Reason for not using transformer as EBM architecture #6
Description
Do you know why EBM architectures are always parametrized as convolutional models and never use self-attention?
{{ message }}
Do you know why EBM architectures are always parametrized as convolutional models and never use self-attention?