Update 2.2-图解transformer.md

docs/篇章2-Transformer相关原理/2.2-图解transformer.md

@@ -156,7 +156,7 @@ The animal didn't cross the street because it was too tired
 先通过一个简单的例子来理解一下：什么是“self-attention自注意力机制”？假设一句话包含两个单词：Thinking Machines。自注意力的一种理解是：Thinking-Thinking，Thinking-Machines，Machines-Thinking，Machines-Machines，共$2^2$种两两attention。那么具体如何计算呢？假设Thinking、Machines这两个单词经过词向量算法得到向量是$X_1, X_2$​：
 $$
 1: q_1 = X_1 W^Q, q_2 = X_2 W^Q; k_1 = X_1 W^K, k_2 = X_2 W^K;v_1 = X_1 W^V, v_2 = X_2 W^V, W^Q, W^K, W^K \in \mathbb{R}^{d_x \times d_k}\\
-2-3: score_{11} = \frac{q_1 \cdot q_1}{\sqrt{d_k}} , score_{12} = \frac{q_1 \cdot q_2}{\sqrt{d_k}} ; score_{21} = \frac{q_2 \cdot q_1}{\sqrt{d_k}}, score_{22} = \frac{q_2 \cdot q_2}{\sqrt{d_k}}; \\
+2-3: score_{11} = \frac{q_1 \cdot k_1}{\sqrt{d_k}} , score_{12} = \frac{q_1 \cdot k_2}{\sqrt{d_k}} ; score_{21} = \frac{q_2 \cdot k_1}{\sqrt{d_k}}, score_{22} = \frac{q_2 \cdot k_2}{\sqrt{d_k}}; \\
 4: score_{11} = \frac{e^{score_{11}}}{e^{score_{11}} + e^{score_{12}}},score_{12} = \frac{e^{score_{12}}}{e^{score_{11}} + e^{score_{12}}}; score_{21} = \frac{e^{score_{21}}}{e^{score_{21}} + e^{score_{22}}},score_{22} = \frac{e^{score_{22}}}{e^{score_{21}} + e^{score_{22}}} \\
 5-6: z_1 = v_1 \times score_{11} + v_2 \times score_{12}; z_2 = v_1 \times score_{21} + v_2 \times score_{22}
 $$