Prompting with LangChain

LangChain standardise messages for various integrations by using three types of messages:

• SystemMessage: Provide context and guidance for the conversation
• HumanMessages: Represent prompts or questions from user
• AIMessages: Represent response from language models

Messages are commonly stored in a list then pass to the Large Language Model (LLM).

Using Messages Abstraction

from langchain_core.messages import SystemMessage, HumanMessage, AIMessage
messages = [SystemMessage("You are a friendly assistant"),
            HumanMessage("What is LangChain?")]

In the example above, we use SystemMessage to set the LLM tone and HumanMessage as a prompt to the LLM.

Using LLM Integrations in LangChain

LangChain also tries to standardise the way we instantiate LLMs which typically follow:

from langchain_{company_name} import Chat{company_name}
llm = Chat{company_name}(..."pass in model + api_key if not in environment"...)

OR

from langchain_community.{provider_name} import Chat{model_name}
llm = Chat{model_name}(..."pass in model + api_key if not in environment"...)

It also implements these standard methods:

• .invoke(): takes a list / coerce messages into a list to be used as input to LLM
• .stream(): stream the output as LLM are generating the content
• .batch(): perform multiple requests to the LLM at the same time
• .bind_tools(): bind tools to your model for tool calling
• .with_structured_output(): set rules to get an expected output

from langchain_core.messages import SystemMessage, HumanMessage, AIMessage
messages = [SystemMessage("You are a friendly assistant"),
            HumanMessage("What is LangChain?")]

from langchain_openai import ChatOpenAI
llm = ChatOpenAI(model = 'gpt-4o-mini', temperature = 0)
response = llm.invoke(messages)
print(response)

With the example above, we have learnt how to connect with different providers and perform basic prompting like how we can do in their respective native APIs.

Hands on: Prompt Adventure

Use the LangChain messages abstraction and LLM integration to ask a question related to culture in your country:

• Your system message should state that the LLM is an expert in culture
• Your human message should ask about a local festival in your country

Solution

from langchain_core.messages import SystemMessage, HumanMessage
from langchain_openai import ChatOpenAI
messages = [
    SystemMessage("You are a Singaporean which is expert in all cultures and heritage here.")
    HumanMessage("Can you explain more on Chingay?")
]
llm = ChatOpenAI(model = 'gpt-4o-mini', temperature = 0)
response = llm.invoke(messages)
print(response)

The message you would get back after invoking is an AIMessage. With this knowledge, create an interactive chatbot by extending your code above:

Extension

from langchain_core.messages import SystemMessage, HumanMessage
from langchain_openai import ChatOpenAI
llm = ChatOpenAI(model = 'gpt-4o-mini', temperature = 0)
messages = [
    SystemMessage("You are a Singaporean which is expert in all cultures and heritage here."),
]
while True:
    question = str(input("Enter your question here:  (Enter q / Q to quit) "))
    if question.lower() == 'q':
        print("Thanks for Prompting!")
        break
    messages.append(HumanMessage(question))
    response = llm.invoke(messages)
    messages.append(response)
    print(response.content)

Further Readings:

LangChain Messages Guide

LangChain Chat Model (LLMs) Conceptual Guide

API Reference for BaseChatModel