app.py

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

# Model configuration - Using only Me-LLaMA 13B-chat
ME_LLAMA_MODEL = "clinicalnlplab/me-llama-13b-chat"

# System prompts for different phases
CONSULTATION_PROMPT = """You are a professional virtual doctor. Your goal is to collect detailed information about the user's health condition, symptoms, medical history, medications, lifestyle, and other relevant data.
Ask 1-2 follow-up questions at a time to gather more details about:
- Detailed description of symptoms
- Duration (when did it start?)
- Severity (scale of 1-10)
- Aggravating or alleviating factors
- Related symptoms
- Medical history
- Current medications and allergies
After collecting sufficient information (4-5 exchanges), summarize findings and suggest when they should seek professional care. Do NOT make specific diagnoses or recommend specific treatments.
Respond empathetically and clearly. Always be professional and thorough."""

MEDICINE_PROMPT = """You are a specialized medical assistant. Based on the patient information gathered, provide:
1. One specific over-the-counter medicine with proper adult dosing instructions
2. One practical home remedy that might help
3. Clear guidance on when to seek professional medical care

Be concise, practical, and focus only on general symptom relief. Do not diagnose. Include a disclaimer that you are not a licensed medical professional.

Patient information: {patient_info}"""

# Global variables
me_llama_model = None
me_llama_tokenizer = None
conversation_turns = 0
patient_data = []

def build_me_llama_prompt(system_prompt, history, user_input):
    """Format the conversation for Me-LLaMA chat model."""
    prompt = f"<s>[INST] <<SYS>>\n{system_prompt}\n<</SYS>>\n\n"
    
    # Add conversation history
    for user_msg, assistant_msg in history:
        prompt += f"{user_msg} [/INST] {assistant_msg} </s><s>[INST] "
    
    # Add the current user input
    prompt += f"{user_input} [/INST] "
    
    return prompt

@spaces.GPU
def load_model_if_needed():
    """Load Me-LLaMA model only when GPU is available."""
    global me_llama_model, me_llama_tokenizer
    
    if me_llama_model is None:
        print("Loading Me-LLaMA 13B-chat model...")
        me_llama_tokenizer = AutoTokenizer.from_pretrained(ME_LLAMA_MODEL)
        me_llama_model = AutoModelForCausalLM.from_pretrained(
            ME_LLAMA_MODEL,
            torch_dtype=torch.float16,
            device_map="auto",
            trust_remote_code=True
        )
        print("Me-LLaMA 13B-chat model loaded successfully!")

@spaces.GPU
def generate_medicine_suggestions(patient_info):
    """Use Me-LLaMA to generate medicine and remedy suggestions."""
    load_model_if_needed()
    
    # Create a simple prompt for medicine suggestions
    prompt = f"<s>[INST] {MEDICINE_PROMPT.format(patient_info=patient_info)} [/INST] "
    
    inputs = me_llama_tokenizer(prompt, return_tensors="pt")
    
    # Move inputs to the same device as the model
    if torch.cuda.is_available():
        inputs = {k: v.to(me_llama_model.device) for k, v in inputs.items()}
    
    with torch.no_grad():
        outputs = me_llama_model.generate(
            inputs["input_ids"],
            attention_mask=inputs["attention_mask"],
            max_new_tokens=300,
            temperature=0.7,
            top_p=0.9,
            do_sample=True,
            pad_token_id=me_llama_tokenizer.eos_token_id
        )
    
    suggestion = me_llama_tokenizer.decode(outputs[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True)
    return suggestion

@spaces.GPU
def generate_response(message, history):
    """Generate response using only Me-LLaMA for both consultation and medicine suggestions."""
    global conversation_turns, patient_data
    
    # Load model if needed
    load_model_if_needed()
    
    # Track conversation turns
    conversation_turns += 1
    
    # Store patient data
    patient_data.append(message)
    
    # Phase 1-3: Information gathering
    if conversation_turns < 4:
        # Build consultation prompt
        prompt = build_me_llama_prompt(CONSULTATION_PROMPT, history, message)
        
        inputs = me_llama_tokenizer(prompt, return_tensors="pt")
        
        # Move inputs to the same device as the model
        if torch.cuda.is_available():
            inputs = {k: v.to(me_llama_model.device) for k, v in inputs.items()}
        
        # Generate consultation response
        with torch.no_grad():
            outputs = me_llama_model.generate(
                inputs["input_ids"],
                attention_mask=inputs["attention_mask"],
                max_new_tokens=400,
                temperature=0.7,
                top_p=0.9,
                do_sample=True,
                pad_token_id=me_llama_tokenizer.eos_token_id
            )
        
        # Decode response
        full_response = me_llama_tokenizer.decode(outputs[0], skip_special_tokens=False)
        response = full_response.split('[/INST]')[-1].split('</s>')[0].strip()
        
        return response
    
    # Phase 4+: Summary and medicine suggestions
    else:
        # First, get summary from consultation
        summary_prompt = build_me_llama_prompt(
            CONSULTATION_PROMPT + "\n\nNow summarize what you've learned and suggest when professional care may be needed.",
            history, 
            message
        )
        
        inputs = me_llama_tokenizer(summary_prompt, return_tensors="pt")
        
        if torch.cuda.is_available():
            inputs = {k: v.to(me_llama_model.device) for k, v in inputs.items()}
        
        # Generate summary
        with torch.no_grad():
            outputs = me_llama_model.generate(
                inputs["input_ids"],
                attention_mask=inputs["attention_mask"],
                max_new_tokens=400,
                temperature=0.7,
                top_p=0.9,
                do_sample=True,
                pad_token_id=me_llama_tokenizer.eos_token_id
            )
        
        summary_response = me_llama_tokenizer.decode(outputs[0], skip_special_tokens=False)
        summary = summary_response.split('[/INST]')[-1].split('</s>')[0].strip()
        
        # Then get medicine suggestions using the same model
        full_patient_info = "\n".join(patient_data) + f"\n\nMedical Summary: {summary}"
        medicine_suggestions = generate_medicine_suggestions(full_patient_info)
        
        # Combine both responses
        final_response = (
            f"**MEDICAL SUMMARY:**\n{summary}\n\n"
            f"**MEDICATION AND HOME CARE SUGGESTIONS:**\n{medicine_suggestions}\n\n"
            f"**DISCLAIMER:** This is AI-generated advice for informational purposes only. Please consult a licensed healthcare provider for proper medical diagnosis and treatment."
        )
        
        return final_response

# Create the Gradio interface
demo = gr.ChatInterface(
    fn=generate_response,
    title="🏥 Complete Medical Assistant - Me-LLaMA 13B",
    description="Comprehensive medical consultation powered by Me-LLaMA 13B-chat. One model handles both consultation and medicine suggestions. Tell me about your symptoms!",
    examples=[
        "I have a persistent cough and sore throat for 3 days",
        "I've been having severe headaches and feel dizzy",
        "My stomach hurts and I feel nauseous after eating"
    ],
    theme="soft"
)

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