Attention

Attention is how a model weighs which available tokens are useful when processing a token.