Yes, the size is 99% 2 models weights required to run inference offline, there no way around it.