Update content src/site/notes/AI/transformers/transformers architecture.md

Author: 1024daniel

src/site/notes/AI/transformers/transformers architecture.md

@@ -13,9 +13,27 @@ https://zhuanlan.zhihu.com/p/82312421
 
 ![Pasted image 20241124140104.png](/img/user/AI/transformers/attachments/Pasted%20image%2020241124140104.png)
 
+https://www.youtube.com/watch?v=wjZofJX0v4M
 
-![Transformers(how LLMs work)](https://www.youtube.com/watch?v=wjZofJX0v4M)
 
 
+<div style="position: relative; width: 100%; max-width: 800px; height: 0; padding-bottom: 56.25%; margin: 0 auto;">
+    <iframe 
+        src="https://www.youtube.com/embed/wjZofJX0v4M?rel=0&modestbranding=1&autoplay=0&showinfo=0&fs=1&disablekb=1" 
+        style="position: absolute; top: 0; left: 0; width: 100%; height: 100%; border: none;" 
+        allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture" 
+        allowfullscreen>
+    </iframe>
+</div>
 
-![Pasted image 20250119001634.png](/img/user/AI/transformers/attachments/Pasted%20image%2020250119001634.png)
+
+
+
+![Pasted image 20250119001634.png](/img/user/AI/transformers/attachments/Pasted%20image%2020250119001634.png)
+
+
+### 1. embedding层
+[[AI/torch/nn/embedding\|embedding]]
+主要将词表表达为vector并映射到高维词表空间，对于英文来说有大概5k+左右的向量，除了表示token的含义之外，还通过位置编码将token在context的位置也包含了进去，将位置编码进去的主要目的是为了可以并行计算不会像RNN网络一样存在前后依赖，可以最大化并行利用好gpu的效率
+### 2. attention 层
+### 3. feed forward layer