app.py

import gradio as gr
import pandas as pd
import os
import re
from datetime import datetime
from huggingface_hub import hf_hub_download
from huggingface_hub import HfApi, HfFolder

LEADERBOARD_FILE = "leaderboard.csv"
GROUND_TRUTH_FILE = "ground_truth.csv"
LAST_UPDATED = datetime.now().strftime("%B %d, %Y")

# Ensure authentication and suppress warnings
os.environ["HF_HUB_DISABLE_SYMLINKS_WARNING"] = "1"
HF_TOKEN = os.getenv("HF_TOKEN")
if not HF_TOKEN:
    raise ValueError("HF_TOKEN environment variable is not set or invalid.")

def initialize_leaderboard_file():
    """
    Ensure the leaderboard file exists and has the correct headers.
    """
    if not os.path.exists(LEADERBOARD_FILE):
        pd.DataFrame(columns=[
            "Model Name", "Overall Accuracy", "Valid Accuracy",
            "Correct Predictions", "Total Questions", "Timestamp"
        ]).to_csv(LEADERBOARD_FILE, index=False)
    elif os.stat(LEADERBOARD_FILE).st_size == 0:
        pd.DataFrame(columns=[
            "Model Name", "Overall Accuracy", "Valid Accuracy",
            "Correct Predictions", "Total Questions", "Timestamp"
        ]).to_csv(LEADERBOARD_FILE, index=False)

def clean_answer(answer):
    if pd.isna(answer):
        return None
    answer = str(answer)
    clean = re.sub(r'[^A-Da-d]', '', answer)
    return clean[0].upper() if clean else None


def update_leaderboard(results):
    """
    Append new submission results to the leaderboard file and push updates to the Hugging Face repository.
    """
    new_entry = {
        "Model Name": results['model_name'],
        "Overall Accuracy": round(results['overall_accuracy'] * 100, 2),
        "Valid Accuracy": round(results['valid_accuracy'] * 100, 2),
        "Correct Predictions": results['correct_predictions'],
        "Total Questions": results['total_questions'],
        "Timestamp": datetime.now().strftime("%Y-%m-%d %H:%M:%S"),
    }

    try:
        # Update the local leaderboard file
        new_entry_df = pd.DataFrame([new_entry])
        file_exists = os.path.exists(LEADERBOARD_FILE)
        
        new_entry_df.to_csv(
            LEADERBOARD_FILE,
            mode='a',  # Append mode
            index=False,
            header=not file_exists  # Write header only if the file is new
        )
        print(f"Leaderboard updated successfully at {LEADERBOARD_FILE}")

        # Push the updated file to the Hugging Face repository using HTTP API
        api = HfApi()
        token = HfFolder.get_token()
        
        api.upload_file(
            path_or_fileobj=LEADERBOARD_FILE,
            path_in_repo="leaderboard.csv",
            repo_id="SondosMB/ss",  # Your Space repository
            repo_type="space",
            token=token
        )
        print("Leaderboard changes pushed to Hugging Face repository.")
        
    except Exception as e:
        print(f"Error updating leaderboard file: {e}")



def load_leaderboard():
    if not os.path.exists(LEADERBOARD_FILE) or os.stat(LEADERBOARD_FILE).st_size == 0:
        return pd.DataFrame({
            "Model Name": [],
            "Overall Accuracy": [],
            "Valid Accuracy": [],
            "Correct Predictions": [],
            "Total Questions": [],
            "Timestamp": [],
        })
    return pd.read_csv(LEADERBOARD_FILE)

def evaluate_predictions(prediction_file, model_name, add_to_leaderboard):
    try:
        ground_truth_path = hf_hub_download(
            repo_id="SondosMB/ground-truth-dataset",
            filename="ground_truth.csv",
            repo_type="dataset",
            use_auth_token=True
        )
        ground_truth_df = pd.read_csv(ground_truth_path)
    except FileNotFoundError:
        return "Ground truth file not found in the dataset repository.", load_leaderboard()
    except Exception as e:
        return f"Error loading ground truth: {e}", load_leaderboard()

    if not prediction_file:
        return "Prediction file not uploaded.", load_leaderboard()

    try:
        #load predition file
        predictions_df = pd.read_csv(prediction_file.name)
         # Validate required columns in prediction file
        required_columns = ['question_id', 'predicted_answer']
        missing_columns = [col for col in required_columns if col not in predictions_df.columns]
        if missing_columns:
            return (f"Error: Missing required columns in prediction file: {', '.join(missing_columns)}.",
                    load_leaderboard())

        # Validate 'Answer' column in ground truth file
        if 'Answer' not in ground_truth_df.columns:
            return "Error: 'Answer' column is missing in the ground truth dataset.", load_leaderboard()
        merged_df = pd.merge(predictions_df, ground_truth_df, on='question_id', how='inner')
        merged_df['pred_answer'] = merged_df['predicted_answer'].apply(clean_answer)

        valid_predictions = merged_df.dropna(subset=['pred_answer'])
        correct_predictions = (valid_predictions['pred_answer'] == valid_predictions['Answer']).sum()
        total_predictions = len(merged_df)
        total_valid_predictions = len(valid_predictions)

        overall_accuracy = correct_predictions / total_predictions if total_predictions > 0 else 0
        valid_accuracy = correct_predictions / total_valid_predictions if total_valid_predictions > 0 else 0

        results = {
            'model_name': model_name if model_name else "Unknown Model",
            'overall_accuracy': overall_accuracy,
            'valid_accuracy': valid_accuracy,
            'correct_predictions': correct_predictions,
            'total_questions': total_predictions,
        }

        if add_to_leaderboard:
            update_leaderboard(results)
            return "Evaluation completed and added to leaderboard.", load_leaderboard()
        else:
            return "Evaluation completed but not added to leaderboard.", load_leaderboard()
  
    except Exception as e:
        return f"Error during evaluation: {str(e)}", load_leaderboard()

initialize_leaderboard_file()

# Function to set default mode
# Function to set default mode
import gradio as gr

# # Custom CSS to match website style
# # Define CSS to match a modern, professional design
# # Define enhanced CSS for the entire layout

css_tech_theme = """
body {
    font-family: 'Roboto', sans-serif;
    background-color: #f4f6fa;
    color: #333333;
    margin: 0;
    padding: 0;
}

/* Header Styling */
header {
    text-align: center;
    padding: 60px 20px;
    background: linear-gradient(135deg, #6a1b9a, #64b5f6);
    color: #ffffff;
    border-radius: 12px;
    margin-bottom: 30px;
    box-shadow: 0 6px 20px rgba(0, 0, 0, 0.2);
}

header h1 {
    font-size: 3.5em;
    font-weight: bold;
    margin-bottom: 10px;
}

header h2 {
    font-size: 2em;
    margin-bottom: 15px;
}

header p {
    font-size: 1em;
    line-height: 1.8;
}

.header-buttons {
    display: flex;
    justify-content: center;
    gap: 15px;
    margin-top: 20px;
}

.header-buttons a {
    text-decoration: none;
    font-size: 1.5em;
    padding: 15px 30px;
    border-radius: 30px;
    font-weight: bold;
    background: #ffffff;
    color: #6a1b9a;
    transition: transform 0.3s, background 0.3s;
    box-shadow: 0 4px 10px rgba(0, 0, 0, 0.1);
}

.header-buttons a:hover {
    background: #64b5f6;
    color: #ffffff;
    transform: scale(1.05);
}

/* Pre-Tabs Section */



#pre-tabs{
    text-align: left !important;
    color:#6a1b9a
}

#pre-tabs h2 {
    font-size: 3em
    font-color:#6a1b9a
    margin-bottom: 15px;
}

#pre-tabs p {
    color: #555555;
    line-height: 1.5;
    font-size: 1.5em
}






/* Tabs Section */
.tabs {
    margin: 0 auto;
    padding: 20px;
    background: #ffffff;
    border-radius: 12px;
    box-shadow: 0 4px 15px rgba(0, 0, 0, 0.1);
    /* max-width: 1300px; /* change 1 */ */
}

/* Post-Tabs Section */
.post-tabs {
    text-align: center;
    padding: 40px 20px;
    background: linear-gradient(135deg, #64b5f6, #6a1b9a);
    color: #ffffff;
    border-radius: 12px;
    margin-top: 30px;
}

.post-tabs h2 {
color: blue;
    font-size: 3.4em;
    margin-bottom: 15px;
}

.post-tabs p {
    font-size: 2em;
    line-height: 1.8;
    margin-bottom: 20px;
}

.post-tabs a {
    text-decoration: none;
    font-size: 1.1em;
    padding: 15px 30px;
    border-radius: 30px;
    font-weight: bold;
    background: #ffffff;
    color: #6a1b9a;
    transition: transform 0.3s, background 0.3s;
    box-shadow: 0 4px 10px rgba(0, 0, 0, 0.1);
}

.post-tabs a:hover {
    background: #6a1b9a;
    color: #ffffff;
    transform: scale(1.05);
}

/* Footer */
#custom-footer {
    background: linear-gradient(135deg, #6a1b9a, #8e44ad);
    color: #ffffff;
    text-align: center;
    padding: 40px 20px;
    margin-top: 30px;
    border-radius: 12px;
    box-shadow: 0 4px 10px rgba(0, 0, 0, 0.2);
}

#custom-footer h2 {
    font-size: 1.5em;
    margin-bottom: 15px;
}

#custom-footer p {
    font-size: 0.8em;
    line-height: 1.6;
    margin-bottom: 20px;
}
/* Link Styling */
.social-links {
    display: flex;
    justify-content: center;
    gap: 15px; /* Space between links */
}

.social-link {
    display: inline-block;
    text-decoration: none;
    color: #ffffff;
    background-color: #6a1b9a; /* Purple button background */
    padding: 10px 20px;
    border-radius: 30px;
    font-size: 16px;
    font-weight: bold;
    transition: all 0.3s ease;
    box-shadow: 0 4px 10px rgba(0, 0, 0, 0.1);
}

.social-link:hover {
    background-color: #8c52d3; /* Darker shade on hover */
    box-shadow: 0 6px 15px rgba(0, 0, 0, 0.2);
    transform: translateY(-2px);
}

.social-link:active {
    transform: translateY(1px);
    box-shadow: 0 3px 8px rgba(0, 0, 0, 0.1);
}


#submission-buttons {
    display: flex;
    justify-content: center;
    gap: 15px;
    margin-top: 20px;
}



/* Buttons Styling */
#submission-buttons button {
    padding: 12px 25px;
    font-size: 1.1em;
    color: #ffffff;
    background: #6a1b9a;
    border: none;
    border-radius: 30px;
    cursor: pointer;
    font-weight: bold;
    transition: all 0.3s ease;
    box-shadow: 0 4px 12px rgba(0, 0, 0, 0.1);
}
#submission-buttons button:hover {
    background: #8c52d3; /* Slightly lighter purple */
    transform: scale(1.05);
    box-shadow: 0 6px 15px rgba(0, 0, 0, 0.2);
}
#submission-buttons button:active {
    background: #5e1287; /* Darker purple */
    transform: scale(0.98);
    box-shadow: 0 3px 10px rgba(0, 0, 0, 0.1);
}
.gradio-container {
    padding-bottom: 0 !important;
    margin-bottom: 0 !important;
}



/* overview */
#Overview {
    padding: 30px;
    border-radius: 12px;
    background: linear-gradient(135deg, #ffffff, #f9f9ff);
    box-shadow: 0 6px 15px rgba(0, 0, 0, 0.1);
    max-width: 900px;
    margin: 20px auto;
}
#Overview h2 {
    font-size: 2.8em;
    color: #6a1b9a  !important;
    text-align: center;
    margin-bottom: 20px;
}
#Overview p {
    font-size: 1.5em;
    color: #333333;
    line-height: 1.8;
    margin-bottom: 15px;
}
#Overview ul, #Overview ol {
    font-size: 1.2em;
    color: #555555;
    margin: 20px 0;
    padding-left: 40px;
}
#Overview ul li, #Overview ol li {
    margin-bottom: 10px;
}
#Overview ul li::marker, Overview ol li::marker {
    color: #6a1b9a;
    font-size: 1.5em;
}
Overview a {
    color: #6a1b9a;
    text-decoration: underline;
}
Overview a:hover {
    color: #8c52d3;
}
"""

# Create the Gradio Interface

with gr.Blocks(css=css_tech_theme) as demo:
    # Header Section
    gr.Markdown("""
    <header>
        <h1>🏆 Mobile-MMLU Challenge</h1>
        <h2>🚀 Pushing the Limits of Mobile LLMs</h2>
    </header>
    """)
#     # Pre-Tabs Section
    gr.Markdown("""
    <section class="pre-tabs">
    <h2>Why Participate?</h2>
    <p>
        The Mobile-MMLU Benchmark Competition offers a unique opportunity to evaluate your LLMs in real-world mobile scenarios. 
        Join the challenge to drive innovation, showcase your expertise, and shape the future of mobile AI.
    </p>
    </section>""", elem_id="pre-tabs")

    
 # Tabs Section
    with gr.Tabs(elem_id="tabs"):
        # Overview Tab
        with gr.TabItem("📖 Overview"):
            gr.Markdown( """
        <div class="tabs">
            <h2 style="color: #6a1b9a; text-align: center;">About the Competition</h2>
            <p>The <strong>Mobile-MMLU Benchmark Competition</strong> is a premier challenge designed to evaluate and advance mobile-optimized Large Language Models (LLMs). This competition is an excellent opportunity to showcase your model's ability to handle real-world scenarios and excel in mobile intelligence.</p>
            <p>With a dataset spanning <strong>80 distinct fields</strong> and featuring <strong>16,186 questions</strong>, the competition emphasizes practical applications, from education and healthcare to technology and daily life.</p>
            <h3 style="color: #8e44ad;">Why Compete?</h3>
            <p>Participating in this competition allows you to:</p>
            <ul>
                <li>🌟 Showcase your expertise in developing and optimizing LLMs for mobile platforms.</li>
                <li>🚀 Benchmark your model’s performance against others in a highly competitive environment.</li>
                <li>📈 Contribute to advancements in mobile AI, shaping the future of user-centric AI systems.</li>
            </ul>
            <h3 style="color: #6a1b9a;">How It Works</h3>
            <ol>
                <li>1️⃣ <strong>Download the Dataset:</strong> Access the dataset and detailed instructions on the <a href="https://github.com/your-github-repo" target="_blank">GitHub page</a>. Follow the steps to ensure your environment is set up correctly.</li>
                <li>2️⃣ <strong>Generate Predictions:</strong> Use the provided script in the GitHub repository to generate answers. Ensure the output file matches the format in the github </li>
                <li>3️⃣ <strong>Submit Predictions:</strong> Upload your CSV file to the <strong>Submission Page</strong> on this platform.</li>
                <li>4️⃣ <strong>Evaluation:</strong> Your submission will be scored based on accuracy. The results will include overall and valid accuracy metrics.</li>
                <li>5️⃣ <strong>Leaderboard:</strong> Optionally, add your results to the real-time leaderboard to compare your model's performance with others.</li>
            </ol>
            <h3 style="color: #8e44ad;">Resources</h3>
            <ul>
                <li>📂 <a href="https://github.com/your-github-repo" target="_blank">GitHub Repository</a>: Contains the dataset, scripts, and detailed instructions.</li>
                <li>📊 <a href="https://github.com/your-dataset-link" target="_blank">Dataset Link</a>: Direct access to the competition dataset.</li>
                <li>❓ <a href="https://github.com/your-github-repo/issues" target="_blank">Support Page</a>: Use this for queries or issues during participation.</li>
            </ul>
        </div>
        """)
            
        with gr.TabItem("📤 Submission"):
            gr.Markdown("""
            <div class="submission-section" style="border: 3px solid #6a1b9a; padding: 20px; border-radius: 12px; box-shadow: 0 4px 10px rgba(106, 27, 154, 0.2);">
            <h2 style="color: #6a1b9a; text-align: center;">Submit Your Predictions</h2>
            <p style="font-size: 1.2em; color: #333; text-align: center;">Upload your prediction file and provide your model name to evaluate and optionally submit your results to the leaderboard.</p>
            </div>
            """)
            with gr.Row(elem_id="submission-fields"):
                file_input = gr.File(label="📂 Upload Prediction CSV", file_types=[".csv"], interactive=True,scale=1, min_width=12000)
                model_name_input = gr.Textbox(label="🏷️ Model Name", placeholder="Enter your model name",scale=1,  min_width=800)
        
            with gr.Row(elem_id="submission-results"):
                overall_accuracy_display = gr.Number(label="📊 Overall Accuracy (%)", interactive=False,scale=1,min_width=1200)
        
            with gr.Row(elem_id="submission-buttons"):
                eval_button = gr.Button("📈 Evaluate",scale=1,min_width=1200)
                submit_button = gr.Button("📤 Prove and Submit to Leaderboard", elem_id="evaluation-status", visible=False,scale=1,min_width=1200)
                eval_status = gr.Textbox(label="🛠️ Evaluation Status", interactive=False,scale=1,min_width=1200)
        
        # Define the functions outside the `with` block
        def handle_evaluation(file, model_name):
            # Check if required inputs are provided
            if not file:
                return "Error: Please upload a prediction file.", 0, gr.update(visible=False)
            if not model_name or model_name.strip() == "":
                return "Error: Please enter a model name.", 0, gr.update(visible=False)
        
            try:
                # Load predictions file
                predictions_df = pd.read_csv(file.name)
        
                # Validate required columns in the prediction file
                required_columns = ['question_id', 'predicted_answer']
                missing_columns = [col for col in required_columns if col not in predictions_df.columns]
                if missing_columns:
                    return (f"Error: Missing required columns in prediction file: {', '.join(missing_columns)}.",
                            0, gr.update(visible=False))
        
                # Perform evaluation
                status, leaderboard = evaluate_predictions(file, model_name, add_to_leaderboard=False)
                if leaderboard.empty:
                    overall_accuracy = 0
                else:
                    overall_accuracy = leaderboard.iloc[-1]["Overall Accuracy"]
        
                # Show the submit button after successful evaluation
                return status, overall_accuracy, gr.update(visible=True)
        
            except Exception as e:
                # Handle unexpected errors
                return f"Error during evaluation: {str(e)}", 0, gr.update(visible=False)
        
        def handle_submission(file, model_name):
            # Handle leaderboard submission
            status, _ = evaluate_predictions(file, model_name, add_to_leaderboard=True)
            return f"Submission to leaderboard completed: {status}"
        
        # Connect button clicks to the functions
        eval_button.click(
            handle_evaluation,
            inputs=[file_input, model_name_input],
            outputs=[eval_status, overall_accuracy_display, submit_button],
        )
        
        submit_button.click(
            handle_submission,
            inputs=[file_input, model_name_input],
            outputs=[eval_status],
        )



        with gr.TabItem("🏅 Leaderboard"):
            leaderboard_table = gr.Dataframe(
                value=load_leaderboard(),
                label="Leaderboard",
                interactive=False,
                wrap=True,
            )
            refresh_button = gr.Button("Refresh Leaderboard")
            refresh_button.click(
                lambda: load_leaderboard(),
                inputs=[],
                outputs=[leaderboard_table],
            )

     # Post-Tabs Section
    gr.Markdown("""
    <section class="post-tabs">
        <h2>Ready to Compete?</h2>
        <h3>
            Submit your predictions today and make your mark in advancing mobile AI technologies. 
            Show the world what your model can achieve!
        <h3>
    </section>
    """)

    # Footer Section
    gr.Markdown("""
    <footer  id="custom-footer">
        <h2>Stay Connected</h2>
        <p>
            Follow us on social media or contact us for any queries. Let's shape the future of AI together!
        </p>
        <div class="social-links">
        <a href="https://website.com" target="_blank" class="social-link">🌐 Website</a>
        <a href="https://github.com/VILA-Lab/Mobile-MMLU" target="_blank" class="social-link">🐙 GitHub</a>
    </div>
    </footer>
    """)

demo.launch()