TraCR-Supported Mechanistic Interpretability

✍️ Compiling RASP code to Transformers and comparing them to trained networks