How to run train script on cuda:1 instead of cuda:0

[SarthakJaingit]

Hi,

Recently, I was looking to train an effecient det model; however, in a shared server only cuda:1 is open. I was wondering what option is there that I can choose to train on cuda:1 instead of cuda:0 and not get a CUDA out of memory error.