Interpreting vision transformers via residual replacement model | Read Paper on Bytez