Skip to main content

The Stuff Documents Chain

The stuff documents chain is the most straightforward chain for combining multiple documents. It simply inserts all the input documents into a prompt and passes it to the LLM.

Overview​

The stuff chain is well-suited for small documents, generally less than 1000 tokens each, with just a few documents per call. Larger or more documents may exceed the context length limits of most LLMs.

It takes a list of input documents, formats them into a prompt using a template, and passes the prompt to the LLM to generate a response.

Limitations​

  • Input documents should be small, under 1000 tokens each
  • Only a few documents should be passed per call, to avoid exceeding context limits

Usage​

Basic usage is simple:

from langchain import StuffDocumentsChain

chain = StuffDocumentsChain(llm)

docs = [doc1, doc2, doc3]

summary = chain.run(docs)

Customizing the Prompt​

You can customize the prompt template used to format the documents:

prompt_template = """Summarize these documents:

{documents}

Summary:"""

chain = StuffDocumentsChain(llm, prompt=PromptTemplate(template=prompt_template))

The {documents} variable will be populated with the input documents. This allows flexibility in prompt formatting.

Conclusion​

The stuff documents chain is a simple way to combine multiple small documents into a single prompt. It has limitations around document size and number. Prompts can be customized. Overall it provides a straightforward option for basic document combination.