Parsing Model Output Demo 2

In this guide, you’ll learn how to generate model output with LangChain and transform plain-text responses into structured JSON using the JsonOutputParser. This approach removes manual parsing, ensures valid JSON, and boosts developer productivity.

Setup

Begin by installing and importing the necessary packages, then initialize your language model:

pip install langchain langchain-openai

from langchain.prompts import PromptTemplate
from langchain_openai import OpenAI
from langchain_core.output_parsers import JsonOutputParser
import json

llm = OpenAI()

1. Basic Prompt without Parsing

First, create a simple prompt template to list three countries and their capitals:

prompt = PromptTemplate(
    template="List 3 countries in {continent} and their capitals",
    input_variables=["continent"]
)

raw_output = llm.invoke(input=prompt.format(continent="Asia"))
print(raw_output)

Example response:

1. Japan – Tokyo
2. China – Beijing
3. India – New Delhi

While human-readable, this plain-text format is hard to consume programmatically.

2. Defining the JSON Output Parser

Instantiate LangChain’s JSON parser and get its format instructions so the LLM knows to return valid JSON:

output_parser = JsonOutputParser()
format_instructions = output_parser.get_format_instructions()
print(format_instructions)

This prints:

Return a JSON object.

3. Prompt Template with Format Instructions

Embed the format instructions into your template to enforce JSON output:

prompt = PromptTemplate(
    template=(
      "List 3 countries in {continent} and their capitals\n"
      "{format_instructions}"
    ),
    input_variables=["continent"],
    partial_variables={"format_instructions": format_instructions}
)

print(prompt.format(continent="North America"))

Resulting prompt:

List 3 countries in North America and their capitals
Return a JSON object.

4. Invoking the Model and Parsing

Invoke the LLM with the enhanced prompt and parse the JSON response:

response = llm.invoke(input=prompt.format(continent="North America"))
print(response)
# Example LLM response (JSON):
# {
#   "USA": "Washington D.C.",
#   "Canada": "Ottawa",
#   "Mexico": "Mexico City"
countries = output_parser.parse(response)
print(type(countries))        # <class 'dict'>
print(json.dumps(countries))  # {"USA":"Washington D.C.","Canada":"Ottawa","Mexico":"Mexico City"}

5. Comparison of LangChain Output Parsers

Conclusion

By incorporating the JsonOutputParser and embedding format instructions in your LangChain prompts, you can automatically obtain structured JSON from your LLM calls. This streamlines data handling in Python and eliminates fragile text parsing.

Links and References

• LangChain Documentation
• OpenAI API Reference
• JSON Specification