Facing ValidationError: 1 validation error for StuffDocumentsChain

[anand-it-is]

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

import os
import pickle
import boto3
import streamlit as st
from PyPDF2 import PdfReader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.vectorstores import FAISS
from langchain.chains.question_answering import load_qa_chain
from langchain.embeddings import BedrockEmbeddings
from langchain.llms import Bedrock
from langchain.prompts import PromptTemplate

# Set AWS credentials and region
my_access_id = os.getenv('AK_ACCESS_ID')
my_secret_key = os.getenv('AK_SECRET')
region_name = 'us-east-1'

# Create a client to interact with the Bedrock service using AWS credentials
bedrock = boto3.client(
    service_name="bedrock-runtime",
    region_name=region_name,
    aws_access_key_id=my_access_id,
    aws_secret_access_key=my_secret_key
)

# Initialize BedrockEmbeddings with a specific model and the Bedrock client
bedrock_embeddings = BedrockEmbeddings(model_id="amazon.titan-embed-text-v1", client=bedrock)

def main():
    st.header("Chat with PDFs")
    
    # Allow the user to upload multiple PDF files
    pdfs = st.file_uploader("Upload PDF files", type="pdf", accept_multiple_files=True)
    
    if pdfs:
        all_chunks = []
        
        for pdf in pdfs:
            st.write(f"File {pdf.name[:-4]} uploaded successfully.")
            
            # Read the uploaded PDF file
            pdf_reader = PdfReader(pdf)
            text = ""
            for page in pdf_reader.pages:
                text += page.extract_text()
            
            # Split the extracted text into chunks
            text_splitter = RecursiveCharacterTextSplitter(
                chunk_size=1000,
                chunk_overlap=200,
                length_function=len
            )
            chunks = text_splitter.split_text(text=text)
            all_chunks.extend(chunks)

            # Create a name for storing the chunks based on the PDF file name
            store_name = pdf.name[:-4]

            # Check if chunks are already saved on disk
            if os.path.exists(f"{store_name}_chunks.pkl"):
                try:
                    # Load the chunks from disk
                    with open(f"{store_name}_chunks.pkl", "rb") as file:
                        chunks = pickle.load(file)
                        st.write(f"Text chunks of file {store_name} loaded from the Disk")
                except EOFError as e:
                    st.error(f"Error loading text chunks: {e}")
                    os.remove(f"{store_name}_chunks.pkl")
            else:
                # Save the chunks to disk
                with open(f"{store_name}_chunks.pkl", "wb") as file:
                    pickle.dump(chunks, file)

        # Use the Bedrock embeddings for the text chunks
        vector_store = FAISS.from_texts(all_chunks, embedding=bedrock_embeddings)

        # Accept a query from the user
        query = st.text_input("Ask questions about your PDF files")
        
        if query:
            # Search for similar text chunks based on the query
            docs = vector_store.similarity_search(query=query, k=3)

            # Reinitialize the Bedrock client for language model
            llm = Bedrock(
                model_id="anthropic.claude-v2:1",
                client=bedrock,
                model_kwargs={"temperature": 0.1, "top_p":0.9, "max_tokens":2048, "top_k":250}
            )

            # Define the prompt template with additional instructions
            prompt_template = """
            You are a helpful assistant. Answer the question based on the provided documents.
            - If you're instructed to answer the questions in 2 lines. Answer the question in 2 lines only.
            - If the answer is not in the provided documents, say "I don't know".
            - Use the provided documents to answer the question in detail.
            Question: {query}
            Context: {input_documents}
            Answer:
            """

            # Load the QA chain with the prompt template
            prompt = PromptTemplate(
                input_variables=["query", "input_documents"], 
                template=prompt_template
            )
            chain = load_qa_chain(
                llm=llm, 
                chain_type="stuff", 
                prompt=prompt
            )

            # Prepare inputs for the chain
            input_documents = " ".join([doc.page_content for doc in docs])  # Concatenate document content into a single string
            inputs = {
                "query": query,  # The user's question
                "input_documents": input_documents  # The concatenated text from relevant documents
            }

            # Get the response from the language model based on the query and documents
            response = chain.run(inputs)
            st.write(response)

if __name__ == '__main__':
    main()

Description

I'm trying to implement Prompt Engineering using Langchain. I'm getting this error "ValidationError: 1 validation error for StuffDocumentsChain root document_variable_name context was not found in llm_chain input_variables: ['input_documents', 'query'] (type=value_error)
Traceback:

I'm not able to figure out the ways to sort this out.

"

System Info

After running python -m langchain_core.sys_info

Let me know if you need any other info.

I'm getting something like this in screenshot below

[dosubot[bot]]

Hello, @anand-it-is! I'm here to assist you with any bugs, questions, or contributions you might have. Feel free to ask for help while waiting for a human maintainer.

To resolve the ValidationError indicating that the context was not found in llm_chain input variables when running the StuffDocumentsChain, you need to ensure that the document_variable_name in your StuffDocumentsChain matches the input variable expected by your PromptTemplate.

In your case, the PromptTemplate expects input_documents and query, but the StuffDocumentsChain is looking for context. You need to align these variable names.

Here's how you can modify your code:

1. Change the document_variable_name in the StuffDocumentsChain to input_documents.
2. Ensure that the PromptTemplate and the StuffDocumentsChain are using the same variable names.

Here's the updated code snippet:

from langchain.chains import LLMChain, StuffDocumentsChain
from langchain.prompts import PromptTemplate

# Define the prompt template with additional instructions
prompt_template = """
You are a helpful assistant. Answer the question based on the provided documents.
- If you're instructed to answer the questions in 2 lines. Answer the question in 2 lines only.
- If the answer is not in the provided documents, say "I don't know".
- Use the provided documents to answer the question in detail.
Question: {query}
Context: {input_documents}
Answer:
"""

# Load the QA chain with the prompt template
prompt = PromptTemplate(
    input_variables=["query", "input_documents"], 
    template=prompt_template
)

# Initialize the LLMChain
llm_chain = LLMChain(llm=llm, prompt=prompt)

# Initialize the StuffDocumentsChain with the correct document_variable_name
chain = StuffDocumentsChain(
    llm_chain=llm_chain,
    document_prompt=prompt,
    document_variable_name="input_documents"
)

# Prepare inputs for the chain
input_documents = " ".join([doc.page_content for doc in docs])  # Concatenate document content into a single string
inputs = {
    "query": query,  # The user's question
    "input_documents": input_documents  # The concatenated text from relevant documents
}

# Get the response from the language model based on the query and documents
response = chain.run(inputs)
st.write(response)

This ensures that the document_variable_name in the StuffDocumentsChain matches the input_documents variable expected by the PromptTemplate, resolving the ValidationError ^[1].

To continue talking to Dosu, mention @dosu.

[anand-it-is]

@dosu after applying the suggested changes, i'm getting "AttributeError: 'str' object has no attribute 'page_content'".

[dosubot[bot]]

To resolve the AttributeError: 'str' object has no attribute 'page_content', ensure that you are working with Document objects and not plain strings. The Document class has the page_content attribute, while a string does not.

Here's how you can modify your code to convert the text chunks into Document objects before using them:

1. Convert the text chunks into Document objects.
2. Use these Document objects in your vector store and chain.

Here's the updated code snippet:

from langchain_core.documents import Document

# Convert chunks to Document objects
docs = [Document(page_content=chunk) for chunk in all_chunks]

# Use the Bedrock embeddings for the text chunks
vector_store = FAISS.from_documents(docs, embedding=bedrock_embeddings)

# Accept a query from the user
query = st.text_input("Ask questions about your PDF files")

if query:
    # Search for similar text chunks based on the query
    docs = vector_store.similarity_search(query=query, k=3)

    # Reinitialize the Bedrock client for language model
    llm = Bedrock(
        model_id="anthropic.claude-v2:1",
        client=bedrock,
        model_kwargs={"temperature": 0.1, "top_p": 0.9, "max_tokens": 2048, "top_k": 250}
    )

    # Define the prompt template with additional instructions
    prompt_template = """
    You are a helpful assistant. Answer the question based on the provided documents.
    - If you're instructed to answer the questions in 2 lines. Answer the question in 2 lines only.
    - If the answer is not in the provided documents, say "I don't know".
    - Use the provided documents to answer the question in detail.
    Question: {query}
    Context: {input_documents}
    Answer:
    """

    # Load the QA chain with the prompt template
    prompt = PromptTemplate(
        input_variables=["query", "input_documents"],
        template=prompt_template
    )
    llm_chain = LLMChain(llm=llm, prompt=prompt)

    # Initialize the StuffDocumentsChain with the correct document_variable_name
    chain = StuffDocumentsChain(
        llm_chain=llm_chain,
        document_prompt=prompt,
        document_variable_name="input_documents"
    )

    # Prepare inputs for the chain
    input_documents = " ".join([doc.page_content for doc in docs])  # Concatenate document content into a single string
    inputs = {
        "query": query,  # The user's question
        "input_documents": input_documents  # The concatenated text from relevant documents
    }

    # Get the response from the language model based on the query and documents
    response = chain.run(inputs)
    st.write(response)

This ensures that the chunks are converted into Document objects, which have the page_content attribute, avoiding the AttributeError ^[1][2].

To continue talking to Dosu, mention @dosu.