01. Prompt

PromptTemplate

Copy

from dotenv import load_dotenv

load_dotenv()

Copy

# Set up LangSmith tracking: https:
# !pip install -qU langchain-teddynote
from langchain_teddynote import logging

# Enter a project name.
logging.langsmith("CH02-Prompt")

Copy

Start tracking LangSmith. 
[Project name] 
CH02-Prompt 

Define LLM objects.

Copy

from langchain_openai import ChatOpenAI

llm = ChatOpenAI()

Method 1. Create a PromptTemplate object using the from_template() method

• Variable to be substituted { variable } Define the template by tying it together.

Copy

from langchain_core.prompts import PromptTemplate 

# template definition. {country} is a variable that will be used as a placeholder for
the value that follows template = "What is the capital of {country}?" 

# Create a PromptTemplate object using the from_template method
prompt = PromptTemplate.from_template(template) prompt

Copy

 PromptTemplate (input_variables=['country'], template='{country}) 

country You can create sentences by putting values in variables.

Copy

# Create a prompt. Use the format method to put values ​​into variables.
prompt = prompt.format(country="South Korea") prompt

Copy

 'Where is the capital of Korea?' 

Copy

# Define a template template = "What is the capital of {country}?" 

#Create a Prompt Template object using the from_template method 
prompt = PromptTemplate.from_template(template) 

# Create a chain

#chain = prompt
chain = prompt | llm

Copy

# The value entered in the country variable is automatically substituted and executed
chain.invoke("Republic of Korea").content

Copy

 'The capital of Korea is Seoul.' 

Method 2. Create PromptTemplate objects simultaneously with prompt

For further validation input_variables Please specify

These variables cause exceptions when inconsistent compared to the variables in the template string during instantiation.

Copy

# template definition
template = "{country}"What is the capital of {country}?"

# Create a prompt_template using the PromptTemplate object
prompt = PromptTemplate(
    template=template,
    input_variables=["country"],
)

prompt

Copy

 PromptTemplate (input_variables=['country'], template='{country}) 

Copy

# Create a prompt 
prompt.format(country="South Korea")

Copy

'Where is the capital of Korea?' 

Copy

# template definition
template = "{country1}class {country2}"Where are the capitals of each?"

# PromptTemplate by using objects prompt_template generation
prompt = PromptTemplate(
    template=template,
    input_variables=["country1"],
    partial_variables={
        "country2": "USA"  # dictionary in the form of partial_variables to pass on
    },
)

prompt

Copy

 PromptTemplate (input_variables=['country1'], partial_variables={'country2':'USA'}, 
 template='{country1} and {country2} 

Copy

prompt.format(country1="Canada")
prompt_partial

Copy

PromptTemplate (input_variables=['country1'], partial_variables={'country2':'Canada'}, 
template='{country1} and {country2} 

Copy

prompt_partial.format(country1="korea")

Copy

 'Where are the capitals of Korea and Canada?' 

Copy

chain = prompt_partial | llm

Copy

chain.invoke("korea").content

Copy

 'The capital of Korea is Seoul, and the capital of Canada is Ottawa.' 

Copy

chain.invoke({"country1": "korea", "country2": "australia"}).content

Copy

'The capital of Korea is Seoul and the capital of Australia is Canberra.'

partial_variables : Fill partial variables

partial The general use of is to partially use the function. This use case Variables you always want to bring in a common way This is the case.

Representative example Date or time is.

Let's say you always have a prompt that you want the current date to be displayed. You can't even hardcode the prompt, and passing it along with other input variables is cumbersome. In this case always present Function to return date It is very convenient if you can partially change the prompt using.

The following code is the Python code that saves the date today.

Copy

from datetime import datetime

# Print today's date
datetime.now().strftime("%B %d")

Copy

'June 20' 

Copy

prompt = PromptTemplate(
    template="Today's date is {today} no see. Celebrities whose birthday is today {n} Please list the names. Please indicate your date of birth.",
    input_variables=["n"],
    partial_variables={
        "today": get_today  # dictionary in the form of partial_variables to pass on
    },
)

Copy

# prompt generation
prompt.format(n=3)

Copy

 'Today's date is June 20. List three celebrities this birthday today.
  Please indicate your date of birth.' 

Copy

# chain creates.
chain = prompt | llm

Copy

# chain after running, check the results.
print(chain.invoke(3).content)

Copy

One. Nicole Kidman-June 1967 20 
2. John Goodman-June 1952 20 
3. Lionel Richie-June 1949

Copy

# chain After running, check the results.
print(chain.invoke({"today": "Jan 02", "n": 3}).content)

Copy

One. Kate Bosworth -2 January 1983 
2. Tia Carrere-2 January 1967 
3. Christy Turlington-2 January 1969 

Reading template from file

Copy

from langchain_core.prompts import load_prompt

prompt = load_prompt("prompts/fruit_color.yaml")
prompt

Copy

 PromptTemplate (input_variables=['fruit'], template='{fruit} What is the color of?') 

If any of the Window users have an error, run with the code below (encoding settings)

Copy

# from langchain_teddynote.prompts import load_prompt

# # Windows user only: encoding cp949 set to
# load_prompt("prompts/fruit_color.yaml", encoding="cp949")

Copy

prompt.format(fruit="apologize")

Copy

 'What is the color of the apple?'

Copy

prompt2 = load_prompt("prompts/capital.yaml")
print(prompt2.format(country="korea"))

Copy

 Please tell us about the capital of Korea. 
Organize the features of the capital to the following form. 
Please write 300 characters inside and outside. 
Please write in Koreanglo. 
---- 
[Form] 
One. area 
2. population 
3. Historical place 
4. Specialties 

#Answer: 

ChatPromptTemplate

ChatPromptTemplate You can take advantage of it when you want to inject a silver chat list as a prompt.

The message is organized in tuple format, ( role , message ) Can be configured to create a list.

role - "system" : System setting message. This is mainly a prompt related to global settings. - "human" : User input message. - "ai" : AI's answer message.

Copy

from langchain_core.prompts import ChatPromptTemplate

chat_prompt = ChatPromptTemplate.from_template("{country}what is the capital of?")
chat_prompt

Copy

 ChatPromptTemplate (input_variables=['country'], messages=[HumanMessagePromptTemplate (prompt=Prompt
 Template) (input_varibles=['country'], template= 

Copy

chat_prompt.format(country="korea")

Copy

 'Human: Where is the capital of Korea?' 

Copy

from langchain_core.prompts import ChatPromptTemplate

chat_template = ChatPromptTemplate.from_messages(
    [
        # role, message
        ("system", "You are kind AI I am an assistant. Your name is {name} no see."),
        ("human", "nice to meet you!"),
        ("ai", "Hello! How can I help you?"),
        ("human", "{user_input}"),
    ]
)

# chat message creates.
messages = chat_template.format_messages(
    name="teddy", user_input="What is your name?"
)
messages

Copy

 [SystemMessage (content=' You are a friendly AI assistant. Your name is Teddy.'), 
 HumanMessage (content='Nice to meet you!'), AIMessage (content='Hello! 
 What can I help you with?'), HumanMessage (content='What is your name?')] 

You can get the result by injecting the message you generated right away.

Copy

llm.invoke(messages).content

Copy

 'My name is Teddy. Please tell me if you need help!' 

This time, let's create a chain.

Copy

chain = chat_template | llm

Copy

chain.invoke({"name": "Teddy", "user_input": "what is your name?"}).content

Copy

'My name is Teddy. How can I help you?' 

MessagePlaceholder

LangChain also has full control over the messages to be rendered during formatting MessagePlaceholder Gives.

It may be useful if you are not sure which role to use for the message prompt template, or if you want to insert a list of messages during formatting.

Copy

from langchain_core.output_parsers import StrOutputParser
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder

chat_prompt = ChatPromptTemplate.from_messages(
    [
        (
            "system",
            "You are a summary expert AI You are an assistant. Your job is to summarize the conversation with key words..",
        ),
        MessagesPlaceholder(variable_name="conversation"),
        ("human", "The conversation so far {word_count} Summarize in words."),
    ]
)
chat_prompt

Copy

 ChatPromptTemplate (input_varibles=['conversation','word_count'], 
 input_types={'conversation': typing.List.AIMessage, langchain_core.mess 
 Your mission is to summarize the conversation with the main keywords.')), 
 MessagesPlaceholder (variable_name='conversation'), 
 HumanMessagePromptTemplate (prompt=PromptTemplate (input_varibles=['word_count'], 
 template<TA 

conversation When I want to add a dialog later MessagesPlaceholder You can use

Copy

formatted_chat_prompt = chat_prompt.format(
    word_count=5,
    conversation=[
        ("human", "Hello! I'm Teddy, a new employee today. Nice to meet you."),
        ("ai", "Nice to meet you! I look forward to working with you in the future."),
    ],
)

print(formatted_chat_prompt)

Copy

 System: You are a summary expert AI assistant. Your mission is to summarize the conversation with the main keyword. 
Human: Hello! I am a new Teddy today. Nice to meet you. 
AI: Nice to meet you! Please do well in the future. 
Human: Summarize the conversation so far in 5 words. 

Copy

# chain generation
chain = chat_prompt | llm | StrOutputParser()

Copy

# chain run and check results
chain.invoke(
    {
        "word_count": 5,
        "conversation": [
            (
                "human",
                "Hello! I'm Teddy, a new employee today. Nice to meet you.",
            ),
            ("ai", "Nice to meet you! I look forward to working with you in the future."),
        ],
    }
)

Copy

 'Meeting a new tenant Teddy.'