app.py

from langchain_community.vectorstores import FAISS, Qdrant
from langchain_community.document_loaders import PyPDFLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.chains import ConversationalRetrievalChain, RetrievalQA
from langchain_community.embeddings import HuggingFaceEmbeddings 
from langchain.memory import ConversationBufferMemory
from langchain_community.llms import HuggingFaceEndpoint
from langchain.prompts import PromptTemplate
from langchain.retrievers import ContextualCompressionRetriever
from langchain.retrievers.document_compressors import FlashrankRerank
from langchain_community.embeddings.fastembed import FastEmbedEmbeddings
from langchain_community.document_loaders import UnstructuredMarkdownLoader
from llama_parse import LlamaParse
from langchain_groq import ChatGroq
from dotenv import load_dotenv

import os
import streamlit as st
import PyPDF2
import tempfile
import markdown

# Load environment variables from .env file
load_dotenv()

# Environment variables for API keys
GROQ_API_KEY = os.getenv("GROQ_API_KEY")
LLAMA_CLOUD_API_KEY = os.getenv("LLAMA_CLOUD_API_KEY")
HF_API_TOKEN = os.getenv("HF_AUTH_TOKEN")

st.set_page_config(page_title="💬 QA Chatbot")

# 
def read_pdf(uploaded_file):
    pdf_reader = PyPDF2.PdfReader(uploaded_file)
    text = ""
    for page in pdf_reader.pages:
        text += page.extract_text()
    return text

def split_chunks(docs):
    text_splitter = RecursiveCharacterTextSplitter(
        chunk_size=2048,
        chunk_overlap=128,
    )
    return text_splitter.split_text(docs)

def create_db(splits):
    #embeddings_model = HuggingFaceEmbeddings(model_name='sentence-transformers/all-MiniLM-L6-v2')
    embeddings_model = FastEmbedEmbeddings(model_name="BAAI/bge-base-en-v1.5")

    #vectordb = FAISS.from_documents(splits, embeddings_model)
    #st.write(vectordb)

    vectordb = Qdrant.from_documents(
                    docs,
                    embeddings_model,
                    location=":memory:",
                    #path="./db",
                    collection_name="document_embeddings",
                )

    return vectordb

def initialize_llmchain(llm_model, temperature, max_tokens, top_k, vector_db):
    #st.write(selected_model)
    #st.write(max_tokens)
    if selected_model == "Llama-3-70B":    
        llm = ChatGroq(
          model_name=llm_model,
          temperature=temperature, 
          #max_tokens=max_tokens,
          #top_k=top_k
        )
    else:
        llm = HuggingFaceEndpoint(
           repo_id=llm_model,
           huggingfacehub_api_token=HF_API_TOKEN,
           temperature=temperature,
           max_new_tokens=max_tokens,
           top_k=top_k,           
        )     
    
    retriever = vector_db.as_retriever(search_kwargs={"k": 3})

    compressor = FlashrankRerank(model="ms-marco-MiniLM-L-12-v2")

    compression_retriever = ContextualCompressionRetriever(
                            base_compressor=compressor, base_retriever=retriever
                        )

    qachain = RetrievalQA.from_chain_type(
                    llm=llm,
                    chain_type="stuff",
                    retriever=compression_retriever,
                    return_source_documents=True,
                    chain_type_kwargs={"prompt": prompt, "verbose": False},
                )
    
    return qachain

def generate_llm_response(prompt_input):

    #llm_model = ChatGroq(model_name="llama3-70b-8192", temperature=temperature, )

    qa_chain = initialize_llmchain(llm_model, temperature, max_tokens, top_k, st.session_state.vector_db)
    #st.write(qa_chain)

    st.write(prompt_input)

    response = qa_chain({"query": prompt_input, "context": "", "question": prompt_input})
    #st.write(response)

    return response

def clear_chat_history():    
    st.session_state.messages = [{"role": "assistant", "content": "How may I assist you today?"}]
    

instruction = """The provided document is Meta First Quarter 2024 Results.
This form provides detailed financial information about the company's performance for a specific quarter.
It includes unaudited financial statements, management discussion and analysis, and other relevant disclosures required by the SEC.
It contains many tables.
Try to be precise while answering the questions"""

parser = LlamaParse(
    api_key=LLAMA_CLOUD_API_KEY,
    result_type="markdown",
    parsing_instruction=instruction,
    max_timeout=5000,
)

# Store LLM generated responses
if "messages" not in st.session_state.keys():
    st.session_state.messages = [{"role": "assistant", "content": "How may I assist you today?"}]
    
if "vector_db" not in st.session_state:
    st.session_state.vector_db = None

if "uploaded_file" not in st.session_state:
    st.session_state.uploaded_file = None

prompt_template = """
                Use the following pieces of information to answer the user's question.
                If you don't know the answer, just say that you don't know, don't try to make up an answer.

                Context: {context}
                Question: {question}

                Answer the question and provide additional helpful information,
                based on the pieces of information, if applicable. Be succinct.

                Responses should be properly formatted to be easily read.
                """

prompt = PromptTemplate(
    template=prompt_template, input_variables=["context", "question"]
)

# Main app layout
st.title("QA Chatbot with Custom PDF")
st.markdown("---")

# Sidebar for model selection and parameters
with st.sidebar:

    uploaded_file = st.sidebar.file_uploader("Upload a PDF", type="pdf") 

    # Check if a new file has been uploaded or the file has been removed
    if uploaded_file != st.session_state.uploaded_file:
        st.session_state.uploaded_file = uploaded_file
        clear_chat_history()
        st.session_state.vector_db = None

    if uploaded_file is not None and st.session_state.vector_db is None:    
        with st.spinner("Converting to Vectors..."):
            #text = read_pdf(uploaded_file)
            #chunks = split_chunks(docs=text)
            #st.write(len(chunks))
            #st.session_state.vector_db = create_db(chunks)

            #file_data = uploaded_file.getvalue()
            #st.write(type(file_data))

            #temp_dir = tempfile.mkdtemp()
            #documents = LlamaParse(result_type="markdown").load_data(uploaded_file.name)
            #st.write(documents[0].text[:1000])

            temp_dir = tempfile.mkdtemp()
            file_path = os.path.join(temp_dir, uploaded_file.name)
            with open(file_path, "wb") as f:
                    f.write(uploaded_file.getvalue())

            documents = LlamaParse(result_type="markdown").load_data(file_path)

            document_path = os.path.join(temp_dir,"parsed_document.md")
            with open(document_path, "w", encoding="utf-8") as f:  # Ensuring UTF-8 encoding
                f.write(str(documents))

            loader = UnstructuredMarkdownLoader(document_path)
            loaded_documents = loader.load()
          
            text_splitter = RecursiveCharacterTextSplitter(chunk_size=2048, chunk_overlap=128)
            docs = text_splitter.split_documents(loaded_documents)

            #st.write(len(docs))

            st.session_state.vector_db = create_db(loaded_documents)

            #st.sidebar.write("PDF processed and vector database created!")
            st.sidebar.markdown('<p style="color:green;">PDF processed and vector database created!</p>', unsafe_allow_html=True)

    st.sidebar.title("Model Settings")

    #selected_model = st.sidebar.selectbox('Choose a LLM model', ['Llama-3-8B', 'Mistral-7B'], key='selected_model')
    #llm_model = 'meta-llama/Meta-Llama-3-8B-Instruct' if selected_model == 'Llama-3-8B' else 'mistralai/Mistral-7B-Instruct-v0.2'

    selected_model = st.sidebar.selectbox('Choose a LLM model', ['Llama-3-70B', 'Llama-3-8B', 'Mistral-7B'], key='selected_model')

    llm_model = 'meta-llama/Meta-Llama-3-8B-Instruct' if selected_model == 'Llama-3-8B' else 'mistralai/Mistral-7B-Instruct-v0.2' if selected_model == 'Mistral-7B' else 'llama3-70b-8192'
    #st.write(selected_model)

    temperature = st.sidebar.slider('Temperature', 0.0, 1.0, 0.1)
    top_k = st.sidebar.slider('Top_k', 1, 10, 3)
    max_tokens = st.sidebar.slider('Max Tokens', 1, 512, 256)

show_clear_button = len(st.session_state.messages) > 1

if st.session_state.vector_db is not None:
    for message in st.session_state.messages:
        with st.chat_message(message["role"]):
            st.write(message["content"])

    user_input = st.chat_input("Ask a question here")
    if user_input:
        st.session_state.messages.append({"role": "user", "content": user_input})
        with st.chat_message("user"):
            st.write(user_input)

    # Generate a new response if last message is not from assistant
    if st.session_state.messages[-1]["role"] != "assistant":
        with st.chat_message("assistant"):
            with st.spinner("Thinking..."):
                response = generate_llm_response(user_input)
                #st.write(response)
                placeholder = st.empty()
                full_response = ''
                for item in response["result"]:
                    full_response += item
                    #placeholder.markdown(full_response)
                placeholder.markdown(full_response)
        message = {"role": "assistant", "content": full_response}
        #st.write("-------")
        #st.write(message)
        st.session_state.messages.append(message)

        show_clear_button = len(st.session_state.messages) > 1

else:
    st.write("Please upload a PDF file to initialize the database.")

if show_clear_button and st.button('Clear Chat History'):
        clear_chat_history()
        st.rerun()  # This line ensures the page reruns to reflect the changes