app.py

"""
DABstep Leaderboard - Genesis Edition
A self-contained leaderboard for the DABstep benchmark.

Original source: https://huggingface.co/spaces/adyen/DABstep
"""
import os

import gradio as gr

from dabstep_benchmark.content import (
    CITATION_BUTTON_LABEL,
    CITATION_BUTTON_TEXT,
    INTRODUCTION_TEXT,
    SUBMISSION_TEXT,
    TEMPORARY_NOTICE,
    TITLE,
    VALIDATION_GUIDELINES,
)
from dabstep_benchmark.leaderboard import (
    generate_leaderboard_df,
    process_submission,
    refresh,
    sync_from_hf_dataset,
)


def download_leaderboard(lb_type: str) -> str:
    """Download the leaderboard as CSV."""
    validated_lb, unvalidated_lb = generate_leaderboard_df()
    
    if lb_type == "validated":
        df_to_download = validated_lb
    else:
        df_to_download = unvalidated_lb
    
    os.makedirs("data", exist_ok=True)
    path = f"data/{lb_type}_leaderboard.csv"
    
    if os.path.exists(path):
        os.remove(path)
    
    df_to_download.to_csv(path, index=False)
    return path


# Custom CSS for better styling
CUSTOM_CSS = """
.markdown-text {
    font-size: 16px;
}
.gradio-container {
    max-width: 1200px !important;
}
#citation-button {
    font-family: monospace;
    font-size: 12px;
}
.notice-box {
    background: transparent !important;
    border: none !important;
    padding: 0 !important;
    margin: 8px 0 16px 0 !important;
}
.notice-box p {
    margin: 0 !important;
    padding: 12px 16px !important;
    background: #2a2a2a !important;
    border: 1px solid #444 !important;
    border-radius: 6px !important;
    color: #ccc !important;
    font-size: 14px !important;
    line-height: 1.5 !important;
}
.notice-box a {
    color: #6cb6ff !important;
}
"""

if __name__ == "__main__":
    # Ensure data directories exist
    os.makedirs("data/task_scores", exist_ok=True)
    os.makedirs("data/submissions", exist_ok=True)
    
    # Sync latest data from HF Dataset repo on startup
    print("===== Application Startup at", __import__("datetime").datetime.now().strftime("%Y-%m-%d %H:%M:%S"), "=====")
    sync_from_hf_dataset()
    
    # Load data once at startup (cached for subsequent calls)
    validated_lb, unvalidated_lb = generate_leaderboard_df()
    
    # Build the Gradio app
    demo = gr.Blocks(title="DABstep Leaderboard - Temporary Mirror")
    
    with demo:
        gr.HTML(f"<style>{CUSTOM_CSS}</style>")
        gr.Markdown(TITLE)
        gr.Markdown(TEMPORARY_NOTICE, elem_classes="notice-box")
        gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")
        
        with gr.Tab("📊 Leaderboard"):
            with gr.Tab("Validated"):
                verified_table = gr.Dataframe(
                    value=validated_lb,
                    datatype=["markdown", "str", "str", "str", "markdown", "str", "str"],
                    interactive=False,
                    column_widths=["20%"],
                    wrap=True,
                )
                verified_download = gr.DownloadButton(
                    label="📥 Download Leaderboard CSV",
                    elem_id="download-verified-lb",
                )
            
            with gr.Tab("Unvalidated"):
                unverified_table = gr.Dataframe(
                    value=unvalidated_lb,
                    datatype=["markdown", "str", "str", "str", "markdown", "str", "str"],
                    interactive=False,
                    column_widths=["20%"],
                    wrap=True,
                )
                unverified_download = gr.DownloadButton(
                    label="📥 Download Full Leaderboard CSV",
                    elem_id="download-unverified-lb",
                )
            
            # Refresh button
            refresh_button = gr.Button("🔄 Refresh Leaderboard", variant="secondary")
            
            def do_refresh():
                """Clear cache and reload leaderboard data."""
                return refresh(only_leaderboard=True)
            
            refresh_button.click(
                fn=do_refresh,
                inputs=None,
                outputs=[verified_table, unverified_table],
            )
        
        # Download handlers
        verified_download.click(
            download_leaderboard,
            inputs=[gr.Textbox(value="validated", visible=False)],
            outputs=[verified_download]
        )
        unverified_download.click(
            download_leaderboard,
            inputs=[gr.Textbox(value="unvalidated", visible=False)],
            outputs=[unverified_download]
        )
        
        
        gr.Markdown(VALIDATION_GUIDELINES, elem_classes="markdown-text")
        
        with gr.Tab("📤 Submit"):
            gr.Markdown(SUBMISSION_TEXT, elem_classes="markdown-text")
            
            # Show login status
            with gr.Row():
                gr.Markdown("**Login with HuggingFace to submit:**")
                login_btn = gr.LoginButton(size="sm")
            
            with gr.Row():
                with gr.Column():
                    split = gr.Radio(["all"], value="all", label="Split", visible=False)
                    agent_name_textbox = gr.Textbox(
                        label="Agent Name",
                        placeholder="e.g., MyDataAgent-v1"
                    )
                    model_family_textbox = gr.Textbox(
                        label="Model Family",
                        placeholder="e.g., GPT-4, Claude, Llama"
                    )
                    repo_url_textbox = gr.Textbox(
                        label="Repository URL (optional)",
                        placeholder="https://github.com/..."
                    )
                
                with gr.Column():
                    organisation = gr.Textbox(
                        label="Organization",
                        placeholder="e.g., Genesis Computing"
                    )
                    mail = gr.Textbox(
                        label="Contact Email",
                        placeholder="your@email.com"
                    )
                    file_output = gr.File(
                        label="Upload Submission (.jsonl)",
                        file_types=[".jsonl", ".json"]
                    )
            
            with gr.Row():
                submit_button = gr.Button("🚀 Submit Answers", variant="primary")
            
            submission_result = gr.Markdown()
            
            def submit_with_user(split, agent_name, model_family, repo_url, file, org, mail, profile: gr.OAuthProfile | None):
                """Wrap submission to include HF username."""
                if profile is None:
                    return "⚠️ Please login with HuggingFace first to submit."
                
                # Append HF username to organization (use dash instead of pipe to pass validation)
                org_with_user = f"{org} - user {profile.username}"
                return process_submission(split, agent_name, model_family, repo_url, file, org_with_user, mail)
            
            submit_button.click(
                submit_with_user,
                inputs=[
                    split,
                    agent_name_textbox,
                    model_family_textbox,
                    repo_url_textbox,
                    file_output,
                    organisation,
                    mail
                ],
                outputs=submission_result,
            )
        
        with gr.Tab("📚 Citation"):
            with gr.Accordion("📙 How to Cite", open=True):
                citation_button = gr.Textbox(
                    value=CITATION_BUTTON_TEXT,
                    label=CITATION_BUTTON_LABEL,
                    lines=len(CITATION_BUTTON_TEXT.split("\n")),
                    elem_id="citation-button",
                )
        
        # Login at bottom (LoginButton shows logout when logged in)
        with gr.Row():
            gr.Markdown("---")
        with gr.Row():
            gr.LoginButton(size="sm")
    
    # Launch the app with HuggingFace OAuth
    demo.launch(
        server_name="0.0.0.0",
        server_port=7860,
        share=False,
        debug=True
    )