app.py

import gradio as gr
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
import spaces

# Model configuration
MODEL_ID = "yasserrmd/DentaInstruct-1.2B"
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"

# Initialize model and tokenizer
print(f"Loading model {MODEL_ID}...")

# Load tokenizer - try the fine-tuned model first, then base model
try:
    tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
    print(f"Loaded tokenizer from {MODEL_ID}")
except Exception as e:
    print(f"Failed to load tokenizer from {MODEL_ID}: {e}")
    print("Using tokenizer from base LFM2 model...")
    try:
        tokenizer = AutoTokenizer.from_pretrained("LiquidAI/LFM2-1.2B")
    except Exception as e2:
        print(f"Failed to load LFM2 tokenizer: {e2}")
        print("Using fallback TinyLlama tokenizer...")
        tokenizer = AutoTokenizer.from_pretrained("TinyLlama/TinyLlama-1.1B-Chat-v1.0")

# Load model with proper dtype for efficiency
model = AutoModelForCausalLM.from_pretrained(
    MODEL_ID,
    torch_dtype=torch.bfloat16 if torch.cuda.is_available() else torch.float32,
    device_map="auto" if torch.cuda.is_available() else None
)

if not torch.cuda.is_available():
    model = model.to(DEVICE)

# Set padding token if not set
if tokenizer.pad_token is None:
    tokenizer.pad_token = tokenizer.eos_token

def format_prompt(message, history):
    """Format the prompt for the model"""
    messages = []
    
    # Add conversation history
    for user_msg, assistant_msg in history:
        messages.append({"role": "user", "content": user_msg})
        if assistant_msg:
            messages.append({"role": "assistant", "content": assistant_msg})
    
    # Add current message
    messages.append({"role": "user", "content": message})
    
    # Apply chat template
    if hasattr(tokenizer, 'apply_chat_template'):
        prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
    else:
        # Fallback formatting
        prompt = ""
        for msg in messages:
            if msg["role"] == "user":
                prompt += f"User: {msg['content']}\n"
            else:
                prompt += f"Assistant: {msg['content']}\n"
        prompt += "Assistant: "
    
    return prompt

@spaces.GPU(duration=60)
def generate_response(
    message,
    history,
    temperature=0.3,
    max_new_tokens=512,
    top_p=0.95,
    repetition_penalty=1.05,
):
    """Generate response from the model"""
    
    # Format the prompt
    prompt = format_prompt(message, history)
    
    # Tokenize input
    inputs = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=2048)
    inputs = {k: v.to(model.device) for k, v in inputs.items()}
    
    # Generate response
    with torch.no_grad():
        outputs = model.generate(
            **inputs,
            max_new_tokens=max_new_tokens,
            temperature=temperature,
            top_p=top_p,
            repetition_penalty=repetition_penalty,
            do_sample=True,
            pad_token_id=tokenizer.pad_token_id,
            eos_token_id=tokenizer.eos_token_id,
        )
    
    # Decode response
    response = tokenizer.decode(outputs[0][inputs['input_ids'].shape[1]:], skip_special_tokens=True)
    
    return response

# Example questions
EXAMPLES = [
    ["What are the main types of dental cavities?"],
    ["Explain the process of root canal treatment"],
    ["What is the difference between gingivitis and periodontitis?"],
    ["How should I care for my teeth after a dental extraction?"],
    ["What are the benefits of fluoride in dental care?"],
    ["Explain the stages of tooth development in children"],
    ["What causes tooth sensitivity and how can it be treated?"],
    ["Describe the different types of dental fillings available"],
]

# Custom CSS for styling
custom_css = """
.disclaimer {
    background-color: #fff3cd;
    border: 1px solid #ffc107;
    border-radius: 5px;
    padding: 10px;
    margin-bottom: 15px;
}
"""

# Create Gradio interface
with gr.Blocks(theme=gr.themes.Soft(), css=custom_css) as demo:
    gr.Markdown(
        """
        # Dental VQA Model Comparison
        
        Interactive comparison of dental visual question answering models. Currently featuring DentaInstruct-1.2B for dental education and oral health information.
        """
    )
    
    gr.HTML(
        """
        <div class="disclaimer">
        <strong>⚠️ Important Disclaimer:</strong><br>
        This model is for educational purposes only. It is NOT a substitute for professional dental care.
        Do not use this model for clinical diagnosis or treatment advice. Always consult a qualified dental professional.
        </div>
        """
    )
    
    chatbot = gr.Chatbot(
        height=400,
        label="Conversation"
    )
    
    msg = gr.Textbox(
        label="Your dental question",
        placeholder="Ask a question about dental health, procedures, or oral care...",
        lines=2
    )
    
    with gr.Row():
        submit = gr.Button("Send", variant="primary")
        clear = gr.Button("Clear")
    
    with gr.Accordion("Advanced Settings", open=False):
        temperature = gr.Slider(
            minimum=0.1,
            maximum=1.0,
            value=0.3,
            step=0.1,
            label="Temperature",
            info="Controls randomness in responses"
        )
        
        max_new_tokens = gr.Slider(
            minimum=64,
            maximum=1024,
            value=512,
            step=64,
            label="Max New Tokens",
            info="Maximum length of the response"
        )
        
        top_p = gr.Slider(
            minimum=0.1,
            maximum=1.0,
            value=0.95,
            step=0.05,
            label="Top-p",
            info="Nucleus sampling parameter"
        )
        
        repetition_penalty = gr.Slider(
            minimum=1.0,
            maximum=1.5,
            value=1.05,
            step=0.05,
            label="Repetition Penalty",
            info="Reduces repetition in responses"
        )
    
    gr.Examples(
        examples=EXAMPLES,
        inputs=msg,
        label="Example Questions"
    )
    
    gr.Markdown(
        """
        ## About This Model
        
        DentaInstruct-1.2B is a specialised language model fine-tuned on dental educational content.
        It's designed to provide educational information about dental health, procedures, and oral care.
        
        **Model Details:**
        - Base Model: LFM2-1.2B
        - Parameters: 1.17B
        - Training Data: Dental subset of MIRIAD dataset
        - Purpose: Educational dental information
        
        **Created by:** @yasserrmd | **Space by:** @chrisvoncsefalvay
        """
    )
    
    # Event handlers
    def respond(message, chat_history, temperature, max_new_tokens, top_p, repetition_penalty):
        response = generate_response(
            message, 
            chat_history,
            temperature,
            max_new_tokens,
            top_p,
            repetition_penalty
        )
        chat_history.append((message, response))
        return "", chat_history
    
    msg.submit(
        respond,
        [msg, chatbot, temperature, max_new_tokens, top_p, repetition_penalty],
        [msg, chatbot]
    )
    
    submit.click(
        respond,
        [msg, chatbot, temperature, max_new_tokens, top_p, repetition_penalty],
        [msg, chatbot]
    )
    
    clear.click(lambda: None, None, chatbot, queue=False)

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