Never update mlflow version within docker-mlflow-server/requirements.txt without first backing up the database (usually the docker volume containing the database). Usually it will end up to losing your data.
Just run docker-compose up -d in the VM that the MLflow server will be deployed.
A Postgres database (parameter logging) and a minIO server (model store) are also automatically deployed.
See here for how to deploy docker on EPU VMs and opening ports to the public:
Alternatively check file at root: How_to_Whitelist_IP_Addresses_to_Access_Desired_Docker_Containers.pdf
Follow the procedure and only allow VPN IPs to the specific port you need to reach.