No option to change gpus when using torch.distributed.init_process_gr

Question

When we run the NVIDIA code for training on Imagenet using mixed precision and the following:

$ python -m torch.distributed.launch --nproc_per_node=n main_amp