MLFlow Model Repository Server on Azure Container Registry

This repository provides necessary artefacts to quicky and easily deploy an MLflow Tracking server on Azure.

MLflow is an open source platform for managing the end-to-end machine learning lifecycle.

Deploy

Requirements:

• az cli 2.0+

Deployment

1. Ensure you are logged-in in the azure cli.
   • Run az login to login.
   • Run az account set -s <SUBSCRIPTION_ID> to set target azure subscription.
2. Ensure you are in the deploy-aci folder.
3. Open deploy-aci.sh and inspect/change top parameters, if necessary.
4. Run ./deploy-aci.sh
5. Validate deployment by navigating to the ACI IP:port (default: 5000). NOTE, that it takes a few moments for the server to startup.
   • You can retrieve IP and port of the deployed Tracking Server on ACI by running:
   • az container show --name <ACI_NAME> --resource-group <ACI_RESOURCE_GROUP> --output table

Logging Data to MLFlow Tracking Server

Documentation for MLFlow Model Repository

Documentaion for mlflow repository and scripts

Current functionality

• Mlflow server deployed on a container instance with:
  • storage container - with blob and file store
  • docker image for mlflow server
• Version 1.8.0 of mlflow
• Models saved to blob storage and meta data saved to SQLlite db inside the container

Build Docker Image

Build image

docker build --tag jaredmagrath/mlflowserver-azure:1.8.2 .\mlflow-tracking-docker\

Push to dockerhub

docker push jaredmagrath/mlflowserver-azure:1.8.2