import gradio as gr from huggingface_hub import snapshot_download from llama_cpp import Llama # System prompt text SYSTEM_PROMPT = ( "Your name is ANIMA, an Advanced Nature Inspired Multidisciplinary Assistant, " "and a leading expert in biomimicry, biology, engineering, industrial design, " "environmental science, physiology, and paleontology. You were instructed to " "understand, learn from, and emulate the strategies used by living things to help " "users create sustainable designs and technologies.\n\n" "Your goal is to help the user work in a step by step way through the Biomimicry Design " "Process to propose biomimetic solutions to a challenge.\n\n" "Use the questions listed below as a guide to help you reflect on your work:\n" "• How does context play a role?\n" "• Are the strategies operating at the same or different scales (nano, micro, macro, meso)?\n" "• Are there repeating shapes, forms, or textures?\n" "• What behaviors or processes are occurring?\n" "• What relationships are at play?\n" "• Does information play a role? How does it flow?\n" "• How do your strategies relate to the different systems they are part of?\n\n" "Consider each of your abstracted design strategies in relation to the original design " "question or problem you identified in the Define step. Ask, “How can this strategy inform " "our design solution?” Write down all of your ideas and then analyze them.\n\n" "Think about how the strategies and design concepts you are working with relate to nature " "unifying patterns. What is their role in the larger system? How can you use a systems view " "to get to a deeper level of emulation or a more life-friendly solution?\n\n" "Nature's Unifying Patterns:\n" "Nature uses only the energy it needs and relies on freely available energy.\n" "Nature recycles all materials.\n" "Nature is resilient to disturbances.\n" "Nature tends to optimize rather than maximize.\n" "Nature provides mutual benefits.\n" "Nature runs on information.\n" "Nature uses chemistry and materials that are safe for living beings.\n" "Nature builds using abundant resources, incorporating rare resources only sparingly.\n" "Nature is locally attuned and responsive.\n" "Nature uses shape to determine functionality." ) SYSTEM_TOKEN = 1587 USER_TOKEN = 2188 BOT_TOKEN = 12435 LINEBREAK_TOKEN = 13 ROLE_TOKENS = { "user": USER_TOKEN, "bot": BOT_TOKEN, "system": SYSTEM_TOKEN } def get_message_tokens(model, role, content): message_tokens = model.tokenize(content.encode("utf-8")) message_tokens.insert(1, ROLE_TOKENS[role]) message_tokens.insert(2, LINEBREAK_TOKEN) message_tokens.append(model.token_eos()) return message_tokens def get_system_tokens(model): system_message = {"role": "system", "content": SYSTEM_PROMPT} return get_message_tokens(model, **system_message) repo_name = "Severian/ANIMA-Phi-Neptune-Mistral-7B-gguf" model_name = "ANIMA-Phi-Neptune-Mistral-7B-gguf" snapshot_download(repo_id=repo_name, local_dir=".", allow_patterns=model_name) model = Llama( model_path=model_name, n_ctx=2000, n_parts=1, ) max_new_tokens = 1500 def user(message, history): new_history = history + [[message, None]] return "", new_history def bot( history, system_prompt, top_p, top_k, temp ): tokens = get_system_tokens(model)[:] tokens.append(LINEBREAK_TOKEN) for user_message, bot_message in history[:-1]: message_tokens = get_message_tokens(model=model, role="user", content=user_message) tokens.extend(message_tokens) if bot_message: message_tokens = get_message_tokens(model=model, role="bot", content=bot_message) tokens.extend(message_tokens) last_user_message = history[-1][0] message_tokens = get_message_tokens(model=model, role="user", content=last_user_message) tokens.extend(message_tokens) role_tokens = [model.token_bos(), BOT_TOKEN, LINEBREAK_TOKEN] tokens.extend(role_tokens) generator = model.generate( tokens, top_k=top_k, top_p=top_p, temp=temp ) partial_text = "" for i, token in enumerate(generator): if token == model.token_eos() or (max_new_tokens is not None and i >= max_new_tokens): break partial_text += model.detokenize([token]).decode("utf-8", "ignore") history[-1][1] = partial_text yield history with gr.Blocks( theme=gr.themes.Soft() ) as demo: #favicon = '' gr.Markdown( f"""

{favicon}ANIMA

""" ) with gr.Blocks(theme=gr.themes.Soft()) as demo: with gr.Row(): with gr.Column(scale=5): system_prompt = gr.Textbox(label="System Prompt", placeholder="", value=SYSTEM_PROMPT, interactive=False) chatbot = gr.Chatbot(label="Dialogue").style(height=400) with gr.Column(min_width=80, scale=1): with gr.Tab(label="Generation Parameters"): # "Top-p" remains the same top_p = gr.Slider( minimum=0.0, maximum=1.0, value=0.9, step=0.05, interactive=True, label="Top-p", ) # "Top-k" remains the same top_k = gr.Slider( minimum=10, maximum=100, value=30, step=5, interactive=True, label="Top-k", ) # "Температура" translates to "Temperature" temp = gr.Slider( minimum=0.0, maximum=2.0, value=0.01, step=0.01, interactive=True, label="Temperature", ) with gr.Row(): with gr.Column(): # "Отправить сообщение" translates to "Send Message" msg = gr.Textbox( label="Send Message", placeholder="Send Message", show_label=False, ).style(container=False) with gr.Column(): with gr.Row(): # "Отправить" translates to "Submit" submit = gr.Button("Submit") # "Остановить" translates to "Stop" stop = gr.Button("Stop") # "Очистить" translates to "Clear" clear = gr.Button("Clear") with gr.Row(): # The warning message translates to "WARNING: The model may generate factually or ethically incorrect text. We are not responsible for this." gr.Markdown( """WARNING: The model may generate factually or ethically incorrect text. We are not responsible for this.""" ) # Pressing Enter submit_event = msg.submit( fn=user, inputs=[msg, chatbot], outputs=[msg, chatbot], queue=False, ).success( fn=bot, inputs=[ chatbot, system_prompt, top_p, top_k, temp ], outputs=chatbot, queue=True, ) # Pressing the button submit_click_event = submit.click( fn=user, inputs=[msg, chatbot], outputs=[msg, chatbot], queue=False, ).success( fn=bot, inputs=[ chatbot, system_prompt, top_p, top_k, temp ], outputs=chatbot, queue=True, ) # Stop generation stop.click( fn=None, inputs=None, outputs=None, cancels=[submit_event, submit_click_event], queue=False, ) # Clear history clear.click(lambda: None, None, chatbot, queue=False) demo.queue(max_size=128, concurrency_count=1) demo.launch()