app.py

import gradio as gr
import pandas as pd
from apscheduler.schedulers.background import BackgroundScheduler
from gradio_leaderboard import ColumnFilter, Leaderboard, SelectColumns

from src.about import CITATION_BUTTON_LABEL, CITATION_BUTTON_TEXT, EVALUATION_QUEUE_TEXT, INTRODUCTION_TEXT, TITLE
from src.datamodel.data import F1Data
from src.display.css_html_js import custom_css
from src.display.formatting import styled_error
from src.display.utils import AutoEvalColumn, ModelType, fields
from src.envs import API, CODE_PROBLEMS_REPO, REPO_ID, RESULTS_REPO, SUBMISSIONS_REPO
from src.logger import get_logger
from src.populate import get_leaderboard_df
from src.submission.submit import add_new_solutions
from src.validation.validate import MAX_INPUT_LENGTH, MIN_INPUT_LENGTH, is_submission_file_valid, is_valid

logger = get_logger(__name__)

ENSURE_ALL_PRESENT = False  # TODO: Switch to True.
SPLIT = "warmup"  # TODO temp

lbdb = F1Data(
    cp_ds_name=CODE_PROBLEMS_REPO,
    sub_ds_name=SUBMISSIONS_REPO,
    res_ds_name=RESULTS_REPO,
    split=SPLIT,
)

leaderboard_df = None

logger.info("Initialized LBDB")


def restart_space():
    logger.info("Restarting space")
    API.restart_space(repo_id=REPO_ID)


def refresh_leaderboard_data():
    """Refresh the leaderboard data from the latest results"""
    global leaderboard_df
    try:
        logger.info("Loading leaderboard data...")
        new_leaderboard_df = get_leaderboard_df(RESULTS_REPO)

        if new_leaderboard_df is not None and not new_leaderboard_df.empty:
            logger.info("Leaderboard data refreshed successfully")
            leaderboard_df = new_leaderboard_df
        else:
            logger.warning("No new leaderboard data found")
            return None
    except Exception as e:
        logger.error(f"Error refreshing leaderboard data: {e}")
        return None


def init_leaderboard(dataframe: pd.DataFrame):

    if dataframe is None or dataframe.empty:
        raise ValueError("Leaderboard DataFrame is empty or None.")

    return Leaderboard(
        value=dataframe,
        datatype=[c.type for c in fields(AutoEvalColumn)],
        select_columns=SelectColumns(
            default_selection=[c.name for c in fields(AutoEvalColumn) if c.displayed_by_default],
            cant_deselect=[c.name for c in fields(AutoEvalColumn) if c.never_hidden],
            label="Select Columns to Display:",
        ),
        search_columns=[AutoEvalColumn.system.name, AutoEvalColumn.system_type.name],
        hide_columns=[c.name for c in fields(AutoEvalColumn) if c.hidden],
        filter_columns=[
            ColumnFilter(AutoEvalColumn.system_type.name, type="checkboxgroup", label="Model types"),
        ],
        bool_checkboxgroup_label="Hide models",
        interactive=False,
    )


blocks = gr.Blocks(css=custom_css)
with blocks:
    gr.Image(
        "assets/banner.png",
        interactive=False,
        show_label=False,
        show_download_button=False,
        container=False,
    )

    gr.HTML(
        """
        <style>
            body {
                background-color: #121212;
                color: white;
                margin: 0;  /* Reset browser default */
            }

            /* Outer container margin & spacing */
            .gradio-container {
                max-width: 1100px;
                margin: 2rem auto;        /* top/bottom spacing + horizontal centering */
                padding: 2rem;            /* inner spacing */
                background-color: rgba(0, 0, 0, 0.6);  /* optional: semi-transparent panel */
                border-radius: 12px;      /* rounded corners */
            }
        </style>
        """
    )

    gr.HTML(TITLE)
    gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")

    with gr.Tabs(elem_classes="tab-buttons") as tabs:
        with gr.TabItem("🏅 FormulaOne Leaderboard", elem_id="formulaone-leaderboar-tab-table", id=0):
            refresh_leaderboard_data()  # updates leaderboard_df
            assert leaderboard_df is not None
            leaderboard_component = init_leaderboard(leaderboard_df)

        with gr.TabItem("🚀 Submit here! ", elem_id="llm-benchmark-tab-table", id=2):
            logger.info("Tab submission")
            with gr.Column():
                with gr.Row():
                    gr.Markdown(EVALUATION_QUEUE_TEXT, elem_classes="markdown-text")

            with gr.Row():
                gr.Markdown("# ✉️✨ Submit your solutions here!", elem_classes="markdown-text")

            with gr.Row():
                with gr.Column():
                    system_name_textbox = gr.Textbox(label=AutoEvalColumn.system.name)
                    org_textbox = gr.Textbox(label=AutoEvalColumn.organization.name)
                    sys_type_dropdown = gr.Dropdown(
                        choices=[t.to_str(" ") for t in ModelType],
                        label=AutoEvalColumn.system_type.name,
                        multiselect=False,
                        value=ModelType.LLM.to_str(" "),
                        interactive=True,
                    )

                    submission_file = gr.File(label="JSONL solutions file", file_types=[".jsonl"])

            logger.info("Submit button")
            submit_button = gr.Button("Submit")
            submission_result = gr.Markdown()

            def add_solution_cbk(
                system_name: str,
                org: str,
                sys_type: str,
                submission_path: str,
            ):

                try:
                    # Validating the submission file.
                    if len(submission_path) == 0:
                        return styled_error("Please upload JSONL submission file.")

                    if not is_submission_file_valid(submission_path):
                        return styled_error("Failed to read JSONL submission file. Please try again later.")

                    # Validating all user-supplied arguments.
                    for val, val_name in [
                        (system_name, "System name"),
                        (org, "Organisation name"),
                        (sys_type, "System type"),
                    ]:
                        if len(val) == 0:
                            return styled_error(f"Please fill in the '{val_name}' field.")

                        if not is_valid(val):
                            return styled_error(
                                f"{val_name} is invalid! Must only contain characters [a-zA-Z0-9], spaces, "
                                + "or the special characters '-' and '.', and be of length between "
                                + f"{MIN_INPUT_LENGTH} and {MAX_INPUT_LENGTH}."
                            )
                except Exception:
                    logger.warning("Failed to process user submission", exc_info=True)
                    return styled_error("An error occurred. Please try again later.")  # Intentionally vague.

                return add_new_solutions(
                    lbdb,
                    system_name,
                    org,
                    sys_type,
                    submission_path,
                    ensure_all_present=ENSURE_ALL_PRESENT,
                )

            submit_button.click(
                add_solution_cbk,
                [
                    system_name_textbox,
                    org_textbox,
                    sys_type_dropdown,
                    submission_file,
                ],
                submission_result,
            )

    with gr.Row():
        logger.info("Citation")
        with gr.Accordion(CITATION_BUTTON_LABEL, open=False):
            gr.Code(
                value=CITATION_BUTTON_TEXT.strip(),
                elem_id="citation-block",
            )

    # UI refresh triggers latest data swap. The work already happened in the background - refresh_leaderboard_data().
    blocks.load(lambda: leaderboard_df, inputs=[], outputs=[leaderboard_component])


logger.info("Scheduler")
scheduler = BackgroundScheduler()
scheduler.add_job(restart_space, "interval", seconds=1800)
scheduler.add_job(refresh_leaderboard_data, "interval", seconds=120)
scheduler.start()
logger.info("Launch")
blocks.queue(default_concurrency_limit=40).launch()
logger.info("Done")