Models

Data scientists requires to repeat training models with various combinations of dataset, feature, parameters etc., and conducting experiments on models, furthermore, to register/to version models which have decent performance according to results. Nowadays, this is one part of MLOps.

Regarding managing versioned models, PrimeHub, by integrating well-known MLflow, provides models management feature, Models that scientists can examine the performance of versioned/registered models and deploy a selected model directly as a service by Deployments on PrimeHub.

MLflow is required

A running installed MLflow instance is required and Group Setting has to be configured with relative information.

MLflow setting is not configured yet

Mlflow instance is not reachable/running

Models

The page displays registered models from binding MLflow.

• See Tutorial: Use MLflow Tracking.

• MLflow UI button: navigate to binding MLflow server in a new tab.

As long as an experimental model is registered on MLflow, it is listed in Models on PrimeHub as well.

Versioned Model List

By clicking each model name, it navigates into the list of versioned models.

• Version: Version number

• Registered At: The registration date/time

• Deployed By: The deployment name if the model is used for a deployment; click to navigate into the deployment detail page.

• Parameters: selected parameters of the model

• Metrics: selected metrics of the model

• Deploy button: Deploying the selected versioned model.

Parameters and Metrics

Clicking Columns and select parameters and metrics to display as columns in the table.

Versioned Model Detail

The page displays the information regarding this version.

• Registered At

• Last Modified

• Source Run: linking to the run on MLflow

• Parameters: if any

• Metrics: if any

• Tags: if any

Deploy Versioned Model

In order to deploy a certain versioned model, click Deploy of a versioned model and select + Create new deployment or update an existing deployment. It will navigate to Deployment page, continue to submit the deployment with mandatory information.

Deployed

The model which is used for the deployment is with the information of the deployment name under Deployed by column. Click the deployment will navigate into the deployment detail page.

From the deployment information page, Model URI presents models:/<model_name>/<model_version>, e.g., models:/tensorflow-model/2.

• models:/: the model which is tracked by MLflow is deployed from Model Management

• <model_name>:the name of the model

• <model_version: the version number of the model

See Tutorial: Manage and Deploy a Model.