Exploring the 'grokking' phenomenon for modular addition using a one-layer transformer with bidirectional attention.