mirror of
https://github.com/skindhu/Build-A-Large-Language-Model-CN.git
synced 2026-07-25 05:10:24 +08:00
add fifth chapter
This commit is contained in:
@@ -267,6 +267,7 @@ tensor([ -9.5042, -10.3796, -11.3677, -11.4798, -9.7764, -12.2561])
|
||||
> $$ ln(0.2)+ln(0.1)+ln(0.05)≈−1.6094+(−2.3026)+(−2.9957)=−6.9077 $$
|
||||
>
|
||||
> 虽然这个和也是负数,但它不会像直接相乘的结果那样接近于0,避免了数值下溢的问题。**对数的累加性质**允许我们将原本的累乘操作转换为累加,使得计算更加稳定和高效。
|
||||
>
|
||||
>
|
||||
> 2. **归属概率在损失函数中的作用**
|
||||
>
|
||||
|
||||
Reference in New Issue
Block a user