Mac M1, llama.cpp and pi harness.

Mac M3, oMLX, pi. 64GB RAM. Qwen 3.8 27b / Gemma 12b

m2 24gb macbook mlx-lm & mlx-vlm own harness