In transformer-based neural networks, a Key Vector is one of three learned linear projections of an input token embedding—alongside the Query Vector and Value Vector—that together implement the scaled dot-product attention mechanism. The key vector represents what a given token has to offer: each query–key dot product measures the compatibility or relevance between a querying token and every other token in the sequence, with the resulting attention weights determining how much each value vector contributes to the output representation.
Semantic Classification
Content
Key Vector — content pending enrichment.