app.py

import streamlit as st
import os
from langchain_community.vectorstores import FAISS
from langchain_community.document_loaders import PyPDFLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_community.embeddings import HuggingFaceEmbeddings 
from langchain_huggingface import HuggingFaceEndpoint  # Updated import
from langchain.chains import ConversationalRetrievalChain
from langchain.memory import ConversationBufferMemory
import tempfile

api_token = os.getenv("HF_TOKEN")
list_llm = ["meta-llama/Meta-Llama-3-8B-Instruct", "mistralai/Mistral-7B-Instruct-v0.2"]
list_llm_simple = [os.path.basename(llm) for llm in list_llm]

def load_doc(uploaded_files):
    try:
        temp_files = []
        for uploaded_file in uploaded_files:
            temp_file = tempfile.NamedTemporaryFile(delete=False, suffix=".pdf")
            temp_file.write(uploaded_file.read())
            temp_file.close()
            temp_files.append(temp_file.name)
        
        loaders = [PyPDFLoader(x) for x in temp_files]
        pages = []
        for loader in loaders:
            pages.extend(loader.load())
        
        text_splitter = RecursiveCharacterTextSplitter(chunk_size=1024, chunk_overlap=64)  
        doc_splits = text_splitter.split_documents(pages)

        for temp_file in temp_files:
            os.remove(temp_file)  # Clean up temporary files

        return doc_splits
    except Exception as e:
        st.error(f"Error loading document: {e}")
        return []

def create_db(splits):
    try:
        embeddings = HuggingFaceEmbeddings()
        vectordb = FAISS.from_documents(splits, embeddings)
        return vectordb
    except Exception as e:
        st.error(f"Error creating vector database: {e}")
        return None

def initialize_llmchain(llm_model, temperature, max_tokens, top_k, vector_db):
    try:
        llm = HuggingFaceEndpoint(
            repo_id=llm_model,
            huggingfacehub_api_token=api_token,
            temperature=temperature,
            max_new_tokens=max_tokens,
            top_k=top_k,
        )
        memory = ConversationBufferMemory(
            memory_key="chat_history",
            output_key='answer',
            return_messages=True
        )

        retriever = vector_db.as_retriever()
        qa_chain = ConversationalRetrievalChain.from_llm(
            llm,
            retriever=retriever,
            chain_type="stuff", 
            memory=memory,
            return_source_documents=True,
            verbose=False,
        )
        return qa_chain
    except Exception as e:
        st.error(f"Error initializing LLM chain: {e}")
        return None

def initialize_database(uploaded_files):
    try:
        doc_splits = load_doc(uploaded_files)
        if not doc_splits:
            return None, "Failed to load documents."
        vector_db = create_db(doc_splits)
        if vector_db is None:
            return None, "Failed to create vector database."
        return vector_db, "Database created!"
    except Exception as e:
        st.error(f"Error initializing database: {e}")
        return None, "Failed to initialize database."

def initialize_LLM(llm_option, llm_temperature, max_tokens, top_k, vector_db):
    try:
        llm_name = list_llm[llm_option]
        qa_chain = initialize_llmchain(llm_name, llm_temperature, max_tokens, top_k, vector_db)
        if qa_chain is None:
            return None, "Failed to initialize QA chain."
        return qa_chain, "QA chain initialized. Chatbot is ready!"
    except Exception as e:
        st.error(f"Error initializing LLM: {e}")
        return None, "Failed to initialize LLM."

def format_chat_history(chat_history):
    formatted_chat_history = []
    for user_message, bot_message in chat_history:
        formatted_chat_history.append(f"User: {user_message}")
        formatted_chat_history.append(f"Assistant: {bot_message}")
    return formatted_chat_history

def conversation(qa_chain, message, history):
    try:
        formatted_chat_history = format_chat_history(history)
        response = qa_chain.invoke({"question": message, "chat_history": formatted_chat_history})
        response_answer = response["answer"]
        if "Helpful Answer:" in response_answer:
            response_answer = response_answer.split("Helpful Answer:")[-1]
        response_sources = response["source_documents"]
        response_source1 = response_sources[0].page_content.strip()
        response_source1_page = response_sources[0].metadata["page"] + 1
        response_source2 = response_sources[1].page_content.strip()
        response_source2_page = response_sources[1].metadata["page"] + 1
        response_source3 = response_sources[2].page_content.strip()
        response_source3_page = response_sources[2].metadata["page"] + 1
        new_history = history + [(message, response_answer)]
        return qa_chain, new_history, response_source1, response_source1_page, response_source2, response_source2_page, response_source3, response_source3_page
    except Exception as e:
        st.error(f"Error in conversation: {e}")
        return qa_chain, history, "", 0, "", 0, "", 0

def main():
    st.title("PDF Chatbot")

    st.markdown("### Step 1 - Upload PDF documents and Initialize RAG pipeline")
    uploaded_files = st.file_uploader("Upload PDF documents", type="pdf", accept_multiple_files=True)

    if uploaded_files:
        if st.button("Create vector database"):
            with st.spinner("Creating vector database..."):
                vector_db, db_message = initialize_database(uploaded_files)
                st.success(db_message)
                st.session_state['vector_db'] = vector_db

    if 'vector_db' not in st.session_state:
        st.session_state['vector_db'] = None

    if 'qa_chain' not in st.session_state:
        st.session_state['qa_chain'] = None

    st.markdown("### Select Large Language Model (LLM) and input parameters")
    llm_option = st.radio("Available LLMs", list_llm_simple)
    temperature = st.slider("Temperature", 0.01, 1.0, 0.5, 0.1)
    max_tokens = st.slider("Max New Tokens", 128, 9192, 4096, 128)
    top_k = st.slider("top-k", 1, 10, 3, 1)

    if st.button("Initialize Question Answering Chatbot"):
        with st.spinner("Initializing QA chatbot..."):
            qa_chain, llm_message = initialize_LLM(list_llm_simple.index(llm_option), temperature, max_tokens, top_k, st.session_state['vector_db'])
            st.session_state['qa_chain'] = qa_chain
            st.success(llm_message)

    st.markdown("### Step 2 - Chat with your Document")
    if st.session_state['qa_chain']:
        history = []
        message = st.text_input("Ask a question")

        if st.button("Submit"):
            with st.spinner("Generating response..."):
                qa_chain, history, response_source1, source1_page, response_source2, source2_page, response_source3, source3_page = conversation(st.session_state['qa_chain'], message, history)
                st.session_state['qa_chain'] = qa_chain

                st.markdown("### Chatbot Response")
                st.text_area("Chatbot Response", value=response_source1, height=100)
                st.text_area("Source 1", value=response_source1, height=100)
                st.text(f"Page: {source1_page}")
                st.text_area("Source 2", value=response_source2, height=100)
                st.text(f"Page: {source2_page}")
                st.text_area("Source 3", value=response_source3, height=100)
                st.text(f"Page: {source3_page}")

if __name__ == "__main__":
    main()