Long time llama.cpp user (and user of other ggml based AI programs) running on my crappy AMD card that only supports vulkan and it's usable thanks to GGML and llama.cpp, great program great project.