in the agent’s clearml.conf file, set agent.docker_force_pull to true .
You can also try in the machine running the ClearML agent to run:
docker pull nvidia/cuda:10.1-runtime-ubuntu18.04