gpt_with_chunks.py

from dotenv import load_dotenv
import streamlit as st
from PyPDF2 import PdfReader
from langchain.text_splitter import CharacterTextSplitter
from langchain.embeddings.openai import OpenAIEmbeddings
from langchain.vectorstores import FAISS
from langchain.chains.question_answering import load_qa_chain
from langchain.llms import OpenAI

def main():
    load_dotenv()
    st.set_page_config(page_title="Ask your pdf", layout="centered", initial_sidebar_state="auto")
    st.header("Ask your pdf(OpenAI) 🤓")

    # Uploading the file
    pdf = st.file_uploader("Upload your pdf", type="pdf")

    # Extracting the text
    if pdf is not None:
        pdf_reader = PdfReader(pdf)
        text = ""
        for page in pdf_reader.pages:
            text += page.extract_text()

        # Split into chunks
        text_splitter = CharacterTextSplitter(
            separator="\n",  # Defines a new line
            chunk_size=1000,
            chunk_overlap=200,
            length_function=len
        )
        chunks = text_splitter.split_text(text)

        # Create embeddings
        embeddings = OpenAIEmbeddings()

        # Creating an object on which we will be able to search FAISS
        knowledge_base = FAISS.from_texts(chunks, embeddings)

        # show user input
        user_question = st.text_input("Ask a question about the PDF: ")

        # if st.button("Refresh Page"):
        #     st.caching.clear_cache()

        similar_chunks = []  # Initialize outside the if block

        if user_question:
            # 1. Similarity search on the vectorstore
            similar_chunks = knowledge_base.similarity_search(user_question)

        # Continue processing only if user_question is not empty
        if user_question:
            # 2. Retrieve the top k chunks (you can adjust k as needed)
            k = 3
            top_k_chunks = similar_chunks[:k]

            # 3. Submit relevant chunks and the original question to the LLM
            llm = OpenAI()
            chain = load_qa_chain(llm, chain_type="stuff")
            response = chain.run(input_documents=top_k_chunks, question=user_question)

            # 4. Determine 'k' value for each chunk retrieval
            k_values = [len(str(chunk)) for chunk in top_k_chunks]

            # Display results
            st.subheader("Similarity Search Results:")
            st.write("Chunks with similar context/meaning as the question:")
            for i, chunk in enumerate(similar_chunks):
                st.write(f"Chunk {i + 1}:", chunk)

            st.subheader("Top {} Chunks Similar to the Question:".format(k))
            for i, chunk in enumerate(top_k_chunks):
                st.write(f"Chunk {i + 1}:", chunk)

            st.subheader("Answer from LLM:")
            st.write(response)

            st.subheader("Determine 'k' value for each chunk retrieval:")
            for i, k_value in enumerate(k_values):
                st.write(f"Chunk {i + 1}: {k_value}")

if __name__ == '__main__':
    main()