also no updates since then. does any library follow this reference implementation? llama.cpp surely is more optimized?