app.py

import streamlit as st
import os
api_token = os.environ.get("Key2")

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_community.llms import HuggingFaceEndpoint
from langchain.chains import ConversationalRetrievalChain
from langchain.memory import ConversationBufferMemory

list_llm = ["meta-llama/Llama-3.2-3B-Instruct", "mistralai/Mistral-7B-Instruct-v0.2"]  

# Load and split PDF document
def load_doc(list_file_path):
    loaders = [PyPDFLoader(x) for x in list_file_path]
    pages = []
    for loader in loaders:
        pages.extend(loader.load())
    text_splitter = RecursiveCharacterTextSplitter(chunk_size=1024, chunk_overlap=64)  
    return text_splitter.split_documents(pages)

# Create vector database
def create_db(splits):
    embeddings = HuggingFaceEmbeddings()
    return FAISS.from_documents(splits, embeddings)

# Initialize langchain LLM chain
def initialize_llmchain(llm_model, temperature, max_tokens, top_k, vector_db):
    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()

    return ConversationalRetrievalChain.from_llm(
        llm,
        retriever=retriever,
        chain_type="stuff", 
        memory=memory,
        return_source_documents=True,
        verbose=False,
    )

st.title("RAG PDF Chatbot")

uploaded_files = st.file_uploader("Upload PDF files", accept_multiple_files=True, type="pdf")

if uploaded_files:
    # Save uploaded files to local disk
    file_paths = []
    for uploaded_file in uploaded_files:
        file_path = os.path.join("temp", uploaded_file.name)
        os.makedirs("temp", exist_ok=True)
        with open(file_path, "wb") as f:
            f.write(uploaded_file.getbuffer())
        file_paths.append(file_path)

    st.session_state["doc_splits"] = load_doc(file_paths)
    st.success("Documents successfully loaded and split!")

if 'vector_db' not in st.session_state and 'doc_splits' in st.session_state:
    st.session_state['vector_db'] = create_db(st.session_state['doc_splits'])

llm_option = st.selectbox("Select LLM", list_llm)

temperature = st.slider("Temperature", 0.01, 1.0, 0.5, 0.1)
max_tokens = st.slider("Max Tokens", 128, 9192, 4096, 128)
top_k = st.slider("Top K", 1, 10, 3, 1)

if 'qa_chain' not in st.session_state and 'vector_db' in st.session_state:
    st.session_state['qa_chain'] = initialize_llmchain(llm_option, temperature, max_tokens, top_k, st.session_state['vector_db'])

if "chat_history" not in st.session_state:
    st.session_state["chat_history"] = []

user_input = st.text_input("Ask a question")

if st.button("Submit") and user_input:
    qa_chain = st.session_state['qa_chain']
    response = qa_chain.invoke({"question": user_input, "chat_history": st.session_state["chat_history"]})

    st.session_state["chat_history"].append((user_input, response["answer"]))

    st.write("### Response:")
    st.write(response["answer"])

    st.write("### Sources:")
    for doc in response["source_documents"][:3]:
        st.write(f"Page {doc.metadata['page'] + 1}: {doc.page_content[:300]}...")

st.write("### Chat History")
for user_msg, bot_msg in st.session_state["chat_history"]:
    st.text(f"User: {user_msg}")
    st.text(f"Assistant: {bot_msg}")