app.py

import gradio as gr
from gradio_leaderboard import Leaderboard, SelectColumns

from src.display.columns import (
    MODEL_COL,
    TEAM_COL,
    main_board_datatype_for_df,
    main_board_default_cols,
    main_board_hidden_columns,
)
from src.display.css_html_js import custom_css
from src.display.datasets_tab import build_datasets_tab
from src.display.i18n import t
from src.display.utils import BENCHMARK_COLS, COLS, AutoEvalColumn
from src.leaderboard.sync import log_storage_status, sync_results_from_hub
from src.populate import (
    get_empty_leaderboard_notice,
    get_leaderboard_df,
    reload_leaderboard_tables,
)
from src.submission.submit import add_zip_submission

log_storage_status()
sync_results_from_hub()
LEADERBOARD_DF = get_leaderboard_df(COLS, BENCHMARK_COLS)


def build_leaderboard(dataframe):
    return Leaderboard(
        value=dataframe,
        datatype=main_board_datatype_for_df(dataframe),
        select_columns=SelectColumns(
            default_selection=main_board_default_cols(),
            cant_deselect=[AutoEvalColumn.model_type_symbol.name, MODEL_COL, TEAM_COL],
            label=t("select_columns"),
        ),
        search_columns=[MODEL_COL, TEAM_COL],
        hide_columns=main_board_hidden_columns(),
        interactive=False,
    )


demo = gr.Blocks(css=custom_css)
with demo:
    title = gr.HTML(t("title"))
    introduction = gr.Markdown(t("introduction"), elem_classes="markdown-text")

    with gr.Tabs(elem_classes="tab-buttons") as tabs:
        with gr.TabItem(t("tab_leaderboard"), elem_id="llm-benchmark-tab-table", id=0):
            empty_notice = gr.Markdown(
                get_empty_leaderboard_notice() if LEADERBOARD_DF.empty else "",
                visible=LEADERBOARD_DF.empty,
                elem_classes="markdown-text",
            )
            leaderboard = build_leaderboard(LEADERBOARD_DF)

        dataset_tables = build_datasets_tab()

        with gr.TabItem(t("tab_about"), elem_id="llm-benchmark-tab-about", id=2):
            gr.Markdown(t("about"), elem_classes="markdown-text")

        with gr.TabItem(t("tab_submit"), elem_id="llm-benchmark-tab-submit", id=3):
            gr.Markdown(t("submit"), elem_classes="markdown-text")
            zip_model_name_textbox = gr.Textbox(
                label=t("model_name_label"),
                placeholder=t("model_name_placeholder"),
            )
            zip_team_name_textbox = gr.Textbox(
                label=t("team_name_label"),
                placeholder=t("team_name_placeholder"),
            )
            submission_zip = gr.File(
                label=t("submission_zip_label"),
                file_types=[".zip"],
                type="filepath",
            )
            zip_submit_button = gr.Button(t("submit_button"))
            zip_submission_result = gr.Markdown()
            zip_submit_button.click(
                add_zip_submission,
                [submission_zip, zip_model_name_textbox, zip_team_name_textbox],
                [
                    zip_submission_result,
                    leaderboard,
                    *dataset_tables,
                    empty_notice,
                    zip_model_name_textbox,
                    zip_team_name_textbox,
                    submission_zip,
                ],
            )

    demo.load(
        reload_leaderboard_tables,
        outputs=[leaderboard, *dataset_tables, empty_notice],
    )

demo.queue(default_concurrency_limit=40).launch()