holy crap, thank you. Its actually running at 54 tok/s which is incredible for local inference.