llama.cpp

5cc672a9 - metal : try to utilize more of the shared memory using smaller views