Is there a way to use a GPU only when training an Index in the process of add_faisis_index?

Motivation - Some FAISS indexes like IVF consist of the training step that clusters the dataset into a given number of indexes. It would be nice if we can use a GPU to do the training step and covert the index back to CPU as mention in [this faiss example](https://gist.github.com/mdouze/46d6bbbaabca0b9778fca37ed2bcccf6).