MLX is often better on Apple Silicon from my experience llama.cpp is my fallback method

→ View original post on X — @theahmadosman