Entity

Self-Attention Mechanism

A mechanism in Transformer models that allows the model to 'attend' to different tokens at different times by computing weighted connections between all tokens.