app.py

import gradio as gr
from gradio_client import Client, handle_file
from huggingface_hub import InferenceClient

moondream_client = Client("vikhyatk/moondream2")
llama_client = InferenceClient("Qwen/QwQ-32B-Preview")

history = []

def describe_image(image, user_message):
    global history
    result = moondream_client.predict(
        img=handle_file(image),
        prompt="Describe this image.",
        api_name="/answer_question"
    )
    
    description = result
    history.append(f"User: {user_message}")
    history.append(f"Assistant: {description}")
    
    full_conversation = "\n".join(history)
    return respond(user_message, history, "System: Describe the image.", 512, 0.7, 0.95)

def respond(message, history, system_message, max_tokens, temperature, top_p):
    messages = [{"role": "system", "content": system_message}]
    
    for val in history:
        if val[0]:
            messages.append({"role": "user", "content": val[0]})
        if val[1]:
            messages.append({"role": "assistant", "content": val[1]})
    
    messages.append({"role": "user", "content": message})
    
    response = ""
    for message in llama_client.chat_completion(
        messages,
        max_tokens=max_tokens,
        stream=True,
        temperature=temperature,
        top_p=top_p,
    ):
        token = message.choices[0].delta.content
        response += token
        yield response

def chat_or_image(image, user_message):
    global history
    if image:
        return describe_image(image, user_message)
    else:
        history.append(f"User: {user_message}")
        return respond(user_message, history, "System: Answer the user's question.", 512, 0.7, 0.95)

with gr.Blocks() as demo:
    with gr.Row():
        with gr.Column(scale=1):
            gr.Image(type="filepath", label="Upload Image (optional)", elem_id="left-column")
        with gr.Column(scale=2):
            gr.Textbox(label="Ask or Chat", placeholder="Ask a question...", lines=2, elem_id="right-column")
    
    demo.launch(show_error=True)