app.py

# -*- coding: utf-8 -*-
import gradio as gr
import pandas as pd
from apscheduler.schedulers.background import BackgroundScheduler
# Removed Hugging Face Hub imports as they are not needed for the simplified leaderboard

# --- Make sure these imports work relative to your file structure ---
try:
    # Assume these contain the *content* without excessive inline styling
    from src.about import (
        CITATION_BUTTON_LABEL,
        CITATION_BUTTON_TEXT,
        EVALUATION_QUEUE_TEXT, # Keep if used by commented-out submit tab
        INTRODUCTION_TEXT,
        LLM_BENCHMARKS_TEXT,
        TITLE, # Expected to have an ID like #main-leaderboard-title
    )
    # Import custom_css if it exists, otherwise it will be defined below
    try:
        from src.display.css_html_js import custom_css
    except ImportError:
        print("Warning: src.display.css_html_js not found. Starting with empty custom_css.")
        custom_css = "" # Start fresh if not found

    from src.envs import REPO_ID # Keep if needed for restart_space or other functions
    from src.submission.submit import add_new_eval # Keep if using the submit tab
    print("Successfully imported from src module.")
# Option 2: Placeholder definitions (REMOVE IF USING OPTION 1)
except ImportError:
    print("Warning: Using placeholder values because src module imports failed.")
    CITATION_BUTTON_LABEL="Citation"
    CITATION_BUTTON_TEXT="Please cite us if you use this benchmark..."
    EVALUATION_QUEUE_TEXT="Current evaluation queue:"
    # Example placeholders with structure for CSS
    TITLE="""<h1 id="main-leaderboard-title" align="center">🏆 MLE-Dojo Benchmark Leaderboard (Placeholder)</h1>"""
    INTRODUCTION_TEXT="""
    <div class="introduction-section">
     <p>Welcome to the MLE-Dojo Benchmark Leaderboard (Placeholder Content).</p>
     <p>Edit <code>src/about.py</code> to set your actual title and introduction text.</p>
    </div>
    """
    LLM_BENCHMARKS_TEXT="""
    ## About Section (Placeholder)
    Information about the benchmarks will go here. Edit <code>src/about.py</code>.
    """
    custom_css="" # Start with empty CSS
    REPO_ID="your/space-id" # Replace with actual ID if needed
    def add_new_eval(*args): return "Submission placeholder."
# --- End Placeholder Definitions ---


# --- Elo Leaderboard Configuration ---
# (Keep your data definition as is)
data = [
    {'model_name': 'gpt-4o-mini', 'url': 'https://openai.com/index/gpt-4o-mini-advancing-cost-efficient-intelligence/', 'organizer': 'OpenAI', 'license': 'Proprietary', 'MLE-Lite_Elo': 753, 'Tabular_Elo': 839, 'NLP_Elo': 758, 'CV_Elo': 754, 'Overall': 778},
    {'model_name': 'gpt-4o', 'url': 'https://openai.com/index/hello-gpt-4o/', 'organizer': 'OpenAI', 'license': 'Proprietary', 'MLE-Lite_Elo': 830, 'Tabular_Elo': 861, 'NLP_Elo': 903, 'CV_Elo': 761, 'Overall': 841},
    {'model_name': 'o3-mini', 'url': 'https://openai.com/index/openai-o3-mini/', 'organizer': 'OpenAI', 'license': 'Proprietary', 'MLE-Lite_Elo': 1108, 'Tabular_Elo': 1019, 'NLP_Elo': 1056, 'CV_Elo': 1207, 'Overall': 1096}, # Fill details later
    {'model_name': 'deepseek-v3', 'url': 'https://api-docs.deepseek.com/news/news1226', 'organizer': 'DeepSeek', 'license': 'DeepSeek', 'MLE-Lite_Elo': 1004, 'Tabular_Elo': 1015, 'NLP_Elo': 1028, 'CV_Elo': 1067, 'Overall': 1023},
    {'model_name': 'deepseek-r1', 'url': 'https://api-docs.deepseek.com/news/news250120', 'organizer': 'DeepSeek', 'license': 'DeepSeek', 'MLE-Lite_Elo': 1137, 'Tabular_Elo': 1053, 'NLP_Elo': 1103, 'CV_Elo': 1083, 'Overall': 1100},
    {'model_name': 'gemini-2.0-flash', 'url': 'https://ai.google.dev/gemini-api/docs/models#gemini-2.0-flash', 'organizer': 'Google', 'license': 'Proprietary', 'MLE-Lite_Elo': 847, 'Tabular_Elo': 923, 'NLP_Elo': 860, 'CV_Elo': 978, 'Overall': 895},
    {'model_name': 'gemini-2.0-pro', 'url': 'https://blog.google/technology/google-deepmind/gemini-model-updates-february-2025/', 'organizer': 'Google', 'license': 'Proprietary', 'MLE-Lite_Elo': 1064, 'Tabular_Elo': 1139, 'NLP_Elo': 1028, 'CV_Elo': 973, 'Overall': 1054},
    {'model_name': 'gemini-2.5-pro', 'url': 'https://deepmind.google/technologies/gemini/pro/', 'organizer': 'Google', 'license': 'Proprietary', 'MLE-Lite_Elo': 1257, 'Tabular_Elo': 1150, 'NLP_Elo': 1266, 'CV_Elo': 1177, 'Overall': 1214},
]
master_df = pd.DataFrame(data)
CATEGORIES = ["Overall", "MLE-Lite", "Tabular", "NLP", "CV"]
DEFAULT_CATEGORY = "Overall"
category_to_column = {
    "MLE-Lite": "MLE-Lite_Elo", "Tabular": "Tabular_Elo",
    "NLP": "NLP_Elo", "CV": "CV_Elo", "Overall": "Overall"
}

# --- Helper function to update leaderboard ---
def update_leaderboard(category):
    """
    Selects relevant columns, sorts by the chosen category's Elo score,
    adds Rank, formats model name as a link, and returns the DataFrame.
    """
    score_column = category_to_column.get(category)
    if score_column is None or score_column not in master_df.columns:
        print(f"Warning: Invalid category '{category}' or column '{score_column}'. Falling back to default.")
        score_column = category_to_column[DEFAULT_CATEGORY]
        if score_column not in master_df.columns:
            print(f"Error: Default column '{score_column}' also not found.")
            # Return empty df with desired display columns
            return pd.DataFrame({
                "Rank": [], "Model": [], "Organizer": [], "License": [], "Elo Score": []
            })

    cols_to_select = ['model_name', 'url', 'organizer', 'license', score_column]
    df = master_df[cols_to_select].copy()
    df.sort_values(by=score_column, ascending=False, inplace=True)
    df.reset_index(drop=True, inplace=True)
    df.insert(0, 'Rank', df.index + 1)

    # Format Model Name as HTML Hyperlink - use a CSS class for styling
    df['Model'] = df.apply(
        lambda row: f"<a href='{row['url'] if pd.notna(row['url']) else '#'}' target='_blank' class='model-link'>{row['model_name']}</a>",
        axis=1
    )

    # Rename columns for final display
    df.rename(columns={score_column: 'Elo Score', 'organizer': 'Organizer', 'license': 'License'}, inplace=True)
    final_columns = ["Rank", "Model", "Organizer", "License", "Elo Score"]
    df = df[final_columns]
    return df

# --- Mock/Placeholder functions/data for other tabs ---
print("Warning: Evaluation queue data fetching is disabled/mocked due to leaderboard changes.")
finished_eval_queue_df = pd.DataFrame(columns=["Model", "Status", "Requested", "Started"])
running_eval_queue_df = pd.DataFrame(columns=["Model", "Status", "Requested", "Started"])
pending_eval_queue_df = pd.DataFrame(columns=["Model", "Status", "Requested", "Started"])
EVAL_COLS = ["Model", "Status", "Requested", "Started"]
EVAL_TYPES = ["str", "str", "str", "str"]

# --- Keep restart function if relevant ---
def restart_space():
    print(f"Attempting to restart space: {REPO_ID}")
    # Replace with your actual space restart mechanism if needed

# --- Enhanced CSS Definition ---
# Define all styles here. Assumes TITLE has id="main-leaderboard-title"
# and INTRODUCTION_TEXT is wrapped in class="introduction-section" (or rendered by gr.Markdown).

enhanced_css = """
/* Base and Theme Overrides */
body {
    font-family: -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, Oxygen, Ubuntu, Cantarell, 'Open Sans', 'Helvetica Neue', sans-serif;
    font-size: 1.3em; /* Base font size */
    line-height: 1.6;
    background-color: #f8f9fa; /* Light background */
    color: #343a40; /* Default text color */
}

/* Container adjustments for better spacing */
.gradio-container {
    max-width: 1200px !important; /* Limit max width */
    margin: 0 auto !important; /* Center the container */
    padding: 2rem !important; /* Add padding around the whole app */
}

/* --- Title Styling --- */
/* Targets the h1 tag with the specific ID from src/about.py */
#main-leaderboard-title {
    font-size: 3.2em; /* Large title */
    font-weight: 700; /* Bolder */
    color: #212529; /* Darker color for title */
    text-align: center; /* Ensure centering */
    margin-bottom: 1.5rem; /* Space below title */
    padding-bottom: 0.5rem; /* Space within the element */
    border-bottom: 2px solid #dee2e6; /* Subtle underline */
}

/* --- Introduction Text Styling --- */
/* Targets the wrapper div or the markdown component */
.introduction-section p, .introduction-wrapper .prose p { /* Target paragraphs within the section */
    font-family: 'Georgia',
    font-size: 1.5em; !important; /* Slightly larger than base */
    color: #495057; /* Slightly lighter text color */
    margin-bottom: 1rem; /* Space between paragraphs */
    max-width: 900px; /* Limit width for readability */
    margin-left: auto;  /* Center the text block */
    margin-right: auto; /* Center the text block */
    text-align: center; /* Center align intro text */
}
.introduction-section, .introduction-wrapper {
    font-family: 'Georgia',
    font-size: 1.5em; !important; /* Slightly larger than base */
    margin-bottom: 2.5rem; /* Space below the intro block */
}


/* --- General Markdown and Header Styling --- */
.markdown-text h2, .tabitem .prose h2 { /* Target section headers */
    font-size: 1.8em;
    font-weight: 500;
    color: #343a40;
    margin-top: 2.5rem; /* More space above sections */
    margin-bottom: 1.2rem;
    padding-bottom: 0.4rem;
    border-bottom: 1px solid #e9ecef;
}
.markdown-text p, .tabitem .prose p {
    font-size: 1.5em; /* Standard paragraph size */
    margin-bottom: 1rem;
    color: #495057;
}
.markdown-text a, .tabitem .prose a { /* Style links within markdown */
    font-size: 1.3em;
    color: #007bff;
    text-decoration: none;
}
.markdown-text a:hover, .tabitem .prose a:hover {
    font-size: 1.3em;
    text-decoration: underline;
}

/* --- Tab Styling --- */
.tab-buttons button { /* Style tab buttons */
    font-size: 1.3em !important;
    padding: 5px 10px !important;
    font-weight: 500;
}

/* --- Leaderboard Table Styling --- */
#leaderboard-table {
    margin-top: 1.5rem; /* Space above table */
    font-size: 1.5em; /* Ensure table font size is consistent */
    border: 1px solid #dee2e6;
    box-shadow: 0 2px 4px rgba(0,0,0,0.05); /* Subtle shadow */
}
#leaderboard-table th {
    background-color: #e9ecef; /* Header background */
    font-size: 1.3em;
    font-weight: 500; /* Header font weight */
    padding: 10px 12px; /* Header padding */
    text-align: left;
    color: #495057;
    white-space: nowrap; /* Prevent header text wrapping */
}
#leaderboard-table td {
    font-size: 1.1em;
    padding: 8px 12px; /* Cell padding */
    border-bottom: 1px solid #e9ecef; /* Horizontal lines */
    vertical-align: middle; /* Center cell content vertically */
}
#leaderboard-table tr:nth-child(even) td {
    font-size: 1.1em;
    background-color: #f8f9fa; /* Zebra striping */
}
#leaderboard-table tr:hover td {
    font-size: 1.1em;
    background-color: #e2e6ea; /* Hover effect */
}
/* Style for the model links within the table */
#leaderboard-table .model-link {
    color: #0056b3; /* Slightly darker blue for links */
    font-size: 1.1em;
    font-weight: 500;
    text-decoration: none;
}
#leaderboard-table .model-link:hover {
    font-size: 1.1em;
    text-decoration: underline;
    color: #003d80;
}

/* --- Radio Button / Category Selector Styling --- */
.gradio-radio label span { /* Target the label text */
   font-size: 1.3em !important;
   font-weight: 500;
   color: #343a40;
}
.gradio-radio fieldset { /* Adjust spacing around radio buttons */
   margin-top: 0.5rem;
   margin-bottom: 1.5rem;
}
.gradio-radio fieldset label { /* Style individual radio choices */
    padding: 8px 12px !important;
}


/* --- Accordion Styling --- */
.gradio-accordion > button { /* Accordion header */
    font-size: 1.2em !important;
    font-weight: 600;
    padding: 12px 15px !important;
    background-color: #f1f3f5 !important;
    border-bottom: 1px solid #dee2e6 !important;
}
.gradio-accordion > div { /* Accordion content area */
    padding: 15px !important;
    border: 1px solid #dee2e6 !important;
    border-top: none !important;
}

/* --- Textbox/Button Styling (e.g., Citation) --- */
#citation-button textarea {
    font-family: 'Courier New', Courier, monospace; /* Monospace for code/citation */
    font-size: 0.95em !important;
    background-color: #e9ecef;
    color: #343a40;
}
#citation-button label span {
    font-weight: 600;
}

"""

# Combine any existing CSS with the new enhanced CSS
# Prioritize enhanced_css rules by placing it last or using more specific selectors
final_css = custom_css + "\n" + enhanced_css

# --- Gradio App Definition ---
# Use a theme for base styling and apply custom CSS overrides
demo = gr.Blocks(css=final_css, theme=gr.themes.Soft(
    # Optional: Customize theme variables if needed
    # primary_hue=gr.themes.colors.blue,
    # secondary_hue=gr.themes.colors.gray,
    # neutral_hue=gr.themes.colors.cool_gray,
))

with demo:
    # Render TITLE from src/about.py (expects <h1 id="main-leaderboard-title">...)
    gr.HTML(TITLE)

    # Render INTRODUCTION_TEXT from src/about.py
    # Add a wrapper class for CSS targeting if the text itself doesn't have one
    with gr.Row():
        gr.Markdown(INTRODUCTION_TEXT, elem_classes="introduction-wrapper") # Use this class for CSS

    with gr.Tabs(elem_classes="tab-buttons") as tabs:
        with gr.TabItem("🏅 MLE-Dojo Benchmark", elem_id="llm-benchmark-tab-table", id=0):
            with gr.Column():
                # Use standard Markdown for the section header, CSS will style it
                gr.Markdown("## Model Elo Rankings by Category", elem_classes="markdown-text")
                category_selector = gr.Radio(
                    choices=CATEGORIES,
                    label="Select Category:", # Label is styled via CSS
                    value=DEFAULT_CATEGORY,
                    interactive=True,
                    elem_classes="gradio-radio" # Add class for styling
                )
                leaderboard_df_component = gr.Dataframe(
                    value=update_leaderboard(DEFAULT_CATEGORY),
                    headers=["Rank", "Model", "Organizer", "License", "Elo Score"],
                    datatype=["number", "html", "str", "str", "number"],
                    interactive=False,
                    row_count=(len(master_df), "fixed"),
                    col_count=(5, "fixed"),
                    wrap=True,
                    elem_id="leaderboard-table" # Used for specific table CSS
                )
                category_selector.change(
                    fn=update_leaderboard,
                    inputs=category_selector,
                    outputs=leaderboard_df_component
                )

        with gr.TabItem("📝 About", elem_id="llm-benchmark-tab-about", id=1):
            # Render LLM_BENCHMARKS_TEXT using Markdown, styled by CSS
            gr.Markdown(LLM_BENCHMARKS_TEXT, elem_classes="markdown-text") # Apply standard markdown styling

        # --- Submit Tab (Keep commented out or uncomment and ensure imports/variables are defined) ---
        # with gr.TabItem("🚀 Submit here! ", elem_id="llm-benchmark-tab-submit", id=2):
        #     # ... (Your submission form code - apply elem_classes="markdown-text" to gr.Markdown) ...
        #     pass # Placeholder


    # --- Citation Row (at the bottom, outside Tabs) ---
    with gr.Accordion("📙 Citation", open=False, elem_classes="gradio-accordion"): # Add class
        citation_button = gr.Textbox(
            value=CITATION_BUTTON_TEXT,
            label=CITATION_BUTTON_LABEL,
            lines=8, # Adjusted lines slightly
            elem_id="citation-button", # Used for specific CSS
            show_copy_button=True,
        )

# --- Scheduler and Launch ---
if __name__ == "__main__":
    try:
        scheduler = BackgroundScheduler()
        if callable(restart_space):
            if REPO_ID and REPO_ID != "your/space-id":
                scheduler.add_job(restart_space, "interval", seconds=1800)
                scheduler.start()
                print("Scheduler started for space restart.")
            else:
                print("Warning: REPO_ID not set or is placeholder; space restart job not scheduled.")
        else:
            print("Warning: restart_space function not available; space restart job not scheduled.")
    except Exception as e:
        print(f"Failed to initialize or start scheduler: {e}")

    print("Launching Gradio App...")
    # demo.queue() # Consider adding queue() for better handling under load
    demo.launch()