Multi-head Attention

A key layer in the Transformer architecture which represents a stack of Scaled-Dot Product Attention attention modules.