Extracting a structured (pydantic) output from documents

[idan-ben-ami]

Checked other resources

• I added a very descriptive title to this question.
• I searched the LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.

Commit to Help

• I commit to help with one of those options 👆

Example Code

from typing import Optional

from langchain.chains.combine_documents import create_stuff_documents_chain
from langchain.prompts import ChatPromptTemplate
from langchain_core.documents import Document
from langchain_openai import ChatOpenAI
from pydantic import BaseModel, Field


class Person(BaseModel):
    """Information about a person."""

    name: Optional[str] = Field(default=None, description="The name of the person")
    hair_color: Optional[str] = Field(
        default=None, description="The color of the person's hair if known"
    )
    height_in_meters: Optional[str] = Field(
        default=None, description="Height measured in meters"
    )

class People(BaseModel):
    """A collection of people."""

    people: list[Person] = Field(
        default_factory=list, description="A list of people"
    )

llm = ChatOpenAI(model="gpt-4o-mini", temperature=0)
prompt = ChatPromptTemplate.from_messages([
    ("system", "You are an expert model excelling at extracting human information\n{context}"),
])
structured_llm = llm.with_structured_output(schema=People)
chain = create_stuff_documents_chain(structured_llm, prompt)

docs = [Document(metadata={"team": "dev"}, page_content="The dev team corresponds of Bob and Alice. Bob is 5'8 with a black hair. Alice has the same hair color but is 20cm shorter than Bob")]
chain.invoke({"context": docs})

Description

The problem is that when creating the chain with create_stuff_documents_chain it adds an unwanted StrOutputParser (output_parser: Output parser. Defaults to StrOutputParser.), and since I already use the with_structured_output which adds a PydanticOutputParser - the input to the (unwanted) StrOutputParser is a Pydantic instance rather than a string.

I ended up patching the "create_stuff_documents_chain" to ignore output_parser when not supplied (or worse - reimplement it in my code).

What is the right way of using it?

System Info

System Information

OS: Darwin
OS Version: Darwin Kernel Version 24.1.0: Thu Oct 10 21:03:15 PDT 2024; root:xnu-11215.41.3~2/RELEASE_ARM64_T6000
Python Version: 3.12.7 (main, Nov 22 2024, 09:38:06) [Clang 16.0.0 (clang-1600.0.26.4)]

Package Information

langchain_core: 0.3.21
langchain: 0.3.8
langchain_community: 0.3.8
langsmith: 0.1.146
langchain_ollama: 0.2.0
langchain_openai: 0.2.9
langchain_text_splitters: 0.3.2
langgraph_sdk: 0.1.36

[SetonLiang]

chain = prompt | llm.with_structured_output(schema=People)
You can use this chain to invoke.

[idan-ben-ami]

This solution lacks the documents combining and formatting, and relies on the repr of the documents. It also lacks the validation of the "context" input (which is not dynamically on invocation).

[SetonLiang]

Actually I use this method and get Entities list successfully.