Hi!
I have deployed a CM server + agent on a BM Kubernetes cluster and am trying to set the default docker image for running trains. I have specified the default docker image in clearml.conf in the agent and added an environment variable to the API server. However, the container is still being created with the default image of ubuntu:18.04. Am I misunderstanding something about how this works?

My goal is to run containers with a specific version of Python and CUDA. Another question is how to make all GPUs visible from inside the training container (--gpus docker string args?).