Spaces:
Runtime error
Runtime error
| import gradio as gr | |
| import os | |
| import openai | |
| from newspaper import Article | |
| import json | |
| import re | |
| from transformers import GPT2Tokenizer | |
| import requests | |
| # define the text summarizer function | |
| def text_prompt(request, system_role, page_urls_str, api_key, api_base, deployment_id, temp): | |
| tokenizer = GPT2Tokenizer.from_pretrained("gpt2") | |
| page_urls = [page_url_str for page_url_str in page_urls_str.split("\n") if page_url_str] | |
| if len(page_urls) == 0: | |
| return "", "urls not found", "" | |
| page_texts = [] | |
| response_texts = [] | |
| total_tokens = 0 | |
| for page_url in page_urls: | |
| try: | |
| headers = {'User-Agent': 'Chrome/83.0.4103.106'} | |
| response = requests.get(page_url, headers=headers) | |
| html = response.text | |
| page = Article('') | |
| page.set_html(html) | |
| page.parse() | |
| except Exception as e: | |
| return "", f"--- An error occurred while processing the URL: {e} ---", "" | |
| sentences = page.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 + ". " | |
| tokens.extend(tokenizer.tokenize(request)) | |
| tokens.extend(tokenizer.tokenize(system_role)) | |
| # Delete the last space | |
| page_text = page_text.strip() | |
| num_tokens = len(tokens) | |
| tokens_condition = num_tokens > 10 | |
| api_key_condition = len(api_key) > 6 | |
| deployment_id_condition = len(deployment_id) > 6 | |
| if tokens_condition and api_key_condition and deployment_id_condition: | |
| openai.api_type = "azure" | |
| openai.api_version = "2023-05-15" | |
| openai.api_base = api_base | |
| openai.api_key = api_key | |
| max_tokens = 4000 - num_tokens # TODO: change 4096 to a dictionary with the max tokens for each deploymend_id | |
| # get the response from openai API | |
| try: | |
| response = openai.ChatCompletion.create( | |
| deployment_id=deployment_id, | |
| messages=[ | |
| {"role": "system", "content": system_role}, | |
| {"role": "user", "content": request + "\n\n" + 'Text:\n\n""""' + page_text + '\n""""'} | |
| ], | |
| max_tokens=max_tokens, | |
| 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}]({page_url})\n\n{response_text.strip()}\n" | |
| except Exception as e: | |
| response_text = f"#### [{page.title}]({page_url})\n\n" | |
| response_text += f"--- An error occurred while processing the request: {e} ---\n" | |
| page_texts.append(page.text) | |
| response_texts.append(response_text) | |
| else: | |
| page_text_temp = "ERROR:\n\n" | |
| if page.text: | |
| page_text_temp += page.text | |
| response_text_temp = "#### " | |
| if page.title: | |
| response_text_temp += f"[{page.title}]({page_url})" | |
| if not tokens_condition: | |
| response_text_temp += "\n\nERROR: Tokens problems! Maybe it can't read the URL. " | |
| if not api_key_condition: | |
| response_text_temp += "\n\nERROR: API Key problems! Copy and paste the API Key (be careful with copying spaces at the beginning or end of the API Key). " | |
| if not deployment_id_condition: | |
| response_text_temp += "\n\nERROR: Deployment_id problems! Copy and paste the deployment_id (be careful with copying spaces at the beginning or end of the deployment_id). " | |
| page_texts.append(page_text_temp) | |
| response_texts.append(response_text_temp) | |
| page_texts_str = "".join([f"====== NEW URL: {URL} ======\n{page_text}\n\n" for page_text, URL in zip(page_texts, page_urls)]) | |
| response_texts_str = "\n\n".join([response_text for response_text in response_texts]) | |
| total_tokens_str = str(total_tokens) + " (${:.2f} USD)".format(total_tokens / 1000 * 0.03) | |
| return page_texts_str, response_texts_str, total_tokens_str | |
| # 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 system-role description here...", label="System Role:", type="text"), | |
| gr.Textbox(lines=10, placeholder="Enter the Articles' URLs here...", label="Articles' URLs to parse (one per line up to 10):", type="text"), | |
| gr.Textbox(lines=1, placeholder="Enter your API-key here...", label="API-Key:", type="password"), | |
| gr.Textbox(lines=1, placeholder="Enter your Azure OpenAI API base here...", label="Enter Azure API base (Endpoint):", type="text"), | |
| gr.Textbox(lines=1, placeholder="Enter your model name here...", label="Deployment ID:", type="text"), | |
| gr.Slider(0.0, 1.0, value=0.0, label="Temperature:") | |
| ], | |
| outputs=[gr.Textbox(label="Input:"), gr.Markdown(label="Output:"), gr.Markdown(label="Total Tokens:")], | |
| title="ChatGPT info extraction from URL", | |
| description="This tool allows querying the text retrieved from the URL with newspaper3k lib and using MSFT Azure 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(debug=True) | |
| except Exception as e: | |
| error_message = "An error occurred: " + str(e) | |
| iface.outputs[1].value = error_message |