Pure Go hardware accelerated local inference on VLMs using llama.cpp | Not Hacker News!