1
00:00:00,630 --> 00:00:06,180
Besides simple prompt templates, there
are also chat prompt templates.

2
00:00:07,430 --> 00:00:12,960
Prompt templates are used for tasks that
involve generating text, such as

3
00:00:12,970 --> 00:00:18,360
answering questions or completing
sentences, and chat prompt templates are

4
00:00:18,370 --> 00:00:23,320
specifically designed for tasks that
involve engaging in conversations.

5
00:00:24,110 --> 00:00:30,180
As we ve already discussed, the
ChatCompletions API consists of a list of

6
00:00:30,190 --> 00:00:35,660
messages, SystemMessage, HumanMessage, or
the prompt and a message.

7
00:00:36,850 --> 00:00:43,820
I will import some classes from the
Langchain library, from LangchainPrompts,

8
00:00:44,530 --> 00:00:53,600
Import, ChatPromptTemplate and the
HumanMessagePromptTemplate, and from

9
00:00:53,610 --> 00:01:03,880
LangchainCore, Messages, Import, SystemMessage.

10
00:01:07,680 --> 00:01:10,090
HumanMessagePromptTemplate is used to

11
00:01:10,100 --> 00:01:16,470
create dynamic prompts from the user s
text and the SystemMessage is used to

12
00:01:16,480 --> 00:01:17,830
format the system message.

13
00:01:19,280 --> 00:01:22,570
Note that there is also a class called

14
00:01:24,600 --> 00:01:29,470
ChatSystemMessagePromptTemplate if you
want to generate dynamic system messages.

15
00:01:33,270 --> 00:01:38,580
Now, I will create a new
ChatPromptTemplate object from a list of messages.

16
00:01:39,510 --> 00:01:52,620
ChatTemplate equals ChatPromptTemplate
.fromMessages and the argument will be a list.

17
00:01:53,330 --> 00:02:02,830
The first message is the SystemMessage
and the second message is the

18
00:02:02,840 --> 00:02:07,730
HumanMessagePromptTemplate that
represents the user s input.

19
00:02:11,990 --> 00:02:12,660
Now, the arguments.

20
00:02:14,130 --> 00:02:18,680
SystemMessage of content equals and the string.

21
00:02:20,840 --> 00:02:31,710
You respond only in the JSON format and
the HumanMessagePromptTemplate

22
00:02:31,720 --> 00:02:45,480
.fromTemplate, I m calling the
fromTemplate function of top n countries

23
00:02:45,490 --> 00:02:51,560
in area in curly braces by population.

24
00:02:53,030 --> 00:02:57,900
The placeholders n and area will be

25
00:02:57,910 --> 00:03:00,600
replaced with the actual input text.

26
00:03:02,250 --> 00:03:04,300
Next, I will format the

27
00:03:04,310 --> 00:03:11,060
ChatPromptTemplate object by replacing
the placeholders with the provided value.

28
00:03:12,490 --> 00:03:28,780
So, messages equals ChatTemplate
.formatMessage of n equals 10 as a string

29
00:03:28,790 --> 00:03:31,020
and area equals Europe.

30
00:03:34,410 --> 00:03:37,980
The resulting messages list contains the

31
00:03:37,990 --> 00:03:41,820
formatted SystemMessage and HumanMessage.

32
00:03:42,410 --> 00:03:45,580
Let s print it print messages and I

33
00:03:45,590 --> 00:03:46,120
running the code.

34
00:03:48,410 --> 00:03:50,460
This is the formatted prompt.

35
00:03:51,290 --> 00:03:54,580
It contains a SystemMessage and a HumanMessage.

36
00:03:57,800 --> 00:04:01,590
n in curly braces was replaced with pen

37
00:04:01,600 --> 00:04:05,970
and area in curly braces was replaced
with Europe.

38
00:04:21,070 --> 00:04:35,560
From going with the defaults output
equals llm .invoke of messages and I m

39
00:04:35,570 --> 00:04:39,560
printing the output printOutput .content.

40
00:04:42,480 --> 00:04:48,420
I m running the code very well.

41
00:04:49,210 --> 00:04:55,680
Notice how it responded in JSON according
to the instructions in the SystemPrompt.

42
00:05:01,830 --> 00:05:05,420
You ve learned about prompt templates and
chat prompt templates.

43
00:05:05,930 --> 00:05:10,320
They give you a lot of flexibility in the
way you build your chat prompts.

44
00:05:11,250 --> 00:05:18,780
Some benefits of using prompt templates
include reusability, scalability and accuracy.

45
00:05:19,670 --> 00:05:26,240
For example, if you want a response with
the top 5 countries by population in the

46
00:05:26,250 --> 00:05:34,700
world, you can simply change the argument
of chatTemplate .formatMessage, 5 and 1.

47
00:05:37,450 --> 00:05:39,380
You don t have to change the entire prompt.

48
00:05:43,140 --> 00:05:45,930
We are done with prompt templates for the

49
00:05:45,940 --> 00:05:51,470
moment, please take your time to fully
understand them as they are a key

50
00:05:51,480 --> 00:05:53,030
component of Langchain.