[Mashimo]

Not for faces, but the CLIP model for the context search https://docs.immich.app/features/searching/

That needs to be in (v)ram for searches.