llama.cpp

253eab8a - ggml : poc for normalizing weights for better quantization (metal)