Skip to main content

List Parser

The ListParser is an output parser in LangChain that parses language model responses into Python lists of comma-separated values. It implements the core parse and get_format_instructions methods required of LangChain output parsers.

Overview​

The ListParser provides a simple way to get structured lists back from language model responses. It is commonly used when you want the language model to return a list of items based on some prompt.

Some examples of when the ListParser is useful:

  • Getting a list of movie recommendations
  • Getting a list of ingredients for a recipe
  • Getting a list of pros/cons for a product

Usage​

Here is a typical workflow for using the ListParser:

  1. Import the ListParser and create an instance:
from langchain.output_parsers import CommaSeparatedListOutputParser

parser = CommaSeparatedListOutputParser()
  1. Get the format instructions from the parser to include in your prompt:
format_instructions = parser.get_format_instructions()
  1. Include the format instructions in your prompt template:
from langchain.prompts import PromptTemplate

prompt = PromptTemplate(
  template="List 3 ice cream flavors: {format_instructions}",
  partial_variables={"format_instructions": format_instructions}
)
  1. Send the prompt to your language model and parse the response:
from langchain.llms import OpenAI

llm = OpenAI()  
response = llm.predict(prompt.to_string())

parsed = parser.parse(response)
# ['Vanilla', 'Chocolate', 'Strawberry']

The ListParser works with any LangChain compatible language model, including OpenAI for standard models and ChatOpenAI for chat models.

Handling Invalid Responses​

If the language model returns a response that is not a valid comma-separated list, the ListParser will fail to parse it.

To handle invalid responses, the OutputFixingParser can be used. It takes the invalid response and format instructions and asks the language model to fix the formatting:

from langchain.output_parsers import OutputFixingParser

fixing_parser = OutputFixingParser.from_llm(parser=parser, llm=ChatOpenAI())

invalid_response = "Vanilla, Chocolate, Strawberry"

fixed = fixing_parser.parse(invalid_response) 
# ['Vanilla', 'Chocolate', 'Strawberry']

Summary​

The ListParser provides an easy way to get Python lists back from language models. It can be used directly or with the OutputFixingParser to handle invalid responses. The parser works seamlessly with prompt templates to inject formatting instructions.