app_old_1.py

import urllib.request
import fitz
import re
import numpy as np
#import tensorflow_hub as hub
import openai
import gradio as gr
import os
from sklearn.neighbors import NearestNeighbors
#from pathlib import Path
#from dotenv import load_dotenv

# dotenv_path = Path('azure_key.env')
# load_dotenv(dotenv_path=dotenv_path)

openai.api_type = "azure"
openai.api_version = "azure"
openai.api_base = "azure"  # Your Azure OpenAI resource's endpoint value.
openai.api_key = "azure"

print(openai.api_key)



def download_pdf(url, output_path):
    try:
        urllib.request.urlretrieve(url, output_path)
    except:
        print('error in download file')


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.use = hub.load(r'C:\Users\u393845\wns\GenAI\universal-sentence-encoder_4')
        #self.use = hub.load('/home/wnsuser/ESG/Django/Dev/Resources/USE')
        self.use = None
        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):
        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, model="gpt-3.5-turbo"):
    openai.api_key = openAI_key
    temperature = 0.7
    max_tokens = 256
    top_p = 1
    frequency_penalty = 0
    presence_penalty = 0

    if model == "text-davinci-003":
        completions = openai.Completion.create(
            engine=model,
            prompt=prompt,
            max_tokens=max_tokens,
            n=1,
            stop=None,
            temperature=temperature,
        )
        message = completions.choices[0].text
    else:
        message = openai.ChatCompletion.create(
            model=model,
            messages=[
                {"role": "system", "content": "You are a helpful assistant."},
                {"role": "assistant", "content": "Here is some initial assistant message."},
                {"role": "user", "content": prompt}
            ],
            temperature=.3,
            max_tokens=max_tokens,
            top_p=top_p,
            frequency_penalty=frequency_penalty,
            presence_penalty=presence_penalty,
        ).choices[0].message['content']
    return message


def generate_answer(question, openAI_key, model):
    topn_chunks = recommender(question)
    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. " \
              "Only answer what is asked. The answer should be short and concise. \n\nQuery: "

    prompt += f"{question}\nAnswer:"
    answer = generate_text(openAI_key, prompt, model)
    return answer


def question_answer(chat_history, url, file, question, model):
    openAI_key = openai.api_key

    qna_dic = {"What is the total scope 1 GHG emission?":"The total scope 1 GHG emission in FY 2022-23 is 2,942 tons of CO2e [Page no. 116].",
               "What is the total scope 2 GHG emission?":"The total scope 2 GHG emission in FY 2022-23 is 19,586 tons of CO2e [Page no. 116].",
               "What is the total scope 3 GHG emission?":"WNS measures and tracks its direct (Scope 1) and indirect (Scope 2) GHG emissions in context with its energy consumption. The methodology used for calculating GHG emissions is aligned with the globally accepted GHG protocol standards developed by the World Resources Institute. Additionally, WNS has joined the “Race to Zero” campaign backed by the United Nations and committed to attaining science-based net-zero targets. WNS is leveraging a high-quality technical tool to monitor and measure data across multiple locations and gather insights for tracking and improving its performance. The total Scope 3 GHG emission is not reported, however, WNS may evaluate this disclosure requirement in the near future. [Page no. 116, 129]",
               "What are the main results of the study?":"The main results of the study include the formulation of a policy to ensure a working environment free of discrimination or harassment and where all employees are treated with dignity and respect [Page no. 69], an extensive internal review of ESG topics previously identified in a detailed ESG materiality survey conducted by ESG advisors from Nasdaq Corporate Solutions [Page no. 16], a 12-week-long fellowship program that focuses on youth leadership [Page no. 21], and a focus on improving the diverse representation of the workforce, linking a portion of executive compensation to diversity targets, and recognizing and reinforcing inclusive behavior in the organization [Page no. 33].",
               "What are the main contributions of this study?":"This study focuses on improving the diverse representation of the workforce, linking a portion of executive compensation to diversity targets, recognizing and reinforcing inclusive behavior in the organization, creating several different categories of awards, and focusing on gender advancement, enhancing inclusivity and mental wellness. [Page no. 33] It also works toward improving gender representation in the organization across levels, runs multiple recruitment initiatives, and focuses on encouraging reading among schoolchildren through the management of 17 community libraries and 176 school libraries in India and one school library in China. [Page no. 19] Additionally, WNS has signed a letter of commitment with the Science Based Targets initiative in December 2022, switched to green power in 14 offices in India and Costa Rica, and spent $1,603,967 on community outreach. [Page no. 7]"}


    try:
        if openAI_key.strip() == '':
            return '[ERROR]: Please enter your Open AI Key. Get your key here : https://platform.openai.com/account/api-keys'
        if url.strip() == '' and file is None:
            return '[ERROR]: Both URL and PDF is empty. Provide at least one.'
        if url.strip() != '' and file is not None:
            return '[ERROR]: Both URL and PDF is provided. Please provide only one (either URL or PDF).'
        if model is None or model == '':
            return '[ERROR]: You have not selected any model. Please choose an LLM model.'
        if url.strip() != '':
            glob_url = url
            download_pdf(glob_url, 'corpus.pdf')
            load_recommender('corpus.pdf')
        else:
            pass
            # old_file_name = file.name
            # file_name = file.name
            # file_name = file_name[:-12] + file_name[-4:]
            # os.rename(old_file_name, file_name)
            # load_recommender(file_name)
            #load_recommender(file)
        if question.strip() == '':
            return '[ERROR]: Question field is empty'
        if model == "text-davinci-003" or model == "gpt-4" or model == "gpt-4-32k":

            answer = qna_dic[question]
            #answer = generate_answer_text_davinci_003(question, openAI_key)
        else:
            answer = generate_answer(question, openAI_key, model)
        chat_history.append([question, answer])
        return chat_history
    except openai.error.InvalidRequestError as e:
        return f'[ERROR]: Either you do not have access to GPT4 or you have exhausted your quota!'


def generate_text_text_davinci_003(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,

        engine="davinci003",
        prompt=prompt,
        temperature=0.1,
        max_tokens=400,
        top_p=1,
        frequency_penalty=0,
        presence_penalty=0,
        stop=None
    )
    message = completions.choices[0].text
    return message


def generate_answer_text_davinci_003(question, openAI_key):
    topn_chunks = recommender(question)
    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 which has 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_text_davinci_003(openAI_key, prompt, "text-davinci-003")
    return answer


# pre-defined questions
questions = [
    "What is the total scope 1 GHG emission?",
    "What is the total scope 2 GHG emission?",
    "What is the total scope 3 GHG emission?",
    "What are the main results of the study?",
    "What are the main contributions of this study?",
]

recommender = SemanticSearch()

# title = 'PDF GPT Turbo'
# description = """ PDF GPT Turbo allows you to chat with your PDF files. It uses Google's Universal Sentence Encoder with Deep averaging network (DAN) to give hallucination free response by improving the embedding quality of OpenAI. It cites the page number in square brackets([Page No.]) and shows where the information is located, adding credibility to the responses."""
#
# with gr.Blocks(css="""#chatbot { font-size: 14px; min-height: 1200; }""") as demo:
#     gr.Markdown(f'<center><h3>{title}</h3></center>')
#     gr.Markdown(description)
#
#     try:
#         with gr.Row():
#             with gr.Group():
#                 #gr.Markdown(
#                 #    f'<p style="text-align:center">Get your Open AI API key <a href="https://platform.openai.com/account/api-keys">here</a></p>')
#                 with gr.Accordion(""):
#                     #openAI_key = gr.Textbox(label='Enter your OpenAI API key here', password=True)
#                     url = gr.Textbox(label='Enter PDF URL here   (Example: https://arxiv.org/pdf/1706.03762.pdf )')
#                     gr.Markdown("<center><h4>OR<h4></center>")
#                     file = gr.File(label='Upload your PDF/ Research Paper / Book here', file_types=['.pdf'])
#                 question = gr.Textbox(label='Enter your question here')
#                 gr.Examples(
#                     [[q] for q in questions],
#                     inputs=[question],
#                     label="PRE-DEFINED QUESTIONS: Click on a question to auto-fill the input box, then press Enter!",
#                 )
#                 model = gr.Radio([
#                     'gpt-3.5-turbo',
#                     'gpt-3.5-turbo-16k',
#                     'gpt-3.5-turbo-0613',
#                     'gpt-3.5-turbo-16k-0613',
#                     'text-davinci-003',
#                     'gpt-4',
#                     'gpt-4-32k'
#                 ], label='Select Model', default='gpt-3.5-turbo')
#                 btn = gr.Button(value='Submit')
#
#                 btn.style(full_width=True)
#
#             with gr.Group():
#                 chatbot = gr.Chatbot(placeholder="Chat History", label="Chat History", lines=50, elem_id="chatbot")
#     except:
#         print("error in gradio")
#     #
#     # Bind the click event of the button to the question_answer function
#
#     try:
#         btn.click(
#             question_answer,
#             inputs=[chatbot, url, file, question, model],
#             outputs=[chatbot],
#         )
#     except:
#         print("error in btn.click")
#
# #demo.launch(server_name="10.31.8.38")
# demo.launch(share=True)

title = 'PDF GPT Turbo'
description = """ PDF GPT Turbo allows you to chat with your PDF files. It uses Google's Universal Sentence Encoder with Deep averaging network (DAN) to give hallucination free response by improving the embedding quality of OpenAI. It cites the page number in square brackets([Page No.]) and shows where the information is located, adding credibility to the responses."""

with gr.Blocks(css="""#chatbot { font-size: 14px; min-height: 1200; }""") as demo:
    gr.Markdown(f'<center><h3>{title}</h3></center>')
    gr.Markdown(description)

    try:
        with gr.Row():
            with gr.Group():
                with gr.Accordion(""):
                    url = gr.Textbox(label='Enter PDF URL here   (Example: https://arxiv.org/pdf/1706.03762.pdf )')
                    gr.Markdown("<center><h4>OR<h4></center>")
                    file = gr.File(label='Upload your PDF/ Research Paper / Book here', file_types=['.pdf'])
                question = gr.Textbox(label='Enter your question here')
                gr.Examples(
                    [[q] for q in questions],
                    inputs=[question],
                    label="PRE-DEFINED QUESTIONS: Click on a question to auto-fill the input box, then press Enter!",
                )
                model = gr.Radio([
                    'gpt-3.5-turbo',
                    'gpt-3.5-turbo-16k',
                    'gpt-3.5-turbo-0613',
                    'gpt-3.5-turbo-16k-0613',
                    'text-davinci-003',
                    'gpt-4',
                    'gpt-4-32k'
                ], label='Select Model', default='gpt-3.5-turbo')

            with gr.Group():
                btn = gr.Button(value='Submit')
                btn.style(full_width=True)
                chatbot = gr.Chatbot(placeholder="Chat History", label="Chat History", lines=50, elem_id="chatbot")
                btn.click(
                    question_answer,
                    inputs=[chatbot, url, file, question, model],
                    outputs=[chatbot],
                )
    except:
        print("error in gradio")

demo.launch(share=True)