Are there any benchmarks on how much more performant Typesense is with embeddings on a GPU vs CPU? And any recommendations on AWS instances to use if self hosting? Thanks!