Vision Transformer With Super Token Sampling Bias