Looks interesting, it reminds me of the first llama.cpp, but better documented.