Thanks for this blog. Do we have certain hardware accelerators for enhancing the performance of caffe models or are most of them just for models built on keras or tensorflow?