llama.cpp

5a9edda7 - gguf-py : Numpy dequantization for most types