This folder contains annotated reading material for the lab page.
Read how a causal mask turns parallel training into left-to-right prediction semantics.
Running code is optional. The expected outcome is that you can explain the mechanism, the relevant state/shape, the common misunderstanding, and the interview answer.