When a model is stored in MLflow format, with an MLmodel file and conda requirements, you can deploy it to batch or online endpoints with no scoring script or environment of your own.
Also called mlflow_model.
Read more: Microsoft Learn
In the Ultra Transcenders books
Each book explains MLflow model in context, with comparison tables and the common traps.
Terms in this definition
- MLflow
An open-source framework for experiment tracking and model management. Azure Machine Learning adopts it as its tracking API, so each workspace acts as a compatible MLflow tracking server.
- FORMAT
A DAX function that turns a value into text according to a format string, for instance "MMMM" to show a month's name. Since the output is text, numeric operations can't use it, and dynamic format strings were introduced to get around that.
- Online management group
Workloads that may talk directly to or from the internet, or that need no virtual network at all, go in this management group. It is a sibling of Corp and Local beneath Landing zones.
- Scoring script
Script for a deployment, usually
score.py:init()loads the model as the container starts andrun()handles scoring for each request. MLflow models do not require one. - Azure Machine Learning environment
Versioned asset that pairs a Docker image with a pip or conda specification, so every job or deployment using it gets identical dependencies. You point to it by name plus a version number, or by name with @latest.
Related terms
- No-code deployment
Azure Machine Learning itself supplies the scoring script and environment, so an MLflow model can go to a batch or online endpoint with neither written by you.
- Registered model
A versioned model asset in Azure Machine Learning, typed as custom_model, mlflow_model or triton_model, with tags and lineage to the producing job. Registering under the same name again increments the version automatically.