Click to multiply each word's value vector by its attention weight, one word at a time. Blending these weighted values is how attention builds its output.
Each value vector is scaled by its attention weight. Note it is the values being blended, not the keys.
