app.py

import gradio as gr
from llama_cpp import Llama
from huggingface_hub import hf_hub_download
import random

# Initialize model
model_path = hf_hub_download(
    repo_id="AstroMLab/AstroSage-8B-GGUF",
    filename="AstroSage-8B-Q8_0.gguf"
)

llm = Llama(
    model_path=model_path,
    n_ctx=2048,
    n_threads=4,
    chat_format="llama-3",
    seed=42,
    f16_kv=True,
    logits_all=False,
    use_mmap=True,
    use_gpu=True
)

# Placeholder responses for when context is empty
GREETING_MESSAGES = [
    "Greetings! I am AstroSage, your guide to the cosmos. What would you like to explore today?",
    "Welcome to our cosmic journey! I am AstroSage. How may I assist you in understanding the universe?",
    "AstroSage here. Ready to explore the mysteries of space and time. How may I be of assistance?",
    "The universe awaits! I'm AstroSage. What astronomical wonders shall we discuss?",
]

def get_random_greeting():
    return random.choice(GREETING_MESSAGES)

def respond(message, history, system_message, max_tokens, temperature, top_p):
    messages = [{"role": "system", "content": system_message}]
    for user_msg, assistant_msg in history:
        if user_msg:
            messages.append({"role": "user", "content": user_msg})
        if assistant_msg:
            messages.append({"role": "assistant", "content": assistant_msg})
    messages.append({"role": "user", "content": message})
    
    try:
        # Stream response from LLM
        stream = llm.create_chat_completion(
            messages=messages,
            max_tokens=max_tokens,
            temperature=temperature,
            top_p=top_p,
            stream=True  # Enable streaming
        )
        response_content = ""
        for chunk in stream:
            response_content += chunk["choices"][0]["delta"]["content"]
            yield response_content  # Stream each chunk back to the frontend
    except Exception as e:
        yield f"Error: {e}"

def clear_context():
    greeting_message = get_random_greeting()
    return [("", greeting_message)], ""

# Gradio Interface
with gr.Blocks() as demo:
    gr.HTML("<div class='header-text'>AstroSage-LLAMA-3.1-8B</div><div class='subheader'>Astronomy-Specialized Chatbot</div>")
    
    chatbot = gr.Chatbot(height=400)
    msg = gr.Textbox(placeholder="Ask about astronomy, astrophysics, or cosmology...", show_label=False)
    
    with gr.Accordion("Advanced Settings", open=False) as advanced_settings:
        system_msg = gr.Textbox(
            value="You are AstroSage, a highly knowledgeable AI assistant specialized in astronomy, astrophysics, and cosmology. Provide accurate, engaging, and educational responses about space science and the universe.",
            label="System Message",
            lines=3
        )
        max_tokens = gr.Slider(1, 2048, value=512, step=1, label="Max Tokens")
        temperature = gr.Slider(0.1, 4.0, value=0.7, step=0.1, label="Temperature")
        top_p = gr.Slider(0.1, 1.0, value=0.9, step=0.05, label="Top-p")
    
    # Automatically handle submission on Enter key press with streaming
    def handle_submit(message, history, system_message, max_tokens, temperature, top_p):
        history.append((message, None))  # Append user's message first
        
        # Stream the assistant's response and update the history
        for response in respond(message, history, system_message, max_tokens, temperature, top_p):
            history[-1] = (message, response)
            yield history, ""  # Yield updated history to display in the chatbox

    # Use the new Gradio `chatbot.update` pattern by returning the updated value
    msg.submit(
        handle_submit,
        inputs=[msg, chatbot, system_msg, max_tokens, temperature, top_p],
        outputs=[chatbot, msg],
        queue=False
    )
    
    # Automatically clear context on reload with a greeting
    demo.load(lambda: clear_context(), None, [chatbot, msg])

if __name__ == "__main__":
    demo.launch()