Add detection of GPU numbers and therefore add multi-gpu usage (--machine.num-devices 1) in train.sh and eval.sh scripts, as shown in https://docs.nerf.studio/quickstart/first_nerf.html#evaluating-runs