Skip to content

Commit

Permalink
Update llama系列模型.md
Browse files Browse the repository at this point in the history
  • Loading branch information
lerogo authored Apr 22, 2024
1 parent 6521d7e commit c992c7d
Showing 1 changed file with 1 addition and 1 deletion.
Original file line number Diff line number Diff line change
Expand Up @@ -67,7 +67,7 @@ $$

![](image/image_6g6JVd5GoX.png)

LLaMA中直接将FFN中的ReLU替换为SwiGLU,并将维度放缩为$(2/3) ⋅ 4d$
LLaMA中直接将FFN中的ReLU替换为SwiGLU,并将维度放缩为$(2/3) ⋅ 4d$。这样设计的原因是:维度放缩为 $(2/3) ⋅ 4d$ 后,其计算复杂度为 $(112/9) ⋅ d^3 + (8/3) ⋅ d$ ,普通的 $4d$ 纬度的计算复杂度为 $20 ⋅ d^3$ 。

## 1.4 旋转位置嵌入(RoPE)

Expand Down

0 comments on commit c992c7d

Please sign in to comment.