llama.cpp

6cfb31f9 - metal : add indirect mat-vec kernels for all quantization types