Skip to content

Fine-Tune Pipeline Documentation

Choreo Setup

Fine-Tune-Pipeline

Choreo setup¶

Before running the Fine-Tune Pipeline, you need to ensure the MLFlow server running in Choreo as well as the MLFLow database is running in choreo.

MLFlow database setup¶

Already have a database running?

If you already have a database running in Choreo, you can skip this step.

1. Deploy the database in Choreo¶

Go to Choreo
Go to your organization
Go to Resources > Databases
Create a new database
Choose PostgreSQL as the database type
Deploy the database

2. Power on the database¶

Go to the database you just created
Click on Power On Servicec to start the database
Copy the configuration details (host, port, username, password, database name)

MLFlow server setup¶

1. Deploy the server in Choreo¶

Already have a server running?

If you already have an MLFlow server running in Choreo, you can skip this step.

Go to Choreo
Create a project under your organization
Deploy the MLFlow server within the project as a web application component
Use this GitHub repository as the source: MLFlow Server

2. Configure the MLFlow server¶

Go to the MLFlow server component
Go to DevOps > Configs & Secrets
Create a new configuration with the details you found in the MLflow database setup:

    DATABASE = "your-database-name"
    USER = "your-database-username"
    PASSWORD = "your-database-password"
    HOST = "your-database-host"
    PORT = "your-database-port"

3. Add storage mounts¶

Go to DevOps > Storage
Create a new storage mount with in-memory type
Add mount path as mlruns

4. Configure container entrypoint¶

Go to DevOps > Containers
Set the command to ["mlflow", "run"]
Set the arguments to ["--backend-store-uri", "postgresql+psycopg2://$(USER):$(PASSWORD)@$(HOST):$(PORT)/$(DATABASE)", "--host", "0.0.0.0", "--port", "5000"]

5. Start the MLFlow server¶

Go to Build and build the latest
Go to 'Deploy' and deploy the latest build
Go to the web app url in the Deploy tab and ensure the server is running.

🚀 Great. Now you can move into fine tuning.

Next Steps¶

With your environment set up, you're ready to: