Maybe a container issue. I know Nvidia recommends their docker containers, but I am usually just running it on the Spark directly.

→ View original post on X — @rasbt