app.py

import gradio as gr
import os
import spaces
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
from threading import Thread

# Set an environment variable
HF_TOKEN = os.environ.get("HF_TOKEN", None)

DESCRIPTION = '''
<div>
  <h1 style="text-align: center;">A.I. Healthcare</h1>
</div>
'''

LICENSE = """
<p>
This Health Assistant is designed to provide helpful healthcare information; however, it may make mistakes and is not designed to replace professional medical care. It is not intended to diagnose any condition or disease. Always consult with a qualified healthcare provider for any medical concerns.
</p>
"""

PLACEHOLDER = """
<div style="padding: 30px; text-align: center; display: flex; flex-direction: column; align-items: center;">
   <h1 style="font-size: 28px; margin-bottom: 2px; opacity: 0.55;">A.I. Healthcare</h1>
   <p style="font-size: 18px; margin-bottom: 2px; opacity: 0.65;">Ask me anything...</p>
</div>
"""

css = """
h1 {
  text-align: center;
  display: block;
}

#duplicate-button {
  margin: auto;
  color: white;
  background: #1565c0;
  border-radius: 100vh;
}
"""

# Load the tokenizer and model
tokenizer = AutoTokenizer.from_pretrained("reedmayhew/HealthCare-Reasoning-Assistant-Llama-3.1-8B-HF", device_map="cuda")
model = AutoModelForCausalLM.from_pretrained("reedmayhew/HealthCare-Reasoning-Assistant-Llama-3.1-8B-HF", device_map="cuda")

terminators = [
    tokenizer.eos_token_id,
    tokenizer.convert_tokens_to_ids("<|eot_id|>")
]

@spaces.GPU(duration=60)
def chat_llama3_8b(message: str, 
                   history: list, 
                   temperature: float, 
                   max_new_tokens: int,
                   confirm: bool) -> str:
    """
    Generate a streaming response using the llama3-8b model.
    
    Args:
        message (str): The input message.
        history (list): The conversation history.
        temperature (float): The temperature for generating the response.
        max_new_tokens (int): The maximum number of new tokens to generate.
        confirm (bool): Whether the user has confirmed the age/disclaimer.
        
    Returns:
        str: The generated response.
    """
    # If the confirmation checkbox is not checked, return a short message immediately.
    if not confirm:
        return "⚠️ You must confirm that you meet the usage requirements before sending a message."
    
    conversation = []
    for user, assistant in history:
        conversation.extend([
            {"role": "user", "content": user}, 
            {"role": "assistant", "content": assistant}
        ])
    
    # Ensure the model starts with "<think>"
    conversation.append({"role": "user", "content": message})
    conversation.append({"role": "assistant", "content": "<think> "})  # Force <think> at start

    input_ids = tokenizer.apply_chat_template(conversation, return_tensors="pt").to(model.device)
    
    streamer = TextIteratorStreamer(tokenizer, timeout=10.0, skip_prompt=True, skip_special_tokens=True)

    generate_kwargs = dict(
        input_ids=input_ids,
        streamer=streamer,
        max_new_tokens=max_new_tokens,
        do_sample=True,
        temperature=temperature,
        eos_token_id=terminators,
    )
    
    if temperature == 0:
        generate_kwargs['do_sample'] = False
        
    t = Thread(target=model.generate, kwargs=generate_kwargs)
    t.start()

    outputs = []
    buffer = ""
    think_detected = False
    thinking_message_sent = False
    full_response = ""  # Store the full assistant response

    for text in streamer:
        buffer += text
        full_response += text  # Store raw assistant response (includes <think>)

        # Send the "thinking" message once text starts generating
        if not thinking_message_sent:
            thinking_message_sent = True
            yield "A.I. Healthcare is Thinking...\n\n"

        # Wait until </think> is detected before streaming output
        if not think_detected:
            if "</think>" in buffer:
                think_detected = True
                buffer = buffer.split("</think>", 1)[1]  # Remove <think> section
        else:
            outputs.append(text)
            yield "".join(outputs)

    # Store the full response (including <think>) in history for context
    history.append((message, full_response))

# Custom JavaScript to disable the send button until confirmation is given.
# (The JS waits for the checkbox with a label containing the specified text and then monitors its state.)
CUSTOM_JS = """
<script>
document.addEventListener("DOMContentLoaded", function() {
    // Poll for the confirmation checkbox and the send button inside the ChatInterface.
    const interval = setInterval(() => {
        // The checkbox is rendered as an <input type="checkbox"> with an associated label.
        const checkbox = document.querySelector('input[type="checkbox"][aria-label*="I hereby confirm that I am at least 18 years of age"]');
        // The send button might be a <button> element with a title or specific text. Adjust the selector as needed.
        const sendButton = document.querySelector('button[title="Send"]');
        if (checkbox && sendButton) {
            sendButton.disabled = !checkbox.checked;
            checkbox.addEventListener('change', function() {
                sendButton.disabled = !checkbox.checked;
            });
            clearInterval(interval);
        }
    }, 500);
});
</script>
"""

with gr.Blocks(css=css, title="A.I. Healthcare") as demo:
    gr.Markdown(DESCRIPTION)
    # Inject the custom JavaScript.
    gr.HTML(CUSTOM_JS)
    
    # The ChatInterface below now includes additional inputs: the confirmation checkbox and the parameter sliders.
    chat_interface = gr.ChatInterface(
        fn=chat_llama3_8b,
        title="A.I. Healthcare Chat",
        chatbot=gr.Chatbot(height=450, placeholder=PLACEHOLDER, label='Conversation'),
        additional_inputs=[
            gr.Checkbox(
                value=False,
                label=("I hereby confirm that I am at least 18 years of age (or accompanied by a legal guardian "
                       "who is at least 18 years old), understand that the information provided by this service "
                       "is for informational purposes only and is not intended to diagnose or treat any medical condition, "
                       "and acknowledge that I am solely responsible for verifying any information provided."),
                elem_id="age_confirm_checkbox"
            ),
            gr.Slider(minimum=0.6, maximum=0.6, step=0.1, value=0.6, label="Temperature", visible=False),
            gr.Slider(minimum=128, maximum=4096, step=64, value=1024, label="Max new tokens", visible=False),
        ],
        examples=[
            ['What are the common symptoms of diabetes?'],
            ['How can I manage high blood pressure with lifestyle changes?'],
            ['What nutritional advice can help improve heart health?'],
            ['Can you explain the benefits of regular exercise for mental well-being?'],
            ['What should I know about the side effects of common medications?']
        ],
        cache_examples=False,
        allow_screenshot=False,
    )
    
    gr.Markdown(LICENSE)

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