He said literature and the arts are critical to what he called normalization after decades of military rule.

他表示文学和艺术过几十年的军事统治后他所谓的一化至关重要。

And so what's the normalization constant that I need to normalize it? 1.

那么，我需要的一化常数是什么呢？1.

Then we find the squared distance from that expected value, and we, I guess you could call it, normalize it by the expected value.

然后我们计算与该期望值的平方距离， 并将其，我猜可以称之为， 通过期望值进行一化处理。

And so sometimes we write them real neatly with the normalizing constants and place the mean and the variance in places that you can recognize very easily.

因此，有时我们会用一化常数将它们整齐地写出来，并将平均值和方差放您很容易识别的地方。

So e to the minus 1 half x bar minus mu squared over sigma squared over n. And I've gotten rid of my normalizing constants right here.

所以 e 到负 1 的一半 x bar 减去 mu 的平方超过 sigma 的平方超过 n。我这里删除了我的一化常量。

So I'll say pi p would be proportional to, so there's a normalizing constant in here, p to the alpha minus 1, 1 minus p, beta minus 1.

所以我会说 pi p 与 成正比，所以这里有一个一化常数，p 到 alpha 减 1,1 减去 p， beta 减去 1。

And a lot of your priors on non-compact spaces, unbounded things like mu, turn out that you want to give a prior that can't be normalized quite often.

而紧空间上的很多先验，比如 mu 这样的无界事物， 结果证明想给出一个不能常一化的先验。

You want to have some basic programming skills, but then you're going to dive into key concepts like reward modeling, supervised fine-tuning, mixture of experts layers, RMS, norm, rope, KV caching and more.

需掌握一些基础编程技能，随后深入研究诸如奖励建模、监督微调、专家层混合、均方根一化、旋转位置编码、KV缓存等关键概念。

It's probably worth mentioning there's another set of parameters associated with those normalization steps that this explanation has skipped over, but like the bias vector, they account for a very trivial proportion of the total.

可能值得一提的是， 还有另一组与这些一化步骤相关的参数被这个解释忽略了，但就像偏置向量一样， 它们整个参数总量中所占的比例常微小。

So this conjugacy thing is great if you know about the distribution because it won't require you to keep track of any of your normalizing constants or solve that integral in the denominator at the end of the day.

所以如果了解分布，这个共轭就很好了，因为它不需要跟踪任何一化常数，也不需要一天结束时解决分母中的积分。