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.

Bottleneck Transformers for Visual Recognition

Pretrained Pytorch Bottleneck Transformers for Visual Recognition including following

botnet50
botnet101
botnet152

Pretrained Vision Transformer Models

Pretrained Pytorch Vision Transformer Models including following

vit_s16_224
vit_b16_224
vit_b16_384
vit_b32_384
vit_l16_224
vit_l16_384
vit_l32_384