large language models (LLMs) have demonstrated impressive capabilities in
natural language processing. However, their internal mechanisms are still
unclear and this lack of transparency poses unwanted risks for downstream
applications. Therefore, understanding and explaining these mode