1
00:00:00,700 --> 00:00:05,950
Let s go ahead and talk about another
core component of Langchain called prompt

2
00:00:05,960 --> 00:00:10,530
templates which are like recipes for
generating prompts for LLMs.

3
00:00:11,540 --> 00:00:15,370
A prompt refers to the input to the model.

4
00:00:16,840 --> 00:00:18,910
Prompt templates are a way to create

5
00:00:18,920 --> 00:00:24,970
dynamic prompts for LLMs that are more
flexible and easier to use than

6
00:00:24,980 --> 00:00:26,570
traditional static prompts.

7
00:00:27,440 --> 00:00:30,830
A prompt template takes a piece of text

8
00:00:30,840 --> 00:00:35,090
and injects a user s input into that
piece of text.

9
00:00:36,560 --> 00:00:42,950
Note that in Langchain, there are prompt
templates and chat prompt templates.

10
00:00:43,920 --> 00:00:49,830
Prompt templates are used for tasks that
involve generating text such as answering

11
00:00:49,840 --> 00:00:55,770
questions or completing sentences, while
chat prompt templates are specifically

12
00:00:55,780 --> 00:01:00,730
designed for tasks that involve engaging
in conversations.

13
00:01:02,040 --> 00:01:07,830
Let s see how to create and use a prompt
template that can be used with any type

14
00:01:07,840 --> 00:01:08,870
of LLM.

15
00:01:10,770 --> 00:01:16,650
From Langchain prompts import prompt

16
00:01:16,660 --> 00:01:28,940
template and from Langchain OpenAI, I am
importing the ChatOpenAI class.

17
00:01:32,360 --> 00:01:37,390
Even though it was probably already
imported in a cell above and it s not

18
00:01:37,400 --> 00:01:42,570
necessary here, I will import it again
just in case you want to run only this

19
00:01:42,580 --> 00:01:46,430
cell and the ones below and not the
entire notebook.

20
00:01:47,020 --> 00:01:51,030
I am creating a multiline string called template.

21
00:01:52,340 --> 00:01:55,790
This is similar to an F string and uses

22
00:01:55,800 --> 00:01:59,250
curly braces for the dynamic part of the template.

23
00:02:00,080 --> 00:02:05,140
You are an experienced virologist, write

24
00:02:05,150 --> 00:02:20,740
a few sentences about the following virus
and the virus in curly braces in and the

25
00:02:20,750 --> 00:02:22,620
language in curly braces.

26
00:02:28,720 --> 00:02:31,850
Now let s define a prompt object of the

27
00:02:31,860 --> 00:02:33,070
class prompt template.

28
00:02:34,320 --> 00:02:39,080
Prompt template equals the prompt

29
00:02:39,090 --> 00:02:49,880
template class constructor from template
of and the template equals template.

30
00:02:50,870 --> 00:02:56,720
This piece of code creates a new prompt
template object from the provided

31
00:02:56,730 --> 00:02:58,000
template string.

32
00:03:00,310 --> 00:03:03,720
Next, we ll format the prompt template by

33
00:03:03,730 --> 00:03:10,880
replacing the placeholders virus and
language in curly braces with the

34
00:03:10,890 --> 00:03:24,400
provided values prompt equals prompt
-template .format and the arguments virus

35
00:03:24,410 --> 00:03:30,880
equals HIV and the language equals German.

36
00:03:31,930 --> 00:03:37,240
In this case, the value HIV is used to

37
00:03:37,250 --> 00:03:44,960
replace the placeholder virus and the
value German is used to replace the

38
00:03:44,970 --> 00:03:46,460
placeholder language.

39
00:03:49,090 --> 00:03:52,200
This will result in the following prompt.

40
00:03:54,830 --> 00:04:01,540
You are an experienced virologist, write
a few sentences about the following virus

41
00:04:01,550 --> 00:04:03,020
in German.

42
00:04:04,610 --> 00:04:09,280
I can write curly braces virus in double quotes.

43
00:04:14,740 --> 00:04:17,310
I am creating the LLM object.

44
00:04:17,960 --> 00:04:24,690
LLM equals chatOpenAi, modelName equals

45
00:04:24,700 --> 00:04:32,180
gpt3 .5 turbo and the temperature equals 0.

46
00:04:33,410 --> 00:04:37,720
Next, I will invoke the LLM with the

47
00:04:37,730 --> 00:04:46,260
prompt string as an argument output
equals LLM .invoke of prompt.

48
00:04:51,220 --> 00:04:56,310
Finally, I am printing the response
output .content.

49
00:05:00,120 --> 00:05:01,190
I am running it.

50
00:05:10,490 --> 00:05:13,920
Ok, gpt3 .5 turbo in single quotes.

51
00:05:19,580 --> 00:05:23,410
I ve got a short response about HIV in German.

52
00:05:24,340 --> 00:05:26,110
That s it with prompt templates.

53
00:05:26,320 --> 00:05:32,190
We ll take a break and in the next video,
we ll go over the other type of prompt

54
00:05:32,200 --> 00:05:34,210
called chat prompt template.