app.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

api_key = os.environ['API_TOKEN'] 

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'[{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):
        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(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 'Text Not Found in PDF'. Ignore outlier "\
              "search results which has nothing to do with the question. Only answer what is asked. The "\
              "answer should be short and concise. Answer step-by-step. \n\nQuery: {question}\nAnswer: "
    
    prompt += f"Query: {question}\nAnswer:"
    answer = generate_text(api_key, prompt,"text-davinci-003")
    return answer


def question_answer(url, file, question):
    #if openAI_key.strip()=='':
    #    return '[ERROR]: Please enter you Open AI Key. Get your key here : https://platform.openai.com/account/api-keys'
    if url.strip() == '' and file == None:
        return '[ERROR]: URL 和 PDF 都是空的。 至少提供一個。'
    
    if url.strip() != '' and file != None:
        return '[ERROR]: 提供了 URL 和 PDF。 請僅提供一個（網址或 PDF）。'

    if url.strip() != '':
        glob_url = url
        download_pdf(glob_url, 'corpus.pdf')
        load_recommender('corpus.pdf')

    else:
        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)

    if question.strip() == '':
        return '[ERROR]: 問題字段為空'

    return generate_answer(question,api_key)


recommender = SemanticSearch()

css = """ 
        .gradio-container {
            background-image: linear-gradient(#d7d7d7, #f2f2f2);
            padding: 0;
            
        }
        .app.svelte-p7tiy3.svelte-p7tiy3 {
            padding: 10;
        }
        .padded.svelte-faijhx {
            padding: 30px 0 30px 0;
            background-color: transparent;
        }
        #markdown-or{
            background-color: transparent;
        }
        
        :root,.gradio-container-3-20-1 :host {
          --color-border-primary:transparent;
        }
        
        #submit_button{
            background-color: #fff;
            font-weight: bold;
            box-shadow: 5px 10px 18px #fff;
        }

        footer {
            visibility: hidden;
        }
"""

title = 'AI Pdf 歸納器'
#description = """ KrystalPDF AI allows you to chat with your PDF file. It gives hallucination free response than other tools. The returned response can even cite the page number in square brackets([]) where the information is located, adding credibility to the responses and helping to locate pertinent information quickly."""

with gr.Blocks(css=css) as demo:

    #gr.Markdown(f'<center><h1>{title}</h1></center>')
    #gr.Markdown(description)

    with gr.Row(css=css):
        
        with gr.Group(css=css):
            #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>')
            #openAI_key=gr.Textbox(label='Enter your OpenAI API key here')
            url = gr.Textbox(label='在此處輸入 PDF 網址')
            gr.Markdown("<center>或</center>", elem_id="markdown-or")
            file = gr.File(label='在此處上傳您的 PDF/研究論文/書籍', file_types=['.pdf'], placeholder="將文件拖放到此處 或 點擊上傳")
            question = gr.Textbox(label='在這裡輸入您的問題', elem_id="question")
            btn = gr.Button(value='提交', elem_id="submit_button")
            btn.style(full_width=True)

            answer = gr.Textbox(label='你的提問的答案是：', show_copy_button=True)

        #openAI_key=api_key
        btn.click(question_answer, inputs=[url, file, question], outputs=[answer])
demo.launch()