llama.cpp

086cfea1 - ggml : nibbles_from_floats() + bytes_from_nibbles() (ARM NEON)