app.py

import os
os.environ['KMP_DUPLICATE_LIB_OK']='True'
import streamlit as st
#from streamlit_chat import message

from streamlit.runtime.scriptrunner.script_run_context import get_script_run_ctx
from datetime import datetime
import pandas as pd
from PyPDF2 import PdfReader
from langchain.text_splitter import CharacterTextSplitter,RecursiveCharacterTextSplitter

# from services.qa_chat_mode import *
# from services.qna import *
# from services.utils import *
# from services.transcripts import *
# from langchain import HuggingFaceHub
# from services.smart_prompt import PDSCoverageChain

@st.cache_data
def get_text_chunks(pdf_docs,chunk_size=2000,overlap=0):
    documents = ""
    for pdf in pdf_docs:
        pdf_reader = PdfReader(pdf)
        for page in pdf_reader.pages:
            documents += page.extract_text()
        
    # text_splitter = CharacterTextSplitter(separator="\n",
    #     chunk_size=chunk_size,
    #     chunk_overlap=overlap,
    #     length_function=len)
    
    text_splitter = RecursiveCharacterTextSplitter(
    # Set a really small chunk size, just to show.
    chunk_size = chunk_size,
    chunk_overlap  = overlap)

    texts = text_splitter.split_text(documents)
    return texts

def build_experimental_ui():
 
    with st.sidebar:
        tabs = st.sidebar.selectbox('SELECT TASK', [
            "Question & Answer",
            "Question & Answer (Chat Mode)",
            "Transcript Intelligence",
        ])
        st.markdown('---')
        # Filters
        if tabs=='Question & Answer':
            selected_model = st.selectbox("Select Model:", options=[], index=0)
            selected_embeddings = st.selectbox("Select Embeddings:", options=[], index=0)
    #        strategy = st.selectbox("Select Strategy:", options=STRATEGY_OPTIONS, index=1, disabled=True)
            strategy = ''
        elif tabs=='Question & Answer (Chat Mode)':
            selected_model = st.selectbox("Select Model:", options=[], index=0)
            selected_embeddings = st.selectbox("Select Embeddings:", options=[], index=0)
        elif tabs == 'Transcript Intelligence':
            selected_model = st.selectbox("Select Model:", options=[], index=0)
    #        selected_embeddings = st.selectbox("Select Embeddings:", options=EMBEDDING_OPTIONS, index=0)
    #        claimnumber = st.selectbox("Select Claim:", options=CLAIM_OPTIONS, index=0, disabled=True)

        pdf_docs = st.file_uploader('Upload a PDF file', type=['pdf'],accept_multiple_files=True)
        st.session_state['pdf_file'] = pdf_docs
        Process = st.button("Process", disabled=(pdf_docs==[]))
        
        if Process: 
            if pdf_docs:
                if pdf_docs!=st.session_state['pdf_file']:
                    st.session_state['pdf_file'] = pdf_docs
                with st.spinner('Creating embeddings...'):
                    texts = get_text_chunks(pdf_docs=pdf_docs)
                    # st.write([len(x) for x in texts])
                    # retriever = get_retriever_from_text(texts, embeddings[selected_embeddings])
                    # st.session_state['retriever'] = retriever

    st.error("Disclaimer: All data processed in this application will be sent to OpenAI API based in the United States.")

    st.markdown('## '+tabs)

    if tabs=='Question & Answer':
          
        st.markdown('---')
        
        # # Question & Answer
        # if st.session_state['retriever'] is None:
        #     disable_query = True
        # else:
        #     disable_query = False
            
        prompt = st.text_input('Input your prompt', disabled=False, key="text")

        questions_file =  st.file_uploader('Upload a CSV file with questions', type=['csv'],accept_multiple_files=False)

        if questions_file:
            questions_df = pd.read_csv(questions_file)
        
        # if strategy=='Without Chain-of-Thought':
        #     instruction = st.text_area('Input your instruction (optional)', value=st.session_state['qa_instruction'], disabled=disable_query)
        #     with st.expander("Sample instruction"):
        #         sample_instruction = "Answer the question based on the context provided. Explain with reason in bullet points. Let's think step by step."
        #         button_sample_instruction = st.button(sample_instruction, key='instruction1', disabled=disable_query, on_click=set_qa_instruction, args=(sample_instruction,))
            
        # if st.session_state.query is None:
        #     disable = True
        # else:
        #     disable = False
            
        button_query = st.button('Submit', disabled=False)
        
        if button_query:

            for question in  questions_df['question']:
                instruction = f'{prompt}.Question:{question}'

            return
            
    
    else:
        st.info("Under Development")
        
build_experimental_ui()