Coding a Vision Transformer (almost) from…

[govind sharma]

I am not getting the higher accuracy, even though the implementation is same.

batch_first=True is set in multi_head attention