import logging from pathlib import Path from time import perf_counter import gradio as gr from jinja2 import Environment, FileSystemLoader import requests from backend.query_llm import generate from backend.semantic_search import retriever proj_dir = Path(__file__).parent # Setting up the logging logging.basicConfig(level=logging.INFO) logger = logging.getLogger(__name__) # Set up the template environment with the templates directory env = Environment(loader=FileSystemLoader(proj_dir / 'templates')) # Load the templates directly from the environment template = env.get_template('template.j2') template_html = env.get_template('template_html.j2') def check_endpoint_status(): # Replace with the actual API URL and headers api_url = os.getenv("ENDPOINT_URL") headers = { 'accept': 'application/json', 'Authorization': f'Bearer {os.getenv("BEARER")}' } try: response = requests.get(api_url, headers=headers) response.raise_for_status() # will throw an exception for non-200 status data = response.json() # Extracting the status information status = data.get('status', {}).get('state', 'No status found') message = data.get('status', {}).get('message', 'No message found') return f"Status: {status}\nMessage: {message}" except requests.exceptions.RequestException as e: return f"Failed to get status: {str(e)}" def add_text(history, text): history = [] if history is None else history history = history + [(text, None)] return history, gr.Textbox(value="", interactive=False) def bot(history, system_prompt=""): top_k = 3 query = history[-1][0] logger.warning('Retrieving documents...') # Retrieve documents relevant to query document_start = perf_counter() documents = retriever(query, top_k=top_k) document_time = document_start - perf_counter() logger.warning(f'Finished Retrieving documents in {round(document_time, 2)} seconds...') # Create Prompt prompt = template.render(documents=documents, query=query) prompt_html = template_html.render(documents=documents, query=query) logger.warning(prompt) history[-1][1] = "" for character in generate(prompt): history[-1][1] = character yield history, prompt_html with gr.Blocks() as demo: with gr.Tab("Application"): output = gr.Textbox(check_endpoint_status, label="Endpoint Status (send chat to wake up)", every=1) chatbot = gr.Chatbot( [], elem_id="chatbot", avatar_images=('https://aui.atlassian.com/aui/8.8/docs/images/avatar-person.svg', 'https://huggingface.co/datasets/huggingface/brand-assets/resolve/main/hf-logo.svg'), bubble_full_width=False, show_copy_button=True, show_share_button=True, ) with gr.Row(): txt = gr.Textbox( scale=3, show_label=False, placeholder="Enter text and press enter", container=False, ) txt_btn = gr.Button(value="Submit text", scale=1) prompt_html = gr.HTML() # Turn off interactivity while generating if you hit enter txt_msg = txt_btn.click(add_text, [chatbot, txt], [chatbot, txt], queue=False).then( bot, chatbot, [chatbot, prompt_html]) # Turn it back on txt_msg.then(lambda: gr.Textbox(interactive=True), None, [txt], queue=False) # Turn off interactivity while generating if you hit enter txt_msg = txt.submit(add_text, [chatbot, txt], [chatbot, txt], queue=False).then( bot, chatbot, [chatbot, prompt_html]) # Turn it back on txt_msg.then(lambda: gr.Textbox(interactive=True), None, [txt], queue=False) gr.Examples(['What is the capital of China, I think its Shanghai?', 'Why is the sky blue?', 'Who won the mens world cup in 2014?',], txt) demo.queue() demo.launch(debug=True)