pytorch

9b58d31d - add a global flag for cuda checker so it doesn't get extra time during non-GPU docker images