app.py

# -- Import libraries
from   langchain.prompts           import PromptTemplate
from   PIL                         import Image
from   streamlit.logger            import get_logger
from streamlit_player              import st_player
import pandas                      as pd
import streamlit                   as st
import urllib.request
import argparse
import together
import logging
import requests
import utils
import spacy
import time
import os
import re

def main():
    st.set_page_config(layout="wide")

    # -- 1. Setup arguments
    parser = argparse.ArgumentParser()
    parser.add_argument('--DEFAULT_SYSTEM_PROMPT_LINK', type=str, default="https://raw.githubusercontent.com/AlbertoUAH/Castena/main/prompts/default_system_prompt.txt", help='Valor para DEFAULT_SYSTEM_PROMPT_LINK')
    parser.add_argument('--PODCAST_URL_VIDEO_PATH', type=str, default="https://raw.githubusercontent.com/AlbertoUAH/Castena/main/data/podcast_youtube_video.csv", help='Valor para PODCAST_URL_VIDEO_PATH')
    parser.add_argument('--TRANSCRIPTION', type=str, default='worldcast_roberto_vaquero', help='Name of the trascription')
    parser.add_argument('--MODEL', type=str, default='togethercomputer/llama-2-13b-chat', help='Model name')
    parser.add_argument('--EMB_MODEL', type=str, default='sentence-transformers/paraphrase-multilingual-mpnet-base-v2', help='Embedding model name')
    os.system("python -m spacy download es_core_news_lg")

    # -- 2. Setup env and logger
    os.environ["TOGETHER_API_KEY"] = "6101599d6e33e3bda336b8d007ca22e35a64c72cfd52c2d8197f663389fc50c5"
    logger = get_logger(__name__)

    # -- 3. Setup constants
    B_INST, E_INST = "[INST]", "[/INST]"
    B_SYS, E_SYS   = "<<SYS>>\n", "\n<</SYS>>\n\n"
    args = parser.parse_args()
    PODCAST_URL_VIDEO_PATH     = args.PODCAST_URL_VIDEO_PATH
    DEFAULT_SYSTEM_PROMPT_LINK = args.DEFAULT_SYSTEM_PROMPT_LINK
    TRANSCRIPTION              = args.TRANSCRIPTION
    TRANSCRIPTION_PATH         = '{}_transcription.txt'.format(TRANSCRIPTION)
    MODEL                      = args.MODEL
    EMB_MODEL                  = args.EMB_MODEL

    together.api_key = os.environ["TOGETHER_API_KEY"]
    together.Models.start(MODEL)
    podcast_url_video_df = pd.read_csv(PODCAST_URL_VIDEO_PATH, sep=';')

    r = requests.get("https://raw.githubusercontent.com/AlbertoUAH/Castena/main/media/castena-animated-icon.gif", stream=True)
    icon = Image.open(r.raw)
    icon = icon.resize((100, 100))
    st.sidebar.image(icon)
    video_option = st.sidebar.selectbox(
        "Seleccione el podcast",
        list(podcast_url_video_df['podcast_name_lit'].apply(lambda x: x.replace("'", "")))
    )
    video_option_joined = '_'.join(video_option.replace(': Entrevista a ', ' ').lower().split(' ')).replace("\'", "")
    video_option_joined_path = "{}_transcription.txt".format(video_option_joined)
    youtube_video_url   = list(podcast_url_video_df[podcast_url_video_df['podcast_name'].str.contains(video_option_joined)]['youtube_video_url'])[0].replace("\'", "")

    # -- 4. Setup request for system prompt
    f = urllib.request.urlopen(DEFAULT_SYSTEM_PROMPT_LINK)
    DEFAULT_SYSTEM_PROMPT = str(f.read(), 'UTF-8')

    # -- 5. Setup app
    translator, nlp, retriever = utils.setup_app(video_option_joined_path, EMB_MODEL, MODEL, logger)


    # -- 6. Setup prompt template + llm chain
    instruction = """CONTEXTO:/n/n {context}/n

PREGUNTA: {question}

RESPUESTA: """
    prompt_template = utils.get_prompt(instruction, DEFAULT_SYSTEM_PROMPT, B_SYS, E_SYS, B_INST, E_INST, logger)

    llama_prompt = PromptTemplate(
        template=prompt_template, input_variables=["context", "question"]
    )
    chain_type_kwargs = {"prompt": llama_prompt}

    qa_chain = utils.create_llm_chain(MODEL, retriever, chain_type_kwargs, logger, video_option_joined_path)

    # ---------------------------------------------------------------------
    # -- 7. Setup Streamlit app
    st.title("[Podcast: {}]({})".format(video_option.replace("'", "").title(), youtube_video_url))

    width = 50
    side = (100 - width) / 2
    _, container, _ = st.columns([side, width, side])
    with container:
        st_player(utils.typewrite(youtube_video_url))

    if "messages" not in st.session_state:
        st.session_state.messages = []
    for message in st.session_state.messages:
        with st.chat_message(message["role"]):
            st.markdown(message["content"])
    if prompt := st.chat_input("¡Pregunta lo que quieras!"):
        with st.chat_message("user"):
            st.markdown(prompt)
        st.session_state.messages.append({"role": "user", "content": prompt})
        with st.chat_message("assistant"):
            llm_response = qa_chain(prompt)
            llm_response = utils.process_llm_response(llm_response, nlp)
            st.markdown(llm_response)
            start_time_str_list = []; start_time_seconds_list = []; end_time_seconds_list = []
            for response in llm_response.split('\n'):
                if re.search(r'(\d{2}:\d{2}:\d{2}(.\d{6})?)', response) != None:
                    start_time_str, start_time_seconds, _, end_time_seconds = utils.add_hyperlink_and_convert_to_seconds(response)
                    start_time_str_list.append(start_time_str)
                    start_time_seconds_list.append(start_time_seconds)
                    end_time_seconds_list.append(end_time_seconds)

            if start_time_str_list:
                width = 40
                side = (100 - width) / 2
                for start_time_seconds, start_time_str, end_time_seconds in zip(start_time_seconds_list, start_time_str_list, end_time_seconds_list):
                    st.markdown("__Fragmento: " + start_time_str + "__")
                    _, container, _ = st.columns([side, width, side])
                    with container:
                        st_player(youtube_video_url.replace("?enablejsapi=1", "") + f'?start={start_time_seconds}&end={end_time_seconds}')

        st.session_state.messages.append({"role": "assistant", "content": llm_response})
# -- Sample: streamlit run app.py -- --DEFAULT_SYSTEM_PROMPT_LINK=https://raw.githubusercontent.com/AlbertoUAH/Castena/main/prompts/default_system_prompt.txt --PODCAST_URL_VIDEO_PATH=https://raw.githubusercontent.com/AlbertoUAH/Castena/main/data/podcast_youtube_video.csv --TRANSCRIPTION=worldcast_roberto_vaquero --MODEL=togethercomputer/llama-2-7b-chat --EMB_MODEL=BAAI/bge-base-en-v1.5
if __name__ == '__main__':
    main()