Prompt Template

• Author: Hye-yoon

• Design:

• Peer Review :

• This is a part of LangChain Open Tutorial

Overview

This tutorial covers how to create and utilize prompt templates using LangChain .

Prompt templates are essential for generating dynamic and flexible prompts that cater to various use cases, such as conversation history, structured outputs, and specialized queries.

In this tutorial, we will explore methods for creating PromptTemplate objects, applying partial variables, managing templates through YAML files, and leveraging advanced tools like ChatPromptTemplate and MessagePlaceholder for enhanced functionality.

Table of Contents

• Overview

• Environment Setup

• Creating a PromptTemplate Object

• Using partial_variables

• Load prompt template from YAML file

• ChatPromptTemplate

• MessagePlaceholder

References

• LangChain Documentation : Prompts

---

Environment Setup

Set up the environment. You may refer to Environment Setup for more details.

[Note]

• langchain-opentutorial is a package that provides a set of easy-to-use environment setup, useful functions and utilities for tutorials.

• You can check out the langchain-opentutorial for more details.

%%capture --no-stderr
%pip install langchain-opentutorial

    [notice] A new release of pip is available: 24.1 -> 24.3.1
    [notice] To update, run: python.exe -m pip install --upgrade pip

# Install required packages
from langchain_opentutorial import package

package.install(
    [
        "langsmith",
        "langchain",
        "langchain_core",
        "langchain_community",
        "langchain_openai",
    ],
    verbose=False,
    upgrade=False,
)

from dotenv import load_dotenv

load_dotenv(override=True)

True

# Set environment variables
from langchain_opentutorial import set_env

set_env(
    {
        # "OPENAI_API_KEY": "",
        # "LANGCHAIN_API_KEY": "",
        "LANGCHAIN_TRACING_V2": "true",
        "LANGCHAIN_ENDPOINT": "https://api.smith.langchain.com",
        "LANGCHAIN_PROJECT": "Prompt-Template",
    }
)

Environment variables have been set successfully.

Let's setup ChatOpenAI with gpt-4o model.

from langchain_openai import ChatOpenAI

# Load the model
llm = ChatOpenAI(model_name="gpt-4o")

Creating a PromptTemplate Object

There are two ways to create a PromptTemplate object.

• 1. Using the from_template() method.

• 1. Creating a PromptTemplate object and generating a prompt simultaneously.

Method 1. Using from_template() method

• Define template with variable as {variable} .

from langchain_core.prompts import PromptTemplate

# Define template. In this case, {country} is a variable
template = "What is the capital of {country}?"

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

PromptTemplate(input_variables=['country'], input_types={}, partial_variables={}, template='What is the capital of {country}?')

You can complete the prompt by assigning a value to the variable country .

# Create prompt. Assign value to the variable using `format` method
prompt = prompt.format(country="United States of America")
prompt

'What is the capital of United States of America?'

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

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

# Create chain
chain = prompt | llm

# Replace the country variable with a value of your choice
chain.invoke("United States of America").content

'The capital of the United States of America is Washington, D.C.'

Method 2. Creating a PromptTemplate object and a prompt all at once.

Explicitly specify input_variables for additional validation.

Otherwise, a mismatch between such variables and the variables within the template string can raise an exception in instantiation.

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

# Create a prompt template with `PromptTemplate` object
prompt = PromptTemplate(
    template=template,
    input_variables=["country"],
)
prompt

PromptTemplate(input_variables=['country'], input_types={}, partial_variables={}, template='What is the capital of {country}?')

# Create prompt
prompt.format(country="United States of America")

'What is the capital of United States of America?'

# Define template
template = "What are the capitals of {country1} and {country2}, respectively?"

# Create a prompt template with `PromptTemplate` object
prompt = PromptTemplate(
    template=template,
    input_variables=["country1"],
    partial_variables={
        "country2": "United States of America"  # Pass `partial_variables` in dictionary form
    },
)
prompt

PromptTemplate(input_variables=['country1'], input_types={}, partial_variables={'country2': 'United States of America'}, template='What are the capitals of {country1} and {country2}, respectively?')

prompt.format(country1="South Korea")

'What are the capitals of South Korea and United States of America, respectively?'

prompt_partial = prompt.partial(country2="India")
prompt_partial

PromptTemplate(input_variables=['country1'], input_types={}, partial_variables={'country2': 'India'}, template='What are the capitals of {country1} and {country2}, respectively?')

prompt_partial.format(country1="South Korea")

'What are the capitals of South Korea and India, respectively?'

chain = prompt_partial | llm

chain.invoke("United States of America").content

'The capital of the United States of America is Washington, D.C., and the capital of India is New Delhi.'

chain.invoke({"country1": "United States of America", "country2": "India"}).content

'The capital of the United States of America is Washington, D.C., and the capital of India is New Delhi.'

Using partial_variables

Using partial_variables , you can partially apply functions. This is particularly useful when there are common variables to be shared.

Common examples are date or time.

Suppose you want to specify the current date in your prompt, hardcoding the date into the prompt or passing it along with other input variables may not be practical. In this case, using a function that returns the current date to modify the prompt partially is much more convenient.

from datetime import datetime

# Print the current date
datetime.now().strftime("%B %d")

'January 14'

# Define function to return the current date
def get_today():
    return datetime.now().strftime("%B %d")

prompt = PromptTemplate(
    template="Today's date is {today}. Please list {n} celebrities whose birthday is today. Please specify their date of birth.",
    input_variables=["n"],
    partial_variables={
        "today": get_today  # Pass `partial_variables` in dictionary form
    },
)

# Create prompt
prompt.format(n=3)

"Today's date is January 14. Please list 3 celebrities whose birthday is today. Please specify their date of birth."

# Create chain
chain = prompt | llm

# Invoke chain and check the result
print(chain.invoke(3).content)

Here are three celebrities born on January 14:
    
    1. **Dave Grohl** - Born on January 14, 1969.
    2. **LL Cool J** - Born on January 14, 1968.
    3. **Jason Bateman** - Born on January 14, 1969.

# Invoke chain and check the result
print(chain.invoke({"today": "Jan 02", "n": 3}).content)

Here are three celebrities born on January 2:
    
    1. **Cuba Gooding Jr.** - Born on January 2, 1968.
    2. **Taye Diggs** - Born on January 2, 1971.
    3. **Kate Bosworth** - Born on January 2, 1983.

Load prompt template from YAML file

You can manage prompt templates in seperate yaml files and load using load_prompt .

from langchain_core.prompts import load_prompt

prompt = load_prompt("prompts/fruit_color.yaml", encoding="utf-8")
prompt

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

prompt.format(fruit="an apple")

'What is the color of an apple?'

prompt2 = load_prompt("prompts/capital.yaml")
print(prompt2.format(country="United States of America"))

Please provide information about the capital city of United States of America.
    Summarize the characteristics of the capital in the following format, within 300 words.
    ----
    [Format]
    1. Area
    2. Population
    3. Historical Sites
    4. Regional Products
    
    #Answer:
    

ChatPromptTemplate

ChatPromptTemplate can be used to include a conversation history as a prompt.

Messages are structured as tuples in the format (role , message ) and are created as a list.

role

• "system" : A system setup message, typically used for global settings-related prompts.

• "human" : A user input message.

• "ai" : An AI response message.

from langchain_core.prompts import ChatPromptTemplate

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

ChatPromptTemplate(input_variables=['country'], input_types={}, partial_variables={}, messages=[HumanMessagePromptTemplate(prompt=PromptTemplate(input_variables=['country'], input_types={}, partial_variables={}, template='What is the capital of {country}?'), additional_kwargs={})])

chat_prompt.format(country="United States of America")

'Human: What is the capital of United States of America?'

from langchain_core.prompts import ChatPromptTemplate

chat_template = ChatPromptTemplate.from_messages(
    [
        # role, message
        ("system", "You are a friendly AI assistant. Your name is {name}."),
        ("human", "Nice to meet you!"),
        ("ai", "Hello! How can I assist you?"),
        ("human", "{user_input}"),
    ]
)

# Create chat messages
messages = chat_template.format_messages(name="Teddy", user_input="What is your name?")
messages

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

You can directly invoke LLM using the messages created above.

llm.invoke(messages).content

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

You can also create a chain to execute.

chain = chat_template | llm

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

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

MessagePlaceholder

LangChain also provides a MessagePlaceholder , which provides complete control over rendering messages during formatting.

This can be useful if you’re unsure which roles to use in a message prompt template or if you want to insert a list of messages during formatting.

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

chat_prompt = ChatPromptTemplate.from_messages(
    [
        (
            "system",
            "You are a summarization specialist AI assistant. Your mission is to summarize conversations using key points.",
        ),
        MessagesPlaceholder(variable_name="conversation"),
        ("human", "Summarize the conversation so far in {word_count} words."),
    ]
)
chat_prompt

ChatPromptTemplate(input_variables=['conversation', 'word_count'], input_types={'conversation': list[typing.Annotated[typing.Union[typing.Annotated[langchain_core.messages.ai.AIMessage, Tag(tag='ai')], typing.Annotated[langchain_core.messages.human.HumanMessage, Tag(tag='human')], typing.Annotated[langchain_core.messages.chat.ChatMessage, Tag(tag='chat')], typing.Annotated[langchain_core.messages.system.SystemMessage, Tag(tag='system')], typing.Annotated[langchain_core.messages.function.FunctionMessage, Tag(tag='function')], typing.Annotated[langchain_core.messages.tool.ToolMessage, Tag(tag='tool')], typing.Annotated[langchain_core.messages.ai.AIMessageChunk, Tag(tag='AIMessageChunk')], typing.Annotated[langchain_core.messages.human.HumanMessageChunk, Tag(tag='HumanMessageChunk')], typing.Annotated[langchain_core.messages.chat.ChatMessageChunk, Tag(tag='ChatMessageChunk')], typing.Annotated[langchain_core.messages.system.SystemMessageChunk, Tag(tag='SystemMessageChunk')], typing.Annotated[langchain_core.messages.function.FunctionMessageChunk, Tag(tag='FunctionMessageChunk')], typing.Annotated[langchain_core.messages.tool.ToolMessageChunk, Tag(tag='ToolMessageChunk')]], FieldInfo(annotation=NoneType, required=True, discriminator=Discriminator(discriminator=, custom_error_type=None, custom_error_message=None, custom_error_context=None))]]}, partial_variables={}, messages=[SystemMessagePromptTemplate(prompt=PromptTemplate(input_variables=[], input_types={}, partial_variables={}, template='You are a summarization specialist AI assistant. Your mission is to summarize conversations using key points.'), additional_kwargs={}), MessagesPlaceholder(variable_name='conversation'), HumanMessagePromptTemplate(prompt=PromptTemplate(input_variables=['word_count'], input_types={}, partial_variables={}, template='Summarize the conversation so far in {word_count} words.'), additional_kwargs={})])

You can use MessagesPlaceholder to add the conversation message list.

formatted_chat_prompt = chat_prompt.format(
    word_count=5,
    conversation=[
        ("human", "Hello! I’m Teddy. Nice to meet you."),
        ("ai", "Nice to meet you! I look forward to working with you."),
    ],
)

print(formatted_chat_prompt)

System: You are a summarization specialist AI assistant. Your mission is to summarize conversations using key points.
    Human: Hello! I’m Teddy. Nice to meet you.
    AI: Nice to meet you! I look forward to working with you.
    Human: Summarize the conversation so far in 5 words.

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

# Invoke chain and check the result
chain.invoke(
    {
        "word_count": 5,
        "conversation": [
            (
                "human",
                "Hello! I’m Teddy. Nice to meet you.",
            ),
            ("ai", "Nice to meet you! I look forward to working with you."),
        ],
    }
)

'Teddy introduces himself, exchanges greetings.'