MLflow Authorization Concepts¶
Authorization is implemented in MLflow at the level of Experiments and Models. This means that all experiment runs inherit permissions settings from the experiment they belong to. Also, it means that all Model Versions inherit permissions from the Model.
MLflow supports three roles:
When an experiment is created by a user, that user is automatically assigned manager permissions.
When it pertains to an experiment, Readers can view experiments and experiment runs.
When it pertains to a Model, Readers can view model details, model versions and they can deploy model versions.
When it pertains to an experiment, Editors can view experiments and experiment runs. They can also edit details of the run, for example, by adding a tag.
When it pertains to a Model, Editors can view model details and model versions. Further, they can edit details of Models and Model versions such as adding tags, etc. Finally, they can transition Model Versions from one stage to another.
When it pertains to an experiment, Managers can delete model versions and also delete the experiment and all of its runs.
When it pertains to a Model, Managers can also delete Model Versions and Models.