Skip to main content
Version: Cloud

Configure MLFlow

With the MLFlow integration you can deploy models and explainers to Deeploy that are available in your own MLFlow model registry. You can make use of MLFlow's model stages or deploy specific model versions.

Configure MLFlow​

The MLFlow integration is controlled on team level. Admins can set up the integration on the Integrations page, which is part of the Admin panel. Click Configure on the MLFlow card and Add credentials to set up the integration.

Pre-requisites​

In order for Deeploy to communicate with your MLFlow tracking server there are a few requirements:

  • Deeploy needs to be able to reach the server, for Deeploy private cloud this means available in the same VPC and for Deeploy cloud it needs to be publicly accessible
  • The MLFlow models need to be stored in one of the compatible object storages: AWS S3, Google Cloud Storage, Azure blob storage or Minio
  • Authentication needs to be enabled, which can be enforced by starting the tracking server with the --app-name basic-auth argument attached.
  • MLFlow version >=2.5.0

Best practices​

Give Deeploy the least access possible when connecting to your MLFlow server. In practice, this means having or creating a read-only user.

Update MLFlow credentials​

Consult the update integration credentials to update your MLFlow credentials. Additionally, updating your MLFlow credentials always requires you to fill in your password.