app.py

import torch, os, argparse, shutil, textwrap, time, streamlit as st 
from langchain.document_loaders import YoutubeLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.vectorstores import Chroma
from langchain.embeddings import OpenAIEmbeddings, HuggingFaceInstructEmbeddings, HuggingFaceBgeEmbeddings
from langchain.chains import RetrievalQA
from langchain.llms import OpenAI 
from langchain.chat_models import ChatOpenAI
from langchain import HuggingFaceHub
from transformers import pipeline 
from deep_translator import GoogleTranslator
from langdetect import detect
from urllib.parse import urlparse, parse_qs

def typewriter(text, speed):
    container = st.empty()
    displayed_text = ''

    for char in text:
        displayed_text += char
        container.markdown(displayed_text)
        time.sleep(1 / speed)

def wrap_text_preserve_newlines(text, width=110):
    lines = text.split('\n')
    wrapped_lines = [textwrap.fill(line, width=width) for line in lines]
    wrapped_text = '\n'.join(wrapped_lines)
    return wrapped_text

def process_llm_response(llm_originalresponse2):
    typewriter(llm_originalresponse2['result'], speed=40)

def extract_video_id(youtube_url):
    try:
        parsed_url = urlparse(youtube_url)
        query_params = parse_qs(parsed_url.query)
        video_id = query_params.get('v', [None])[0]

        return video_id
    except Exception as e:
        print(f"Error extracting video ID: {e}")
        return None 

def chat():
    HF_TOKEN = os.environ.get('HF_TOKEN', False)
    model_name = "BAAI/bge-base-en"
    encode_kwargs = {'normalize_embeddings': True}

    st.title('YouTube ChatBot')

    video_url = st.text_input('Insert video URL', placeholder='Format should be like: https://www.youtube.com/watch?v=pSLeYvld8Mk')
    query = st.text_input("Ask any question about the video")

    if st.button('Submit', type='primary'):
        with st.spinner('Processing the video...'):
            video_id = extract_video_id(video_url)
            loader = YoutubeLoader(video_id)
            documents = loader.load()

            text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=100)
            documents = text_splitter.split_documents(documents)

            vector_db = Chroma.from_documents(
                documents,
                embedding = HuggingFaceBgeEmbeddings(model_name=model_name, model_kwargs={'device': 'cuda' if torch.cuda.is_available() else 'cpu'}, encode_kwargs=encode_kwargs)
            )
            repo_id = "tiiuae/falcon-7b-instruct"
            qa_chain = RetrievalQA.from_chain_type(
                llm=HuggingFaceHub(
                    huggingfacehub_api_token=HF_TOKEN, 
                    repo_id=repo_id, 
                    model_kwargs={'temperature': 0.1, 'max_new_tokens': 1000},
                ),
                retriever=vector_db.as_retriever(),
                return_source_documents=False,
                verbose=False
            )

        with st.spinner('Generating Answer...'):
            llm_response = qa_chain(query)
            process_llm_response(llm_response)

chat()