1
00:00:00,700 --> 00:00:07,410
Welcome back, the core element of any LLM
application is the AI model.

2
00:00:08,080 --> 00:00:14,010
Langchain doesn t have its own LLMs, but
it provides a standard interface for

3
00:00:14,020 --> 00:00:20,370
interacting with various LLMs, including
OpenAI s GPT models, Google s Gemini,

4
00:00:20,860 --> 00:00:22,450
Metas, LAMA and more.

5
00:00:23,040 --> 00:00:25,790
In Langchain terminology, these are

6
00:00:25,800 --> 00:00:27,350
called LLM providers.

7
00:00:28,540 --> 00:00:31,850
Now, let s dive into how to invoke OpenAI

8
00:00:31,860 --> 00:00:38,530
s GPT models like GPT -3 .5 Turbo, GPT -4
or GPT -4 Turbo within Langchain.

9
00:00:39,020 --> 00:00:46,750
These models expose an interface where
chat messages or conversations serve as

10
00:00:46,760 --> 00:00:48,290
inputs and outputs.

11
00:00:49,580 --> 00:00:55,360
From Langchain OpenAI, I am importing

12
00:00:55,370 --> 00:01:04,160
chatOpenA. Now, let s create an LLM
object, LLM equals chatOpenA.

13
00:01:06,410 --> 00:01:14,440
Next, I ll use LLM .invoke to send a
request to the OpenAI GPT model.

14
00:01:15,370 --> 00:01:18,200
This method takes the prompt as an argument.

15
00:01:20,070 --> 00:01:24,380
For example, explain quantum mechanics in

16
00:01:24,390 --> 00:01:25,040
one sentence.

17
00:01:33,510 --> 00:01:36,720
The invoke method returns an output

18
00:01:36,730 --> 00:01:45,320
object, and the content attribute of the
output object contains the text response

19
00:01:45,330 --> 00:01:47,760
from the OpenAI GPT model.

20
00:01:48,470 --> 00:01:53,540
I am running the code very well.

21
00:01:54,510 --> 00:02:01,320
Notice that I called LLM .invoke without
any arguments, and you might be wondering

22
00:02:01,330 --> 00:02:04,100
which model and parameters it used.

23
00:02:05,090 --> 00:02:08,360
To view the constructor s arguments along

24
00:02:08,370 --> 00:02:12,660
with their defaults, let s check the
built -in help of the class.

25
00:02:13,590 --> 00:02:16,180
Help, chatOpenA.

26
00:02:22,630 --> 00:02:24,680
As you can see, there are several

27
00:02:24,690 --> 00:02:30,760
available arguments of particular
interest to us are model and temperature.

28
00:02:31,830 --> 00:02:38,520
Note that the default model is GPT 3 .5
turbo and the default value for the

29
00:02:38,530 --> 00:02:41,640
temperature is 0 .7.

30
00:02:42,330 --> 00:02:45,080
I ve delved into OpenAI s chat

31
00:02:45,090 --> 00:02:50,580
completions parameters during the OpenAI
API with Python course.

32
00:02:51,510 --> 00:02:59,280
If you prefer to use GPT 4 or GPT 4 turbo
instead of GPT 3 .5 turbo, simply change

33
00:02:59,290 --> 00:03:00,280
the model name.

34
00:03:01,510 --> 00:03:04,780
I am adding a new argument model equals

35
00:03:04,790 --> 00:03:07,040
GPT 4 turbo preview.

36
00:03:07,610 --> 00:03:10,880
This is the name for GPT 4 turbo, and I m

37
00:03:10,890 --> 00:03:11,480
running the code.

38
00:03:14,370 --> 00:03:18,340
Note that at this moment, GPT 4 turbo is

39
00:03:18,350 --> 00:03:20,820
available only to paid accounts.

40
00:03:21,550 --> 00:03:24,700
For this course, we ll stick with GPT 3

41
00:03:24,710 --> 00:03:29,540
.5 turbo because it s both cost
-effective and powerful.

42
00:03:35,070 --> 00:03:39,040
Let s take a look at the OpenAI chat
completions API.

43
00:03:43,410 --> 00:03:52,280
The OpenAI API for GPT 3 .5 turbo, GPT 4,
GPT 4 turbo and so on uses a list of

44
00:03:52,290 --> 00:03:54,260
dictionaries called messages.

45
00:03:55,130 --> 00:03:59,980
These messages define three roles System,

46
00:04:00,670 --> 00:04:02,280
User and Assistant.

47
00:04:03,870 --> 00:04:07,820
The System role helps set the behavior of

48
00:04:07,830 --> 00:04:14,960
the Assistant, the User message serves as
the prompt or the question we ask the

49
00:04:14,970 --> 00:04:20,320
Assistant, and the Assistant messages
store prior responses.

50
00:04:21,310 --> 00:04:28,060
To use these messages when calling the
LLM from Langchain, we need to import a

51
00:04:28,070 --> 00:04:29,680
schema for the messages.

52
00:04:30,670 --> 00:04:39,120
From Langchain Schema Import, and I will

53
00:04:39,130 --> 00:04:42,680
import three classes for the three
message types,

54
00:04:57,820 --> 00:05:05,210
In the OpenAI chat completions API, a
message is equivalent to the Assistant

55
00:05:05,220 --> 00:05:10,950
message and a human message represents
the user s message or prompt.

56
00:05:11,760 --> 00:05:14,330
Now, let s create a list called messages.

57
00:05:16,310 --> 00:05:19,360
The list will contain a System message

58
00:05:19,370 --> 00:05:21,480
and a human message which is the prompt.

59
00:05:22,350 --> 00:05:29,060
System message of content equals you are

60
00:05:29,070 --> 00:05:48,120
a physicist and respond only in German
and the human message content equals and

61
00:05:48,130 --> 00:05:52,700
the same task explain quantum mechanics
in one sentence.

62
00:05:57,430 --> 00:06:05,900
Finally, let s make the API call to
OpenAI output equals LLM .invoke of

63
00:06:05,910 --> 00:06:13,340
messages and print output .content.

64
00:06:16,180 --> 00:06:22,840
I am running the code very well, the LLM

65
00:06:22,850 --> 00:06:28,720
followed my instructions and responded in
one sentence in German.