A Transformer Based Representation Learning