Support for mlflow 3.x.x in azure-ml?

Hi Daniel Gröger,

Azure Machine Learning is MLflow‑compatible for tracking, logging, model registration, and deployment. You can use current MLflow packages (e.g., mlflow / mlflow‑skinny) with Azure ML today for experiment tracking and to deploy MLflow models to online or batch endpoints.

MLflow 3 introduced new capabilities (such as Logged Models, expanded registry views, and GenAI‑centric tracing/evaluation). These are documented in the MLflow 3 materials and are fully available in Databricks‑managed MLflow; the Azure ML documentation currently emphasizes the established MLflow integration for tracking and deployment rather than these newer 3.x‑specific features.

How to proceed now:

1. Use Azure ML as the MLflow tracking server by pointing your code to the workspace; continue to log runs, metrics, parameters, and artifacts with MLflow.
2. Register and deploy MLflow models to Azure ML endpoints. Azure ML supports no‑code deployment for MLflow models and autogenerates the scoring environment from the model’s conda.yaml.
3. If you specifically need MLflow 3 features like Logged Models and end‑to‑end 3.x registry views, consider running those workflows in Databricks‑managed MLflow and where appropriate promoting artifacts to Azure ML for serving.

Links for setup and details:

• MLflow + Azure ML overview and compatibility: MLflow and Azure Machine Learning
• Deploy MLflow models on Azure ML (no‑code & custom options): Guidelines for deploying MLflow models
• MLflow 3 summary & docs: MLflow 3 overview • MLflow releases
• https://github.com/mlflow/mlflow/issues/16215

If you’re hitting a specific issue (for example, model logging compatibility or a package conflict during deployment), please share:

• your mlflow package version,
• the conda.yaml/environment spec, and
• the exact error text or job log snippet.

That will help us provide a targeted to fix.

thanks for the answer. We currently do not want to use another third party tool like databricks. So this option is out.

We actually facing the issue during logging and receive the error message that the loggeg_models endpoint does not exist.

We are using mlflow 3.5.1 . We do not want to downgrade.