, start a local tracking server on port 5000, and log your first run from Python. For local development, MLflow’s environment guide recommends SQLite; use a shared server, Docker Compose, or Databricks Managed MLflow when you need collaboration or managed infrastructure.
Choose where MLflow will store runs
MLflow Tracking records experiment metadata—such as parameters, metrics, and run details—and can store model artifacts. The tracking URI determines where tracking data is written, so choose a backend before connecting multiple users or machines.
| Option | Setup and persistence | Collaboration, artifacts, and ownership |
|---|---|---|
| Local file store | Fastest to try; when no tracking URI is specified, MLflow can use local file storage and create an mlruns directory. The environment guide says this backend is in Keep-the-Light-On mode. |
Best suited to simple work on one machine, not a shared team setup. Move toward a database for ongoing local development. |
| SQLite | A lightweight persistent database; MLflow’s environment guide recommends sqlite:///mlflow.db for quickstarts and local development. |
Useful for local experiments. It does not by itself provide a shared, secured team service or remote artifact store. |
| Self-hosted server | Run an MLflow tracking server and configure a database backend and artifact storage as needed. | Provides a shared UI and API, but your team owns operations and security configuration. Production guidance uses a remote artifact root such as s3://my-mlflow-bucket/artifacts. |
| Docker Compose | The official Compose flow starts MLflow with PostgreSQL and MinIO and exposes port 5000. | A reproducible fuller local stack with a database and object store; you still operate the stack and should configure access appropriately. |
| Databricks Managed MLflow | Uses Databricks workspace setup and authentication rather than a server you operate yourself. | Managed infrastructure and workspace integration; availability and use are subject to Databricks account and program terms. |
For the shortest path, use SQLite and a local server. Use local file storage only when the simplicity of a one-machine experiment matters more than a database-backed setup. Choose a shared or managed option when collaborators need access to the same tracking service.
Install MLflow and start the local UI
-
Install the Python package in the environment where you will run your training code:
pip install mlflow.Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy. -
Start a tracking server with SQLite as its backend:
mlflow server --backend-store-uri sqlite:///mlflow.db --port 5000Leave this process running while you use the UI or send tracking requests. The SQLite URI follows MLflow’s environment-guide recommendation for quickstarts and local development.
-
Open http://localhost:5000 in a browser on the same machine. The server listens on port 5000 unless you choose another port.
If you omit a tracking URI in a simple local script, MLflow can create an mlruns file-store directory instead. That is convenient for trying the APIs, but the environment guide describes file storage as being in Keep-the-Light-On mode and recommends moving toward a database.
Log and inspect a first experiment
MLflow’s Tracking Quickstart describes its goal as “to provide a quick guide to the most essential core APIs of MLflow Tracking.” A minimal workflow is to name an experiment, enable supported framework autologging, train a model, then inspect the recorded run.
-
In your Python training script, select the local server and create or select an experiment:
import mlflow import mlflow.sklearn mlflow.set_tracking_uri("http://localhost:5000") mlflow.set_experiment("MLflow Quickstart") mlflow.sklearn.autolog() -
Run your usual scikit-learn training code in the same script after enabling autologging. For supported frameworks, autologging can capture parameters, metrics, model artifacts, and metadata without requiring you to log each item manually.
-
Return to the UI at http://localhost:5000, open the
MLflow Quickstartexperiment, and select the run to review its recorded details and artifacts.Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.Rank #3
The tracking server must be running at the URI your code uses. If you change the server port or host, update the URI accordingly.
Load a logged model for inference
When autologging or your own logging code has recorded a model, load it through MLflow’s generic pyfunc interface using the run’s model URI. The exact URI depends on the run ID and artifact path shown for that run.
import mlflow.pyfunc
model = mlflow.pyfunc.load_model("runs:/<run_id>/model")
predictions = model.predict(input_data)Replace <run_id> with the actual run ID, and use the model artifact path recorded by your run if it differs from model. Provide input in the format expected by the logged model.
Connect code to a remote tracking server
For a shared MLflow server, point your code at its reachable HTTP address rather than assuming localhost. For example, replace the host below with the server address supplied by its administrator:
Recommended Free Tools
Rank #4
import mlflow
mlflow.set_tracking_uri("http://<server-host>:5000")Alternatively, configure the tracking URI through the environment variable before running the script:
export MLFLOW_TRACKING_URI=http://<server-host>:5000The tracking URI directs experiment metadata to that server. A remote deployment also needs artifact storage that the server and users can access; production self-hosting guidance shows a remote artifact root such as s3://my-mlflow-bucket/artifacts. The team operating a self-hosted service is responsible for its operations and security configuration.
When Docker Compose or Databricks is a better fit
Use Docker Compose for a reproducible local stack
The official MLflow Compose flow brings up MLflow with PostgreSQL and MinIO and exposes port 5000. Choose it when you want a fuller, repeatable local environment rather than a single SQLite-backed server. Follow the official Compose setup for its files and commands, then connect clients to the exposed server address.
Use Databricks Managed MLflow for workspace-managed infrastructure
Databricks documents Managed MLflow with workspace setup and Databricks authentication. It can suit teams that want MLflow integrated with their Databricks workspace without operating the tracking infrastructure themselves. The workspace and account terms govern access; this is not simply a public local server.
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallCrashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteUse a self-hosted service for production control
Self-hosting gives an organization control over its shared tracking endpoint and storage choices, but it also means owning deployment, security, database, and artifact-storage decisions. MLflow’s self-hosting guidance includes remote artifact roots such as S3 and an official Helm chart for Kubernetes.
Common setup problems
-
The browser cannot reach the UI: confirm the server process is still running and that the browser’s host and port match the server command. A server running on another machine is not reached through that machine’s
localhost. -
A script creates local runs instead of showing them in the UI: set
mlflow.set_tracking_uri("http://localhost:5000")in the script or exportMLFLOW_TRACKING_URIbefore launching it. Confirm the URI points to the running server. -
The run is missing an expected model or metric: verify that the framework’s autologging integration is enabled before training and that the run completed. Autologging applies to supported frameworks; use explicit MLflow logging APIs for information your workflow does not capture automatically.
Recommended: PC Feels Slow? A Free Scan Shows What's Dragging Windows Down →Recommended: Crashes or Glitches? A Free Driver Scan Usually Finds the Culprit →Recommended: Fix Windows Errors and Clear Junk Files in Minutes - Free Scan →Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy. -
Remote runs appear but artifacts cannot be used: metadata tracking and artifact access are related but distinct concerns. Check that the configured artifact store is reachable with the required access from the relevant server and client environment.
Quick Recap
Bestseller No. 1Bestseller No. 3SaleBestseller No. 4
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
