hulma
New Contributor II

my workaround is to add protobuf<=3.20.1 in conda_env when i train the model in mlflow.