Nvidia Runtime in Docker

We are using Docker 19 & Nvidia Docker v2 on CentOS 7.5. The platform has CUDA 9.5.

The following command on CentOS to setup Nvidia as default runtime is consistently failing.
usr/bin/dockerd -H fd:// --default-runtime=nvidia --node-generic-resource gpu=${GPU_ID}

We are using docker to run machine learning applications - the command above works just fine on Ubuntu but fails on CentOS
– Is there a workaround for this on CentOS, or
– a specific set of steps that we need to follow to configure the environment