app.py

import streamlit as st
import urllib.request
import fitz
import re
import numpy as np
import tensorflow_hub as hub
import openai
from sklearn.neighbors import NearestNeighbors
import os
import time
import csv
from io import StringIO
import pandas as pd
from io import BytesIO
import base64
import threading
from queue import Queue
import logging

logging.basicConfig(level=logging.INFO)

def download_pdf(url, output_path):
    urllib.request.urlretrieve(url, output_path)

def preprocess(text):
    text = text.replace('\n', ' ')
    text = re.sub('\s+', ' ', text)
    return text

def pdf_to_text(path, start_page=1, end_page=None):
    doc = fitz.open(path)
    total_pages = doc.page_count

    if end_page is None:
        end_page = total_pages

    text_list = []

    for i in range(start_page-1, end_page):
        text = doc.load_page(i).get_text("text")
        text = preprocess(text)
        text_list.append(text)

    doc.close()
    return text_list

def text_to_chunks(texts, word_length=150, start_page=1):
    text_toks = [t.split(' ') for t in texts]
    page_nums = []
    chunks = []

    for idx, words in enumerate(text_toks):
        for i in range(0, len(words), word_length):
            chunk = words[i:i+word_length]
            if (i+word_length) > len(words) and (len(chunk) < word_length) and (
                len(text_toks) != (idx+1)):
                text_toks[idx+1] = chunk + text_toks[idx+1]
                continue
            chunk = ' '.join(chunk).strip()
            chunk = f'[Page no. {idx+start_page}]' + ' ' + '"' + chunk + '"'
            chunks.append(chunk)
    return chunks

class SemanticSearch:

    def __init__(self):
        self.use = hub.load('https://tfhub.dev/google/universal-sentence-encoder/4')
        self.fitted = False

    def fit(self, data, batch=1000, n_neighbors=5):
        self.data = data
        self.embeddings = self.get_text_embedding(data, batch=batch)
        n_neighbors = min(n_neighbors, len(self.embeddings))
        self.nn = NearestNeighbors(n_neighbors=n_neighbors)
        self.nn.fit(self.embeddings)
        self.fitted = True

    def call(self, text, return_data=True):
        if not self.fitted:
            raise Exception("The fit method must be called before the call method.")
        inp_emb = self.use([text])
        neighbors = self.nn.kneighbors(inp_emb, return_distance=False)[0]

        if return_data:
            return [self.data[i] for i in neighbors]
        else:
            return neighbors

    def get_text_embedding(self, texts, batch=1000):
        embeddings = []
        for i in range(0, len(texts), batch):
            text_batch = texts[i:(i+batch)]
            emb_batch = self.use(text_batch)
            embeddings.append(emb_batch)
        embeddings = np.vstack(embeddings)
        return embeddings

def load_recommender(path, start_page=1):
    global recommender
    texts = pdf_to_text(path, start_page=start_page)
    chunks = text_to_chunks(texts, start_page=start_page)
    recommender.fit(chunks)
    return 'Corpus Loaded.'

def generate_text(openAI_key,prompt, engine="text-davinci-003"):
    openai.api_key = openAI_key
    completions = openai.Completion.create(
    engine=engine,
    prompt=prompt,
    max_tokens=512,
    n=1,
    stop=None,
    temperature=0.7,
    )
    message = completions.choices[0].text
    return message

def generate_answer(question,openAI_key):
    topn_chunks = recommender.call(question)
    if not recommender.fitted:
        st.error('The recommender is not fitted yet.')
        return
    prompt = ""
    prompt += 'search results:\n\n'
    for c in topn_chunks:
        prompt += c + '\n\n'

    prompt += "Instructions: Compose a comprehensive reply to the query using the search results given. "\
              "Cite each reference using [Page Number] notation (every result has this number at the beginning). "\
              "Citation should be done at the end of each sentence. If the search results mention multiple subjects "\
              "with the same name, create separate answers for each. Only include information found in the results and "\
              "don't add any additional information. Make sure the answer is correct, and don't output false content. "\
              "If the text does not relate to the query, simply state 'Found Nothing'. Ignore outlier "\
              "search results that have nothing to do with the question. Only answer what is asked. The "\
              "answer should be short and concise. \n\nQuery: {question}\nAnswer: "

    prompt += f"Query: {question}\nAnswer: "
    answer = generate_text(openAI_key, prompt,"text-davinci-003")

    answer = answer.strip()

    return answer

recommender = SemanticSearch()

st.title('PDF GPT Multi-Line')

description = """ PDF GPT allows you to chat with your PDF file using Universal Sentence Encoder and Open AI. The returned response can cite the page number in square brackets([]) where the information is located, adding credibility to the responses and helping to locate pertinent information quickly."""

st.markdown(description)

openAI_key = st.sidebar.text_input('API Key', value='sk-')

data_section = st.sidebar.text_area("Paste Data:")
paste_data = st.sidebar.button("Paste Data")

add_row = st.sidebar.button("Add row")
row_count = st.session_state.get("row_count", 1)

num_concurrent_calls = st.sidebar.number_input("Concurrent Calls:", min_value=1, max_value=2000, value=10, step=1)
generate_all = st.sidebar.button("Generate All")

reset = st.sidebar.button("Reset")

if reset:
    for i in range(row_count):
        st.session_state[f"url{i}"] = ''
        st.session_state[f"question{i}"] = ''
        st.session_state[f'session_answer{i}'] = ''

    st.session_state.row_count = 1

    st.experimental_rerun()

if add_row:
    row_count += 1
    st.session_state.row_count = row_count

if paste_data:
    data = StringIO(data_section.strip())
    reader = csv.reader(data, delimiter='\t', quotechar='"') # Changed delimiter to '\t'
    urls_questions = [row for row in reader]

    row_count = len(urls_questions)
    st.session_state.row_count = row_count

    for i, url_question in enumerate(urls_questions): # Directly iterate over urls_questions
        if len(url_question) >= 2:
            st.session_state[f"url{i}"] = url_question[0]
            st.session_state[f"question{i}"] = url_question[1]
        else:
            st.error(f"Row {i+1} does not have enough columns.")

for i in range(row_count):
    col1, col2, col3, col4 = st.columns(4)

    with col1:
        url = st.text_input(f'PDF URL {i+1}', key=f'url{i}')
    with col2:
        question = st.text_input(f'Question {i+1}', key=f'question{i}')
    with col3:
        if f'session_answer{i}' not in st.session_state:
            st.session_state[f'session_answer{i}'] = ''
    with col4:
        if st.button(f'Submit {i+1}'):
            if openAI_key.strip()=='':
                st.error('Please enter you Open AI Key')
            elif url.strip() == '':
                st.error('URL field is empty')
            elif question.strip() == '':
                st.error('Question field is empty')
            else:
                glob_url = url
                download_pdf(glob_url, 'corpus.pdf')
                load_recommender('corpus.pdf')

                answer = generate_answer(question,openAI_key)
                st.session_state[f'session_answer{i}'] = answer
    with col3:
        answer_placeholder = st.empty()
        answer_placeholder.text_area(f'Answer {i+1}', key=f'answer{i}', value=st.session_state[f'session_answer{i}'])

def get_table_download_link(df, filename="data.csv", text="Download CSV file"):
    csv = df.to_csv(index=False)
    b64 = base64.b64encode(csv.encode()).decode() 
    href = f'<a href="data:file/csv;base64,{b64}" download="{filename}">{text}</a>'
    return href

data = [[st.session_state.get(f'url{i}', ''), st.session_state.get(f'question{i}', ''), st.session_state.get(f'session_answer{i}', '')] for i in range(row_count)]

df = pd.DataFrame(data, columns=['URL', 'Question', 'Answer'])

st.markdown(get_table_download_link(df), unsafe_allow_html=True)

class WorkerThread(threading.Thread):
    def __init__(self, jobs, results):
        super().__init__()
        self.jobs = jobs
        self.results = results

    def run(self):
        while True:
            job = self.jobs.get()
            if job is None:
                break
            try:
                i, question = job
                result = generate_answer(question, openAI_key)
                self.results.put((i, result))
                logging.info(f"Job {i} completed successfully.")
            except Exception as e:
                self.results.put((i, str(e)))
                logging.error(f"Error on job {i}: {str(e)}")

if generate_all:
    questions = [st.session_state.get(f"question{i}", "") for i in range(row_count)]
    urls = [st.session_state.get(f"url{i}", "") for i in range(row_count)]

    jobs = Queue()
    results = Queue()

    workers = [WorkerThread(jobs, results) for _ in range(num_concurrent_calls)]

    for i, (url, question) in enumerate(zip(urls, questions)):
        download_pdf(url, 'corpus.pdf')
        load_recommender('corpus.pdf')
        jobs.put((i, question))

    for worker in workers:
        worker.start()

    for worker in workers:
        jobs.put(None)

    for worker in workers:
        worker.join()

    logging.info("All worker threads have finished.")

    answers = {}
    while not results.empty():
        i, answer = results.get()
        if isinstance(answer, str) and 'Error' in answer:
            st.error(f"Error on row {i}: {answer}")
        else:
            answers[i] = answer

    logging.info(f"Collected {len(answers)} answers.")

    for i, answer in answers.items():
        st.session_state[f'session_answer{i}'] = answer

    logging.info("Session state updated with answers.")
    
    # Rerun the app after all answers are generated
    st.experimental_rerun()