app.py

import streamlit as st  #Web App
import os
from PIL import Image
from utils import *
import asyncio

import pickle
docs = None
api_key = ' '

st.set_page_config(layout="wide")

image = Image.open('arxiv_decode.png')
st.image(image, width=1000)

#title
st.title("Answering questions from scientific papers")
st.markdown("##### This tool will allow you to ask questions and get answers based on scientific papers. It uses OpenAI's GPT models, and you must have your own API key. Each query is about 10k tokens, which costs about only $0.20 on your own API key, charged by OpenAI.")
st.markdown("##### Current version searches on [ArXiv](https://arxiv.org) papers only. 🚧Under development🚧")
st.markdown("Used libraries:\n * [PaperQA](https://github.com/whitead/paper-qa) \n* [langchain](https://github.com/hwchase17/langchain)")

api_key_url = 'https://help.openai.com/en/articles/4936850-where-do-i-find-my-secret-api-key'

api_key = st.text_input('OpenAI API Key',
         placeholder='sk-...', 
         help=f"['What is that?']({api_key_url})",
         type="password",
         value = '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')     

os.environ["OPENAI_API_KEY"] = f"{api_key}" # 
if len(api_key) != 51:
    st.warning('Please enter a valid OpenAI API key.', icon="⚠️")


max_results_current = 5
max_results = max_results_current
def search_click_callback(search_query, max_results, XRxiv_servers=[]):
    global pdf_info, pdf_citation
    search_engines = XRxivQuery(search_query, max_results, XRxiv_servers=XRxiv_servers)
    pdf_info = search_engines.call_API()
    search_engines.download_pdf()

    return pdf_info

with st.form(key='columns_in_form', clear_on_submit = False):
    c1, c2, c3 = st.columns([5, 0.8, 4])
    with c1:
        search_query = st.text_input("Input search query here:", placeholder='Keywords for most relevant search...', value='CFD Modeling'
                                       )#search_query, max_results_current))

    with c2:
        max_results = st.number_input("Max papers", value=max_results_current)
        max_results_current = max_results_current
    with c3:
        with st.container():
            st.markdown('Pre-print server')
            checks = st.columns(4)
            with checks[0]:
                ArXiv_check = st.checkbox('ArXiv')
            with checks[1]:
                ChemArXiv_check = st.checkbox('ChemArXiv')
            with checks[2]:
                BioArXiv_check = st.checkbox('BioArXiv')
            with checks[3]:
                MedrXiv_check = st.checkbox('MedrXiv')

    searchButton = st.form_submit_button(label = 'Search')

if searchButton:
    # checking which pre-print servers selected
    XRxiv_servers = []
    if ArXiv_check:
        XRxiv_servers.append('rxiv')
    if ChemArXiv_check:
        XRxiv_servers.append('chemrxiv')
    if BioArXiv_check:
        XRxiv_servers.append('biorxiv')
    if MedrXiv_check:
        XRxiv_servers.append('medrxiv')
    global pdf_info
    pdf_info =  search_click_callback(search_query, max_results, XRxiv_servers=XRxiv_servers)
    if 'pdf_info' not in st.session_state:
        st.session_state.key = 'pdf_info'
    st.session_state['pdf_info'] = pdf_info


def answer_callback(question_query, word_count):
    import paperqa
    global docs
    if docs is None:
        pdf_info = st.session_state['pdf_info']
        docs = paperqa.Docs()
        pdf_paths = [f"{p[4]}/{p[0].replace(':','').replace('/','').replace('.','')}.pdf" for p in pdf_info]
        pdf_citations = [p[5] for p in pdf_info]
        print(list(zip(pdf_paths, pdf_citations)))
        for d, c in zip(pdf_paths, pdf_citations):
            docs.add(d, c)
    docs._build_faiss_index()
    answer = docs.query(question_query,  length_prompt=f'use {word_count:d} words')
    st.success('Voila! 😃')
    return answer.formatted_answer

with st.form(key='question_form', clear_on_submit = False):
    c1, c2 = st.columns([6, 2])
    with c1:
        question_query = st.text_input("What do you wanna know from these papers?", placeholder='Input questions here...',
                value='Write about CFD modeling')
        # search_query = st.text_input("Input search query here:", placeholder='Keywords for most relevant search...', value='CFD Modeling'
        #                                )#search_query, max_results_current))

    with c2:
        word_count = st.slider("Suggested number of words in your answer?", 30, 300, 100)
    submitButton = st.form_submit_button('Submit')

if submitButton:
    with st.expander("Found papers:", expanded=True):
        st.write(f"{st.session_state['all_reference_text']}")
    with st.spinner('⏳ Please wait...'):
        start = time.time()
        final_answer = answer_callback(question_query, word_count)
        length_answer = len(final_answer)
        st.text_area("Answer:", final_answer, height=max(length_answer//4, 100))
        end = time.time()
        clock_time = end - start
        with st.empty():
            st.write(f"✔️ Task completed in {clock_time:.2f} seconds.")