Fixed minor issues with existing vision transformer

Fixed issue with pretrained model of vision transformer.

• Attention layer convolution bias was set to false by default causing failure in loading model weights.