Modified Open-Llama-13B to 16k out of box (without finetuning) following the ntk-scaling recipe from this reddit thread.