import logging from functools import partial from pathlib import Path from time import perf_counter import gradio as gr from jinja2 import Environment, FileSystemLoader from transformers import AutoTokenizer from backend.query_llm import check_endpoint_status, 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') # Initialize tokenizer tokenizer = AutoTokenizer.from_pretrained('derek-thomas/jais-13b-chat-hf') # Examples examples = ['من كان طرفي معركة اكتيوم البحرية؟', 'لم السماء زرقاء؟', "من فاز بكأس العالم للرجال في عام 2014؟",] 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, hyde=False): top_k = 5 query = history[-1][0] logger.warning('Retrieving documents...') # Retrieve documents relevant to query document_start = perf_counter() if hyde: hyde_document = generate(f"Write a wikipedia article intro paragraph to answer this query: {query}").split('### Response: [|AI|]')[-1] logger.warning(hyde_document) documents = retriever(hyde_document, top_k=top_k) else: documents = retriever(query, top_k=top_k) document_time = perf_counter() - document_start logger.warning(f'Finished Retrieving documents in {round(document_time, 2)} seconds...') # Function to count tokens def count_tokens(text): return len(tokenizer.encode(text)) # Create Prompt prompt = template.render(documents=documents, query=query) # Check if the prompt is too long token_count = count_tokens(prompt) while token_count > 2048: # Shorten your documents here. This is just a placeholder for the logic you'd use. documents.pop() # Remove the last document prompt = template.render(documents=documents, query=query) # Re-render the prompt token_count = count_tokens(prompt) # Re-count tokens prompt_html = template_html.render(documents=documents, query=query) history[-1][1] = "" response = generate(prompt) history[-1][1] = response.split('### Response: [|AI|]')[-1] return history, prompt_html intro_md = """ # Arabic RAG This is a project to demonstrate Retreiver Augmented Generation (RAG) in Arabic and English. It uses [Arabic Wikipedia](https://ar.wikipedia.org/wiki) as a base to answer questions you have. A retriever ([sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2](https://huggingface.co/sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2/discussions/8)) will find the articles relevant to your query and include them in a prompt so the reader ([core42/jais-13b-chat](https://huggingface.co/core42/jais-13b-chat)) can then answer your questions on it. You can see the prompt clearly displayed below the chatbot to understand what is going to the LLM. # Read this if you get an error I'm using Inference Endpoint's Scale to Zero to save money on GPUs. If the staus shows its not "Running" send a chat to wake it up. You will get a `500 error` and it will take ~7 min to wake up. """ with gr.Blocks() as demo: gr.Markdown(intro_md) endpoint_status = gr.Textbox(check_endpoint_status, label="Inference Endpoint Status", every=1) with gr.Tab("Arabic-RAG"): 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 query in Arabic or English and press enter", container=False, ) txt_btn = gr.Button(value="Submit text", scale=1) gr.Examples(examples, txt) prompt_html = gr.HTML() # Turn off interactivity while generating if you click 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) with gr.Tab("Arabic-RAG + HyDE"): hyde_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(): hyde_txt = gr.Textbox( scale=3, show_label=False, placeholder="Enter text and press enter", container=False, ) hyde_txt_btn = gr.Button(value="Submit text", scale=1) gr.Examples(examples, hyde_txt) hyde_prompt_html = gr.HTML() # Turn off interactivity while generating if you click hyde_txt_msg = hyde_txt_btn.click(add_text, [hyde_chatbot, hyde_txt], [hyde_chatbot, hyde_txt], queue=False).then( partial(bot, hyde=True), [hyde_chatbot], [hyde_chatbot, hyde_prompt_html]) # Turn it back on hyde_txt_msg.then(lambda: gr.Textbox(interactive=True), None, [hyde_txt], queue=False) # Turn off interactivity while generating if you hit enter hyde_txt_msg = hyde_txt.submit(add_text, [hyde_chatbot, hyde_txt], [hyde_chatbot, hyde_txt], queue=False).then( partial(bot, hyde=True), [hyde_chatbot], [hyde_chatbot, hyde_prompt_html]) # Turn it back on hyde_txt_msg.then(lambda: gr.Textbox(interactive=True), None, [hyde_txt], queue=False) demo.queue() demo.launch(debug=True)