Fix transpose in attention scores equation

[taziksh]

9.3.2: Remove outer transpose in attention scores equation

The outer transpose in the attention scores equation appears inconsistent with the stated shape $a_i \in \mathbb{R}^{1 \times n}$.

The current expression is:

$$ a_i = \text{softmax}\left( \frac{[q_i^T k_1, q_i^T k_2, \dots, q_i^T k_n]} {\sqrt{d_k}} \right)^T \in \mathbb{R}^{1 \times n} $$

However, the vector

$$ [q_i^T k_1, q_i^T k_2, \dots, q_i^T k_n] $$

is treated elsewhere as a row vector of shape $1 \times n$. Applying softmax preserves that shape, so the final transpose would make it $n \times 1$, contradicting the stated shape $1 \times n$.

[shensquared]

awesome. thanks!