app.py

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

from src.about import TITLE
from src.display.css_html_js import custom_css
from src.display.utils import COLS, AutoEvalColumn, fields
from src.envs import API, EVAL_REQUESTS_PATH, EVAL_RESULTS_PATH, QUEUE_REPO, REPO_ID, RESULTS_REPO, TOKEN
from src.populate import get_leaderboard_df

import threading
import time

def restart_space():
    time.sleep(30 * 60)  # 30 minutes
    os._exit(0)  # Force exit, HF will restart the space

# Start restart timer in background
threading.Thread(target=restart_space, daemon=True).start()


def restart_space():
    API.restart_space(repo_id=REPO_ID)


def get_best_per_team(df: pd.DataFrame) -> pd.DataFrame:
    """Get the best (max global_score) row for each team."""
    if "team" not in df.columns or "global_score" not in df.columns:
        return df  # fallback
    return df.sort_values("global_score", ascending=False).groupby("team", as_index=False).first()


def init_leaderboard(dataframe):
    field_names = [f.name for f in AutoEvalColumn.__dataclass_fields__.values()]
    datatypes = [getattr(AutoEvalColumn, f).type for f in field_names]
    hide_columns = [f for f in field_names if getattr(AutoEvalColumn, f).hidden]
    filter_columns=[
            ColumnFilter("Team", type="dropdown", label="Filter by Team"),
            ColumnFilter("Submitter", type="dropdown", label="Filter by Submitter"),
        ]
    return Leaderboard(
        value=dataframe,
        datatype=datatypes,
        search_columns=[
            AutoEvalColumn.team.name,
        ],
        # hide_columns=hide_columns,
        filter_columns=filter_columns,
        interactive=True,
    )


### Download Data
try:
    snapshot_download(
        repo_id=QUEUE_REPO, local_dir=EVAL_REQUESTS_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30, token=TOKEN
    )
except Exception:
    restart_space()

try:
    snapshot_download(
        repo_id=RESULTS_REPO, local_dir=EVAL_RESULTS_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30, token=TOKEN
    )
except Exception:
    restart_space()


# Initial Data
FULL_DF = get_leaderboard_df(EVAL_RESULTS_PATH, EVAL_REQUESTS_PATH, COLS)

def filter_best_submissions(show_best):
    if show_best:
        # For each team, get the row with highest global_score
        filtered_df = FULL_DF.loc[FULL_DF.groupby("Team")["Global Score"].idxmax()]
    else:
        filtered_df = FULL_DF
    return filtered_df

demo = gr.Blocks(css=custom_css)
with demo:
    gr.HTML(TITLE)

    with gr.Tabs(elem_classes="tab-buttons"):
        with gr.TabItem("🏅 E2LMC Leaderboard", elem_id="llm-benchmark-tab-table", id=0):
            show_best_checkbox = gr.Checkbox(label="Show only best submission per team", value=False)
            leaderboard_container = gr.Column()

            # Initialize leaderboard
            with leaderboard_container:
                # leaderboard_component = init_leaderboard(get_best_per_team(FULL_DF))
                leaderboard_component = init_leaderboard(filter_best_submissions(show_best_checkbox.value))


            # Update leaderboard on checkbox change
            show_best_checkbox.change(
                fn=filter_best_submissions,
                inputs=[show_best_checkbox],
                outputs=[leaderboard_component],
                queue=False,
            )


# Scheduler
scheduler = BackgroundScheduler()
scheduler.add_job(restart_space, "interval", seconds=1800)
scheduler.start()
demo.queue(default_concurrency_limit=40).launch()