llama.cpp

33dcb44a - convert : handle naive-quantized models