Multi-head Attention

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