Snap interview question

How to use multiple GPU training and describe optimisation techniques for DL inference?