1
00:00:00,440 --> 00:00:01,480
Well, hello.
嗯，你好。

2
00:00:01,520 --> 00:00:02,240
I'm back.
我回来了。

3
00:00:02,240 --> 00:00:03,960
And you're back too, which is a great thing.
你也回来了，这是一件很棒的事情。

4
00:00:04,000 --> 00:00:07,920
It means it's time for us to start day three of our journey together.
这意味着我们一起开始第三天的旅程。

5
00:00:08,200 --> 00:00:13,040
And day three, we've already done building an agentic workflow for the first time.
第三天，我们已经第一次完成了代理工作流程的构建。

6
00:00:13,200 --> 00:00:14,480
And we've talked about patterns.
我们已经讨论过模式。

7
00:00:14,480 --> 00:00:19,520
And now we're going to talk about orchestrating between LMS and many of them.
现在我们要讨论 LMS 和其中许多之间的编排。

8
00:00:20,000 --> 00:00:22,960
So this is going to be a practical day.
所以这将是务实的一天。

9
00:00:22,960 --> 00:00:24,080
You'll be pleased to hear.
你会很高兴听到的。

10
00:00:24,080 --> 00:00:25,920
It's about time we got some more coding.
是时候我们进行更多编码了。

11
00:00:25,920 --> 00:00:28,320
And we're going to be calling a lot of LMS.
我们将调用很多 LMS。

12
00:00:28,600 --> 00:00:31,560
And I just want to of course, say a few things up front.
当然，我只想提前说几句话。

13
00:00:31,800 --> 00:00:37,280
We're going to be calling both paid APIs and also open source models, and we're going to be calling
我们将调用付费 API 和开源模型，并且我们将调用

14
00:00:37,280 --> 00:00:42,360
them both in the cloud and also open source models locally and doing it throughout this course.
它们既在云端，也在本地开源模型，并在整个课程中进行。

15
00:00:42,560 --> 00:00:48,400
And I want to be clear that you have complete flexibility to decide which models you pick, at which
我想明确的是，您可以完全灵活地决定选择哪种型号、选择哪种型号

16
00:00:48,400 --> 00:00:53,600
point I'm going to have coded it one way, but a great exercise is to take what I've done and apply
我将以一种方式对其进行编码，但一个很好的练习是采用我所做的并应用

17
00:00:53,600 --> 00:00:54,880
it to other models.
到其他型号。

18
00:00:54,880 --> 00:00:57,640
And if you don't want to spend a dime, you don't need to.
如果你不想花一毛钱，那就不需要。

19
00:00:57,680 --> 00:01:00,280
You can do this all using local models.
您可以使用本地模型来完成这一切。

20
00:01:00,280 --> 00:01:05,050
Although performance might vary, you can see the results I get from the models I use, and try and
尽管性能可能会有所不同，但您可以看到我从我使用的模型中获得的结果，并尝试

21
00:01:05,050 --> 00:01:08,770
see what you can achieve with free open source models as well.
看看您可以使用免费的开源模型实现什么。

22
00:01:09,250 --> 00:01:15,290
And if if you want more information on what it's like to select models, whether they're open or closed
如果您想了解有关选择模型的更多信息，无论它们是开放式还是封闭式

23
00:01:15,290 --> 00:01:19,290
source, apply them and deploy them, then you should take a look at my other course.
源码，应用它们并部署它们，那么你应该看看我的其他课程。

24
00:01:19,330 --> 00:01:23,610
We're not going to be covering that in detail here, because otherwise I feel like it's going to be
我们不会在这里详细讨论这一点，因为否则我觉得它会是

25
00:01:23,610 --> 00:01:26,410
too much of a of a rabbit hole and we'll get distracted.
太多的兔子洞我们会分心。

26
00:01:26,410 --> 00:01:31,690
So I'm assuming you're coming into this knowing about the different kinds of models and having a sense
所以我假设您了解不同类型的模型并有一定的感觉

27
00:01:31,690 --> 00:01:34,650
of what makes sense for closed source, open source, and so on.
什么对闭源、开源等有意义。

28
00:01:34,890 --> 00:01:38,970
If you don't have that, you can either just go along with it, just just sort of pick it up as we go.
如果你没有，你也可以跟着它走，只是在我们走的时候把它捡起来。

29
00:01:39,210 --> 00:01:43,330
Or of course, you could always turn to look at my other course, or I'll try and put some more background
或者当然，您可以随时查看我的其他课程，或者我会尝试提供更多背景知识

30
00:01:43,330 --> 00:01:46,170
information in the guides as well, because that won't be required.
指南中的信息也是如此，因为这不是必需的。

31
00:01:46,170 --> 00:01:50,650
You can just just go, go with the flow in terms of how we pick models.
你可以随波逐流地选择模型。

32
00:01:51,330 --> 00:01:56,490
So let's talk about the cast of characters, the different models that we're going to be experiencing
那么让我们来谈谈角色阵容以及我们将要体验的不同模型

33
00:01:56,490 --> 00:01:56,890
now.
现在。

34
00:01:57,290 --> 00:02:00,050
So the first model needs no introduction really.
所以第一个模型确实不需要介绍。

35
00:02:00,090 --> 00:02:05,090
It is of course, the model GPT four mini Many from open AI.
当然是来自 open AI 的 GPT 4 mini 模型。

36
00:02:05,290 --> 00:02:07,530
We've already used it in a couple of calls.
我们已经在几次通话中使用过它。

37
00:02:07,530 --> 00:02:11,090
It's for sure the most well known of the models out there.
它肯定是目前最知名的模型。

38
00:02:11,090 --> 00:02:15,370
And of course, there's also GPT four, the bigger cousin of GPT four.
当然，还有 GPT 4，它是 GPT 4 的近亲。

39
00:02:16,050 --> 00:02:21,330
And then there are the reasoning models, which are models that have been trained to think through their
然后是推理模型，这些模型经过训练可以通过其自身的能力进行思考

40
00:02:21,330 --> 00:02:26,530
steps in an Agentic like way, in like a workflow of thinking through the different steps before they
以类似于 Agentic 的方式进行步骤，就像在执行之前思考不同步骤的工作流程

41
00:02:26,530 --> 00:02:27,650
arrive at their conclusion.
得出他们的结论。

42
00:02:27,650 --> 00:02:33,330
Because it turns out that when you ask an LLM to think through its steps, you get much better outcomes.
因为事实证明，当你要求法学硕士仔细思考其步骤时，你会得到更好的结果。

43
00:02:33,810 --> 00:02:39,330
So we may take a look at some point at oh one and oh three mini, but it's less essential for this course.
因此，我们可能会看一下哦一和哦三迷你的某个点，但这对于本课程来说不是那么重要。

44
00:02:39,330 --> 00:02:40,010
Most of the time.
大多数时候。

45
00:02:40,010 --> 00:02:45,810
We're going to be sticking with GPT four mini now and OpenAI's great rival.
我们现在将继续使用 GPT 4 mini，这是 OpenAI 的强大竞争对手。

46
00:02:45,810 --> 00:02:48,370
Of course, their competitor is anthropic.
当然，他们的竞争对手是人类。

47
00:02:48,410 --> 00:02:53,170
That was actually started by a couple of people from OpenAI originally, and we'll be looking at some
这实际上最初是由 OpenAI 的几个人发起的，我们将关注一些

48
00:02:53,170 --> 00:02:53,730
of their models.
他们的模型。

49
00:02:53,730 --> 00:02:58,730
But Claude 37 sonnet is the model that I will spend most time on.
但克劳德 37 十四行诗是我花最多时间研究的模型。

50
00:02:58,850 --> 00:03:04,290
And if you want to have a cheaper version, you can go with the Claude three haiku, which is significantly
如果你想要一个更便宜的版本，你可以选择克劳德三俳句，这是显着的

51
00:03:04,290 --> 00:03:05,210
lower cost.
成本更低。

52
00:03:05,620 --> 00:03:09,100
But but Sonet is also fairly cheap for Google.
但 Sonet 对于 Google 来说也相当便宜。

53
00:03:09,100 --> 00:03:12,180
We're going to be using Gemini two zero flash.
我们将使用 Gemini 两个零闪光。

54
00:03:12,220 --> 00:03:15,300
There is also the Pro version of that too, but I think we'll stick with flash.
还有专业版，但我想我们会坚持使用闪光灯。

55
00:03:15,300 --> 00:03:20,980
And as of right now, flash is actually free at least as long as you use it within certain usage limits.
截至目前，闪存实际上是免费的，至少只要您在一定的使用限制内使用它。

56
00:03:21,020 --> 00:03:25,940
I don't know how long that will be the case for, but by all means, if you want to use an open frontier
我不知道这种情况会持续多久，但无论如何，如果你想使用开放的边界

57
00:03:25,980 --> 00:03:29,820
model without paying for it, then Gemini might be your path.
模型而不付钱，那么双子座可能是你的道路。

58
00:03:29,860 --> 00:03:33,180
Do look into that deep Sikh.
一定要研究一下那个深刻的锡克教徒。

59
00:03:33,180 --> 00:03:40,100
Deep Sikh, of course, is the Chinese upstart startup that shocked us all by coming up with such a
当然，Deep Sikh 是一家中国新贵初创公司，它提出了这样一个让我们所有人震惊的方案。

60
00:03:40,100 --> 00:03:43,580
powerful model in the form of Deep Sikh V3 and R1.
Deep Sikh V3 和 R1 形式的强大模型。

61
00:03:43,580 --> 00:03:48,660
And it's important to understand that what made Deep Sikh so sensational was not necessarily that their
重要的是要明白，让深锡克教徒如此轰动的并不一定是他们的

62
00:03:48,660 --> 00:03:51,380
model was the strongest in the world, because it wasn't.
模型是世界上最强的，因为事实并非如此。

63
00:03:51,420 --> 00:03:58,020
It was slightly behind the latest from open AI, but that they developed such powerful techniques to
它稍微落后于开放人工智能的最新成果，但他们开发了如此强大的技术

64
00:03:58,060 --> 00:04:06,260
train Deep Sikh to be that good, that it cost them a fraction of the spend that OpenAI had spent to
训练 Deep Sikh 变得如此优秀，他们只花费了 OpenAI 花费的一小部分

65
00:04:06,350 --> 00:04:09,870
train GPT four and train GPT 40101.
训练 GPT 4 并训练 GPT 40101。

66
00:04:10,150 --> 00:04:16,110
Deep seek was able to achieve very similar performance, pretty much comparable at a fraction.
深度搜索能够实现非常相似的性能，几乎可以相媲美。

67
00:04:16,110 --> 00:04:18,830
I think it's like 30 times less spend.
我认为这相当于减少了 30 倍的支出。

68
00:04:18,870 --> 00:04:20,590
That was the true innovation.
这才是真正的创新。

69
00:04:20,590 --> 00:04:22,590
That's the remarkable thing about deep seek.
这就是深度探索的非凡之处。

70
00:04:22,670 --> 00:04:25,350
And also that they open sourced the model so that you can use it.
而且他们还开源了该模型，以便您可以使用它。

71
00:04:25,350 --> 00:04:32,470
But the main the major model has 671 billion parameters, which means it's far too big for anyone to
但主要模型有 6710 亿个参数，这意味着它对于任何人来说都太大了

72
00:04:32,510 --> 00:04:33,950
run that on their computers.
在他们的计算机上运行它。

73
00:04:34,150 --> 00:04:40,070
But there are versions of it, small versions of it called the distilled versions, which are in fact
但它有一些版本，它的小版本称为蒸馏版本，实际上是

74
00:04:40,070 --> 00:04:42,230
themselves just smaller models.
本身只是较小的模型。

75
00:04:42,230 --> 00:04:48,950
They're versions of Llama and Quen, two different models that have been fine tuned on data generated
它们是 Llama 和 Quen 的版本，这两个不同的模型已根据生成的数据进行了微调

76
00:04:48,950 --> 00:04:56,710
by the big Deep Seek and those smaller distilled versions of Deep Seek for sure available free of charge.
由大型 Deep Seek 和那些较小的 Deep Seek 精炼版本提供，当然可以免费使用。

77
00:04:57,150 --> 00:04:57,910
Grok.
格洛克。

78
00:04:57,950 --> 00:05:02,510
We will also be using grok, and there are two groks.
我们还将使用 grok，有两个 grok。

79
00:05:02,910 --> 00:05:10,040
Confusingly, if you don't know this grok spelt with a k at the end is the name of the model that comes
令人困惑的是，如果您不知道最后拼写为 k 的 grok 是该模型的名称

80
00:05:10,040 --> 00:05:12,200
from from the company formerly known as Twitter.
来自以前称为 Twitter 的公司。

81
00:05:12,240 --> 00:05:16,600
Now x x is grok and we might use grok with a k at some point as well.
现在 x x 是 grok，我们也可能在某些时候将 grok 与 k 一起使用。

82
00:05:16,640 --> 00:05:20,120
Not not in today's lab, but grok with a Q is something different.
在今天的实验室里并非如此，但用 Q 来理解却是不同的。

83
00:05:20,160 --> 00:05:29,560
Grok with a Q is a company that has come up with a really cheap, fast way to run inference runtime
Grok with a Q 是一家公司，它提出了一种非常便宜、快速的方式来运行推理运行时

84
00:05:29,560 --> 00:05:37,040
models like llama 3.3, which is the massive version of llama with 70 billion parameters.
像 llama 3.3 这样的模型，它是 llama 的大规模版本，拥有 700 亿个参数。

85
00:05:37,040 --> 00:05:44,960
So you can run llama 3.3 really fast, really low cost on Groks infrastructure and along with many other
因此，您可以在 Groks 基础设施以及许多其他基础设施上非常快速、非常低成本地运行 llama 3.3

86
00:05:44,960 --> 00:05:48,280
open source models including Deep Seek variants.
开源模型，包括 Deep Seek 变体。

87
00:05:48,280 --> 00:05:51,000
So grok is great to use for that.
所以 grok 非常适合用于此目的。

88
00:05:52,280 --> 00:05:53,880
And then Olama.
然后是奥拉马。

89
00:05:53,880 --> 00:05:58,280
So Olama is itself a more of a platform.
所以 Olama 本身更像是一个平台。

90
00:05:58,280 --> 00:06:05,360
It's something that you can use to run something locally that provides endpoints locally that are consistent,
您可以使用它在本地运行某些内容，并在本地提供一致的端点，

91
00:06:05,360 --> 00:06:11,560
very similar to the endpoints that OpenAI and other models here have so that you can make local calls
与 OpenAI 和其他模型的端点非常相似，因此您可以进行本地调用

92
00:06:11,560 --> 00:06:18,080
to an API, which is in fact going to just run an open source model locally on your computer in high
到一个 API，它实际上只是在你的计算机上本地运行一个开源模型

93
00:06:18,080 --> 00:06:22,360
performance optimized cplusplus using a library called llama CPP.
使用名为 llama CPP 的库对 cplusplus 进行了性能优化。

94
00:06:22,880 --> 00:06:26,040
And so a llama is something that we will use as well.
所以我们也会使用美洲驼。

95
00:06:26,760 --> 00:06:30,400
Now, if some of these terms are unfamiliar to you, terms like inference.
现在，如果您不熟悉其中一些术语，例如推理等术语。

96
00:06:30,440 --> 00:06:34,600
If you're not sure about that, if you don't know, if you don't fully understand the difference between
如果你对此不确定，如果你不知道，如果你不完全理解两者之间的区别

97
00:06:34,600 --> 00:06:37,760
running something over a llama or grok, then I can.
在美洲驼或大神身上跑一些东西，然后我就可以了。

98
00:06:37,760 --> 00:06:43,200
I can suggest a background materials, look in the guides, and also consider whether you'd like to
我可以建议背景材料，查看指南，并考虑您是否愿意

99
00:06:43,200 --> 00:06:45,960
look at my LLM engineering course, which does cover all of this.
看看我的法学硕士工程课程，它确实涵盖了所有这些内容。

100
00:06:46,720 --> 00:06:53,640
And the final point that I will make is that you may know that I'm something of a fanatic on leaderboards,
我要说的最后一点是，你可能知道我是排行榜的狂热分子，

101
00:06:53,640 --> 00:06:58,960
on places you can go to read about metrics and performance of different models.
在您可以阅读有关不同模型的指标和性能的地方。

102
00:06:58,960 --> 00:07:05,320
So much so that I was called a leaderboard, uh, humorously by my my great friend John Crone on his
以至于我被我的好朋友 John Crone 幽默地称为排行榜。

103
00:07:05,320 --> 00:07:06,680
Super Data Science podcast.
超级数据科学播客。

104
00:07:07,000 --> 00:07:12,730
But there is a website called called leaderboard, which I've given the web address right there.
但是有一个名为排行榜的网站，我在那里给出了网址。

105
00:07:12,730 --> 00:07:18,570
And that is a great resource because it compares a number of the leading closed source and open source
这是一个很好的资源，因为它比较了许多领先的闭源和开源

106
00:07:18,570 --> 00:07:20,570
models together side by side.
模型并排在一起。

107
00:07:20,570 --> 00:07:26,730
And it has things like the costs and it has the context window size, if you're familiar with that,
它有诸如成本和上下文窗口大小之类的东西，如果你熟悉的话，

108
00:07:26,850 --> 00:07:31,210
and it has the cost, you can calculate based on the number of input and output tokens.
它是有成本的，你可以根据输入和输出代币的数量来计算。

109
00:07:31,450 --> 00:07:37,810
And it's also got the performance, the results of key benchmarks across a number of dimensions.
它还获得了性能以及多个维度的关键基准测试结果。

110
00:07:37,810 --> 00:07:41,410
So I strongly encourage people to go and check out the vellum leaderboard.
所以我强烈鼓励人们去查看牛皮纸排行榜。

111
00:07:41,410 --> 00:07:43,650
And there'll be a link in the resources.
资源中会有一个链接。

112
00:07:43,650 --> 00:07:46,050
And it's great to have that bookmarked up.
很高兴能把它加入书签。

113
00:07:46,090 --> 00:07:51,810
And as you go through different APIs, use that to gauge the the costs and capabilities associated with
当您使用不同的 API 时，请使用它来衡量与以下内容相关的成本和功能：

114
00:07:51,810 --> 00:07:52,130
them.
他们。

115
00:07:53,010 --> 00:07:56,890
And now I'm going to put up one more time something that I said to you in the first lecture.
现在我将再次提出我在第一堂课中对你们说过的话。

116
00:07:56,890 --> 00:08:03,130
But just to really double down on this, I want to remind you that there are great resources all over
但为了真正加倍努力，我想提醒您，世界各地都有丰富的资源

117
00:08:03,130 --> 00:08:04,090
the place for this course.
本课程的地点。

118
00:08:04,130 --> 00:08:07,330
There's a resource for the whole course with videos and links.
有包含视频和链接的整个课程的资源。

119
00:08:07,330 --> 00:08:10,290
There's GitHub, the repo that has guides.
GitHub 是一个包含指南的存储库。

120
00:08:10,290 --> 00:08:11,570
It has the troubleshooting.
它有故障排除。

121
00:08:11,570 --> 00:08:14,500
And I'm always updating the labs to keep them up to date.
我总是更新实验室以使其保持最新状态。

122
00:08:14,500 --> 00:08:18,140
I might add on new models, the models that we just went through a moment ago.
我可能会添加新模型，即我们刚才讨论过的模型。

123
00:08:18,300 --> 00:08:20,700
Those are the models that right now are the latest and greatest.
这些是目前最新、最好的型号。

124
00:08:20,700 --> 00:08:23,660
But as new versions come out, I will update the labs.
但随着新版本的出现，我将更新实验室。

125
00:08:23,660 --> 00:08:25,380
So you've got the latest in there.
所以你已经得到了最新的信息。

126
00:08:25,380 --> 00:08:29,980
And so, you know, what you've heard is, is is one story, but you'll get an even better story when
所以，你知道，你听到的是一个故事，但当你听到时，你会得到一个更好的故事

127
00:08:29,980 --> 00:08:31,900
you go through the labs yourselves.
你自己去实验室。

128
00:08:32,300 --> 00:08:37,180
And I will also urge you to keep a thick skin.
我也会劝你保持厚脸皮。

129
00:08:37,460 --> 00:08:38,260
There's that.
就是这样。

130
00:08:38,260 --> 00:08:39,420
You will hit roadblocks.
你会遇到障碍。

131
00:08:39,420 --> 00:08:40,460
There will be problems.
会有问题。

132
00:08:40,460 --> 00:08:45,740
That's one thing I can say for sure, but see them as this is where real learning happens.
这是我可以肯定地说的一件事，但请把它们视为真正学习发生的地方。

133
00:08:45,740 --> 00:08:46,700
It's in debugging.
正在调试中。

134
00:08:46,700 --> 00:08:51,620
It's in diagnosing and figuring things out, even if it's painful to start with.
它在于诊断和解决问题，即使一开始很痛苦。

135
00:08:51,660 --> 00:08:57,020
It's super satisfying when you fix it and get it on track, and that is the way to do it.
当你修复它并使其走上正轨时，这是非常令人满意的，这就是做到这一点的方法。

136
00:08:57,340 --> 00:09:04,980
And and if all else fails, or as I say, or if not all else fails, you just want to then reach out
如果所有其他方法都失败了，或者正如我所说，或者如果不是所有其他方法都失败了，那么您只想伸出援手

137
00:09:04,980 --> 00:09:06,020
and contact me.
并联系我。

138
00:09:06,060 --> 00:09:07,060
Of course, one more time.
当然，还有一次。

139
00:09:07,060 --> 00:09:10,180
I've got my LinkedIn right there, but also you can email me.
我的 LinkedIn 就在那里，您也可以给我发电子邮件。

140
00:09:10,220 --> 00:09:13,900
I've got my details all over the place and I like hearing from people.
我到处都有我的详细信息，而且我喜欢听取人们的意见。

141
00:09:13,900 --> 00:09:15,020
I'm responsive.
我有反应。

142
00:09:15,060 --> 00:09:19,940
But most importantly, I don't want you to be suffering in pain with problems.
但最重要的是，我不希望你因问题而遭受痛苦。

143
00:09:19,940 --> 00:09:21,220
I want to be fixing them.
我想修复它们。

144
00:09:21,220 --> 00:09:22,460
So keep this in mind.
所以请记住这一点。

145
00:09:22,460 --> 00:09:28,540
And also, I mean ask ChatGPT you do tend to get great, great answers.
而且，我的意思是询问 ChatGPT，您确实往往会得到非常非常好的答案。

146
00:09:28,540 --> 00:09:30,700
Honestly, it's amazing and I tend to.
老实说，这太棒了，我也倾向于这样做。

147
00:09:30,700 --> 00:09:36,380
When I have a really difficult question, I will often ask ChatGPT and Claude, both because, uh,
当我有一个非常困难的问题时，我经常会问 ChatGPT 和 Claude，因为，呃，

148
00:09:36,420 --> 00:09:41,420
sometimes you get answers that are too long winded or take you in too many different directions, and
有时你得到的答案过于冗长或带你走向太多不同的方向，并且

149
00:09:41,420 --> 00:09:43,780
so asking a couple of different models can help.
因此询问几个不同的型号会有所帮助。

150
00:09:43,820 --> 00:09:48,020
And you can also basically have like an agent workflow that you do manually.
您基本上也可以像手动执行的代理工作流程一样。

151
00:09:48,020 --> 00:09:54,100
You can ask a question to ChatGPT and then to Claude, you can say, I've got this response to my question.
你可以向 ChatGPT 提问，然后向 Claude 提问，你可以说，我的问题得到了这样的答复。

152
00:09:54,140 --> 00:09:54,820
Do you agree?
你同意？

153
00:09:54,820 --> 00:09:55,900
Is this accurate?
这准确吗？

154
00:09:55,940 --> 00:10:01,540
You can have it be the evaluator, like the evaluator optimizer pattern that we looked at a moment ago.
您可以将其作为评估器，就像我们刚才看到的评估器优化器模式一样。

155
00:10:01,540 --> 00:10:03,980
So that's a that's really cool that you can do it manually.
所以这真的很酷，你可以手动完成。

156
00:10:03,980 --> 00:10:09,700
And it's a great technique to get good answers from Llms when you're stuck yourself.
当你自己陷入困境时，这是一个很好的技巧，可以从法学硕士那里得到很好的答案。

157
00:10:10,300 --> 00:10:10,820
Okay.
好的。

158
00:10:11,100 --> 00:10:14,500
With that preamble, it's time for our next lab.
有了这个序言，我们的下一个实验室就到了。

159
00:10:14,660 --> 00:10:15,700
Let's get to it.
让我们开始吧。