Formatting Outputs for ChatPrompt Templates(two)

https://python.langchain.com.cn/docs/modules/model_io/prompts/prompt_templates/format_output

This guide explains how to use the format method of ChatPrompt (in LangChain) to get outputs in three useful formats. All examples use the same core task: creating a prompt for translating English to French. We'll keep code and outputs exactly as in the original source---no changes.

Key Background First

Before diving into formats, remember: A ChatPrompt typically includes a SystemMessage (tells the AI its role) and a HumanMessage (the user's input). For our examples, the ChatPrompt is set up to translate text from an input_language to an output_language (we'll use English → French).

1. Output as a String

The simplest format: a plain text string that combines the system message and human message. There are two equivalent ways to get this.

Method 1: Use chat_prompt.format()

This directly returns the prompt as a string.

Code (From Original Source)

# Assume `chat_prompt` is already set up for translation (English → French)
output = chat_prompt.format(
    input_language="English", 
    output_language="French", 
    text="I love programming."
)
print(output)

Output (From Original Source)

System: You are a helpful assistant that translates English to French.
Human: I love programming.

Method 2: Use chat_prompt.format_prompt().to_string()

This is a two-step way to get the same string. First, format_prompt() creates a ChatPromptValue (see Section 2), then to_string() converts it to text.

Code (From Original Source)

output_2 = chat_prompt.format_prompt(
    input_language="English", 
    output_language="French", 
    text="I love programming."
).to_string()

# Check if both outputs are identical (they will be!)
assert output == output_2  # No error means they match

What This Means

Both methods give you a readable text string. Use this if you want to quickly check or share the prompt content.

2. Output as a ChatPromptValue

ChatPromptValue is a special LangChain object that stores the full prompt (with messages). It's not just text---it keeps track of the message types (system vs. human).

Code (From Original Source)

chat_prompt_value = chat_prompt.format_prompt(
    input_language="English", 
    output_language="French", 
    text="I love programming."
)
print(chat_prompt_value)

Output (From Original Source)

ChatPromptValue(messages=[
    SystemMessage(content='You are a helpful assistant that translates English to French.', additional_kwargs={}), 
    HumanMessage(content='I love programming.', additional_kwargs={})
])

What This Means

• ChatPromptValue has a messages attribute that holds a list of message objects (here: SystemMessage and HumanMessage).
• Use this if you need to work with the prompt as a structured object (not just text) in LangChain workflows.

3. Output as a List of Message Objects

You can convert the ChatPromptValue into a list of SystemMessage and HumanMessage objects. This list is ready to pass directly to Chat models (e.g., ChatOpenAI), since models accept message objects as input.

Code (From Original Source)

message_list = chat_prompt.format_prompt(
    input_language="English", 
    output_language="French", 
    text="I love programming."
).to_messages()
print(message_list)

Output (From Original Source)

[
    SystemMessage(content='You are a helpful assistant that translates English to French.', additional_kwargs={}), 
    HumanMessage(content='I love programming.', additional_kwargs={})
]

What This Means

• The list contains actual LangChain message objects (not just text).
• This is the most useful format for running the prompt with a Chat model ---you can pass message_list directly to the model's predict or generate methods.

Quick Summary of All 3 Formats

Format Type	How to Get It	Use Case
String	chat_prompt.format(...) or format_prompt().to_string()	Quick checks/sharing prompt text
ChatPromptValue	chat_prompt.format_prompt(...)	Working with structured prompt objects
List of Message Objects	format_prompt().to_messages()	Passing input directly to a Chat model

All code, outputs, and logic match the original source---no extra changes or additions.