app.py

import gradio as gr
import os
import openai
from newspaper import Article
from newspaper import Config

import json
import re
from transformers import GPT2Tokenizer
import requests

# define the text summarizer function
def text_prompt(request, system_role, page_url, api_key, temp):
    try:
        USER_AGENT = 'Mozilla/5.0 (Macintosh; Intel Mac OS X 10.15; rv:78.0) Gecko/20100101 Firefox/78.0'

        config = Config()
        config.browser_user_agent = USER_AGENT
        config.request_timeout = 10

        article = Article(page_url, config=config)
        article.download()
        article.parse()

    except Exception as e:
        return "", f"--- An error occurred while processing the URL: {e} ---", ""
    
    tokenizer = GPT2Tokenizer.from_pretrained("gpt2")
    #TODO: for chinese, separator is '。'
    sentences = article.text.split('.')
    
    tokens = []
    page_text = ""
    
    for sentence in sentences:
        tokens.extend(tokenizer.tokenize(sentence))
        
        # Trim text to a maximum of 3100 tokens
        if len(tokens) > 3100:
            break
        page_text += sentence + ". "
        
    # Delete the last space
    page_text = page_text.strip()

    num_tokens = len(tokens)

    if num_tokens > 10 and api_key.startswith("sk-"):
        openai.api_key = api_key
        # get the response from openai API
        try:
            response = openai.ChatCompletion.create(
                model="gpt-3.5-turbo",
                messages=[
                    {"role": "system", "content": system_role},
                    {"role": "user", "content": request + "\n\n" + 'Text:\n\n"' + page_text + '\n"'}
                ],
                max_tokens=512,
                temperature=temp,
                top_p=1.0,
            )
            # get the response text
            response_text = response['choices'][0]['message']['content']
            total_tokens = response["usage"]["total_tokens"]

            # clean the response text
            response_text = re.sub(r'\s+', ' ', response_text)
            response_text = f"#### [{page.title}]({article_url})\n\n{response_text.strip()}"
            total_tokens_str = str(total_tokens) + " (${:.2f} USD)".format(total_tokens/1000*0.002)


            return article.text, response_text, total_tokens_str
        except Exception as e:
            return article.text, f"--- An error occurred while processing the request: {e} ---", num_tokens
    return article.text, "--- Check API-Key or Min number of tokens:", str(num_tokens)

# define the gradio interface
iface = gr.Interface(
    fn=text_prompt,
    inputs=[gr.Textbox(lines=1, placeholder="Enter your prompt here...", label="Prompt:", type="text"),
            gr.Textbox(lines=1, placeholder="Enter your gpt-role description here...", label="GPT Role:", type="text"),
            gr.Textbox(lines=1, placeholder="Enter the Article's URL here...", label="Article's URL to parse:", type="text"),
            gr.Textbox(lines=1, placeholder="Enter your API-key here...", label="API-Key:", type="password"),
            gr.Slider(0.0,1.0, value=0.7, label="Temperature:")
            ],
    outputs=[gr.Textbox(label="Output:"), gr.Markdown(label="Output:"), gr.Markdown(label="Total Tokens:")],
    examples=[["请用简体中文生成一段200字的摘要，并提取5个关键词.", "作为新闻编辑", "https://openai.com/blog/planning-for-agi-and-beyond","",0.7],
            ["Generate a summary of the following text. Give me an overview of the main business impact from the text following this template:\n- Summary:\n- Business Impact:\n- Companies:", "Act as a Business Consultant", "https://ai.googleblog.com/2019/10/quantum-supremacy-using-programmable.html","",0.7],
            ["Generate the next insights based on the following text. Indicates N/A if the information is not available in the text.\n- Summary:\n- Acquisition Price:\n- Why is this important for the acquirer:\n- Business Line for the acquirer:\n- Tech Focus for the acquired (list):","Act as a Business Consultant", "https://techcrunch.com/2022/09/28/eqt-acquires-billtrust-a-company-automating-the-invoice-to-cash-process-for-1-7b/","",0.3]
    ],
    title="ChatGPT info extraction from URL",
    description="This tool allows querying the text retrieved from the URL with newspaper3k lib and using OpenAI's [gpt-3.5-turbo] engine.\nThe URL text can be referenced in the prompt as \"following text\".\nA GPT2 tokenizer is included to ensure that the 1.800 token limit for OpenAI queries is not exceeded. Provide a prompt with your request, the description for the system role, the url for text retrieval, your api-key and temperature to process the text."
)

# error capturing in integration as a component

error_message = ""

try:
    iface.queue(concurrency_count=20)
    iface.launch()
except Exception as e:
    error_message = "An error occurred: " + str(e)
    iface.outputs[1].value = error_message