import streamlit as st

from model import predictor
from streamlit.scriptrunner import add_script_run_ctx
import audiobot
import chatbot
import os
import threading


def runInThread():
    print('Initialize model in thread')
    st.session_state['predictor'] = predictor.Predictor()
    print('Model is initialized')


def run():
    st.set_page_config(
        page_title='Welcome to Visual Question Answering - Bot',
        page_icon=':robot:',
        layout='wide'
    )

    os.environ['TOKENIZERS_PARALLELISM'] = 'false'
    if 'thread' not in st.session_state:
        st.session_state.thread = threading.Thread(target=runInThread)
        add_script_run_ctx(st.session_state.thread)
        st.session_state.thread.start()

    st.sidebar.title('VQA Bot')
    st.sidebar.write('''
        VQA Bot addresses the challenge of visual question answering with the chat and voice assistance.
        Here, we merged ViLT(Vision-and-Language Transformer) model fine-tuned on VQAv2 into T5-small (Text-to-Text Transfer Transformer). 
        We pretrained and finetuned our model on Language transformer to get the desired result.
        Please use the radio buttons below to navigate.
        ''')

    selected_page = st.sidebar.radio('Go to', ('VQA Chatbot', 'VQA Audiobot'))
    if selected_page == 'VQA Chatbot':
        chatbot.show()
    elif selected_page == 'VQA Audiobot':
        audiobot.show()

    st.caption("Created by Madhuri Sakhare - [Github](https://github.com/msak1612/vqa_chatbot) [Linkedin](https://www.linkedin.com/in/madhuri-sakhare/)")


run()