5

Is there a way to manage permissions at an experiment level in MLflow? We would like to have a shared server but would like to be able to manage permissions at an experiment level - e.g. admin can view all experiments, user_group1 can manage experiment1 - perhaps different groups can see results vs post results.

It looks like it is possible in databricks: https://docs.databricks.com/administration-guide/access-control/workspace-acl.html#experiment-permissions but I can't find anything in the opensource APIdocs.

Thanks.

prk
  • 319
  • 1
  • 3
  • 10

1 Answers1

2

I'm not sure the concept of permissions makes sense unless mlflow is part of a broader application. You would need the concept of users, then users trying to access an experiment. Open source mlflow doesnt support that.

You get that functionality in Databricks because mlflow is hosted as one feature on the broader platform. Once you are logging in to Databricks it is possible to have permissions at various levels. No one is logging into mlflow in open source, so you'd have to implement your own solution here.

Raphael K
  • 2,265
  • 1
  • 16
  • 23