Update 12-3.md

Self-Attention Paragraph Typos: Issues Atcold#779

docs/en/week12/12-3.md

@@ -36,7 +36,7 @@ $$
 \boldsymbol{h} = \boldsymbol{X} \boldsymbol{a}
 $$
 
-where $\boldsymbol{a} \in \mathbb{R}^n$ is a column vector with components $\alpha_i$.
+where $\boldsymbol{a} \in \mathbb{R}^t$ is a column vector with components $\alpha_i$.
 
 Note that this differs from the hidden representation we have seen so far, where the inputs are multiplied by a matrix of weights.