Implementation of the procedure in Language Models Can Control Their Own Attention, from Namgyu Ho et al. of KAIST AI - View it on GitHub