To be fair, llama.cpp had this feature for over a year now. It just applies to GGUF.