I am trying to run the parameter server architecture using MXNet and Nuclio on kubernetes. Using one worker and one parameter server the training works fine with the exception o