[D] Attention Mystery: Which Is Which - q, k, or v?

I'm finally wrapping my head around the attention mechanism, but one piece still eludes me: the matrix magic behind q, k, and v.

I get the whole matrix multiplication dance at a theoretical level, but what **mathematical property** actually dictates which matrix gets to be the **query (q)**, the **key (k)**, and the **value (v)**? Is it just some random assignment, or is there deeper logic at play?


Here's what I've gathered so far:

* All three matrices come from …

attention dance head machinelearning magic matrix matrix multiplication property query the matrix