Comment by ashater

Thank you. Agreed, we are exploring different ways to apply these interpretability methods to a wide range of transformer based methods, not just decoder based generative applications.