app.py

import gradio as gr
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from datasets import load_dataset
import random
import re

SYSTEM_PROMPT = """
You are a medical expert. Answer the medical question with careful analysis and explain why the selected option is correct in 2 sentences without repeating.
Respond in the following format:
<answer>
[correct answer]
</answer>
<reasoning>
[explain why the selected option is correct]
</reasoning>
"""

model_name = "abaryan/BioXP-0.5B-MedMCQA"
model = AutoModelForCausalLM.from_pretrained(model_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)
dataset = load_dataset("openlifescienceai/medmcqa")

# Move model to GPU if available
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = model.to(device)
model.eval()

def get_random_question():
    """Get a random question from the dataset"""
    index = random.randint(0, len(dataset['validation']) - 1)
    question_data = dataset['validation'][index]
    return (
        question_data['question'],
        question_data['opa'],
        question_data['opb'],
        question_data['opc'],
        question_data['opd'],
        question_data.get('cop', None),  # Correct option (0-3)
        question_data.get('exp', None)   # Explanation
    )

def predict(question: str, option_a: str = "", option_b: str = "", option_c: str = "", option_d: str = "", 
           correct_option: int = None, explanation: str = None,
           temperature: float = 0.6, top_p: float = 0.9, max_tokens: int = 256):
    
    # Determine if this is an MCQ by checking if any option is provided
    is_mcq = any(opt.strip() for opt in [option_a, option_b, option_c, option_d])
    
    if is_mcq:
        options = []
        if option_a.strip(): options.append(f"A. {option_a}")
        if option_b.strip(): options.append(f"B. {option_b}")
        if option_c.strip(): options.append(f"C. {option_c}")
        if option_d.strip(): options.append(f"D. {option_d}")
        
        formatted_question = f"Question: {question}\n\nOptions:\n" + "\n".join(options)
        system_prompt = SYSTEM_PROMPT
    else:
        # Format regular question
        formatted_question = f"Question: {question}"
        system_prompt = SYSTEM_PROMPT
    
    prompt = [
        {'role': 'system', 'content': system_prompt},
        {'role': 'user', 'content': formatted_question}
    ]
    
    text = tokenizer.apply_chat_template(prompt, tokenize=False, add_generation_prompt=True)    
    model_inputs = tokenizer([text], return_tensors="pt").to(device)
    
    with torch.inference_mode():
        generated_ids = model.generate(
            **model_inputs,
            max_new_tokens=max_tokens,
            temperature=temperature,
            top_p=top_p,
        )
    
    generated_ids = generated_ids[0, model_inputs.input_ids.shape[1]:]
    model_response = tokenizer.decode(generated_ids, skip_special_tokens=True)
    
    # Clean up the response by removing tags and formatting
    cleaned_response = model_response
    cleaned_response = re.sub(r'<answer>\s*([A-D])\s*</answer>', r'Answer: \1', cleaned_response, flags=re.IGNORECASE)
    cleaned_response = re.sub(r'<reasoning>\s*(.*?)\s*</reasoning>', r'Reasoning:\n\1', cleaned_response, flags=re.IGNORECASE | re.DOTALL)
    
    # Format output with evaluation if available (only for MCQs)
    output = cleaned_response
    
    # if is_mcq and correct_option is not None:
    #     correct_letter = chr(65 + correct_option)
    #     answer_match = re.search(r"Answer:\s*([A-D])", cleaned_response, re.IGNORECASE)
    #     model_answer = answer_match.group(1).upper() if answer_match else "Not found"
        
    #     is_correct = model_answer == correct_letter
    #     output += f"\n\n---\nEvaluation:\n"
    #     output += f"Correct Answer: {correct_letter}\n"
    #     output += f"Model's Answer: {model_answer}\n"
    #     output += f"Result: {'✅ Correct' if is_correct else '❌ Incorrect'}\n"
    #     if explanation:
    #         output += f"\nExpert Explanation:\n{explanation}"
    
    return output

with gr.Blocks(
    title="BioXP Medical MCQ Assistant",
    theme=gr.themes.Soft(
        primary_hue="blue",
        secondary_hue="blue",
        neutral_hue="slate",
        radius_size="md",
        font=["Inter", "ui-sans-serif", "system-ui", "sans-serif"],
    )
) as demo:
    gr.Markdown("""
    # BioXP Medical MCQ Assistant
    A specialized AI assistant for medical multiple-choice questions.
    """)
    
    with gr.Row():
        with gr.Column(scale=1):
            question = gr.Textbox(
                label="Medical Question",
                placeholder="Enter your medical question here...",
                lines=3,
                interactive=True,
                elem_classes=["mobile-input"]
            )
            
            with gr.Accordion("Options", open=True):
                option_a = gr.Textbox(
                    label="Option A",
                    placeholder="Enter option A...",
                    interactive=True,
                    elem_classes=["mobile-input"]
                )
                option_b = gr.Textbox(
                    label="Option B",
                    placeholder="Enter option B...",
                    interactive=True,
                    elem_classes=["mobile-input"]
                )
                option_c = gr.Textbox(
                    label="Option C",
                    placeholder="Enter option C...",
                    interactive=True,
                    elem_classes=["mobile-input"]
                )
                option_d = gr.Textbox(
                    label="Option D",
                    placeholder="Enter option D...",
                    interactive=True,
                    elem_classes=["mobile-input"]
                )
            
            with gr.Accordion("Advanced Settings", open=False):
                with gr.Row():
                    with gr.Column(scale=1):
                        temperature = gr.Slider(
                            minimum=0.1,
                            maximum=1.0,
                            value=0.6,
                            step=0.1,
                            label="Temperature",
                            info="Higher = more creative, Lower = more focused"
                        )
                    with gr.Column(scale=1):
                        top_p = gr.Slider(
                            minimum=0.1,
                            maximum=1.0,
                            value=0.9,
                            step=0.1,
                            label="Top P",
                            info="Controls response diversity"
                        )
                max_tokens = gr.Slider(
                    minimum=50,
                    maximum=512,
                    value=256,
                    step=32,
                    label="Max Response Length",
                    info="Maximum length of the response"
                )
            
            # Hidden fields
            correct_option = gr.Number(visible=False)
            expert_explanation = gr.Textbox(visible=False)
            
            with gr.Row():
                predict_btn = gr.Button("Get Answer", variant="primary", size="lg", elem_classes=["mobile-button"])
                random_btn = gr.Button("Random Question", variant="secondary", size="lg", elem_classes=["mobile-button"])
        
        with gr.Column(scale=1):
            output = gr.Textbox(
                label="Model's Response",
                lines=12,
                elem_classes=["response-box", "mobile-output"]
            )
    
    # Set up button actions
    predict_btn.click(
        fn=predict,
        inputs=[
            question, option_a, option_b, option_c, option_d,
            correct_option, expert_explanation,
            temperature, top_p, max_tokens
        ],
        outputs=output
    )
    
    random_btn.click(
        fn=get_random_question,
        inputs=[],
        outputs=[question, option_a, option_b, option_c, option_d, correct_option, expert_explanation]
    )

    gr.HTML("""
    <style>
        .container {
            max-width: 100%;
            padding: 0.5rem;
        }
        
        /* Input styling */
        .mobile-input textarea {
            font-size: 1rem;
            padding: 0.75rem;
            border-radius: 0.5rem;
            min-height: 2.5rem;
        }
        
        /* Button styling */
        .mobile-button {
            width: 100%;
            margin: 0.5rem 0;
            padding: 0.75rem;
            font-size: 1rem;
            font-weight: 500;
        }
        
        .response-box {
            font-family: 'Inter', sans-serif;
            line-height: 1.6;
        }
        .response-box textarea {
            font-size: 1rem;
            padding: 1rem;
            border-radius: 0.5rem;
        }
        
        /* Mobile-specific adjustments */
        @media (max-width: 768px) {
            .gr-form {
                padding: 0.75rem;
            }
            .gr-box {
                margin: 0.5rem 0;
            }
            .gr-button {
                min-height: 2.5rem;
            }
            .gr-accordion {
                margin: 0.5rem 0;
            }
            .gr-input {
                margin-bottom: 0.5rem;
            }
        }
        
        /* Dark mode support */
        @media (prefers-color-scheme: dark) {
            .gr-box {
                background-color: #1a1a1a;
            }
            .mobile-input textarea,
            .response-box textarea {
                background-color: #2a2a2a;
                color: #ffffff;
            }
        }
    </style>
    """)

# Launch the app
if __name__ == "__main__":
    demo.launch(share=False)