llama.cpp

Single allocation of encode_async block with non-ARC capture in ggml-metal.m

#9747

Merged