app.py

import gradio as gr
from gradio_leaderboard import Leaderboard, ColumnFilter, SelectColumns
import pandas as pd
from apscheduler.schedulers.background import BackgroundScheduler
from huggingface_hub import snapshot_download
from src.about import CITATION_BUTTON_LABEL, CITATION_BUTTON_TEXT, EVALUATION_QUEUE_TEXT, INTRODUCTION_TEXT, LLM_BENCHMARKS_TEXT, TITLE
from src.tasks import TASK_DESCRIPTIONS, MEASURE_DESCRIPTION
from src.display.css_html_js import custom_css
from src.display.utils import BENCHMARK_COLS, COLS, EVAL_COLS, EVAL_TYPES, AutoEvalColumn, ModelType, fields, WeightType, Precision
from src.envs import API, EVAL_REQUESTS_PATH, EVAL_RESULTS_PATH, QUEUE_REPO, REPO_ID, RESULTS_REPO, TOKEN
from src.populate import get_evaluation_queue_df, get_leaderboard_df
from src.submission.submit import add_new_eval
import random

# Define task metadata (icons, names, descriptions)
TASK_METADATA_MULTIPLECHOICE = {
    "TE": {"icon": "📊", "name": "Textual Entailment", "tooltip": ""},
    "SA": {"icon": "😃", "name": "Sentiment Analysis", "tooltip": ""},
    "HS": {"icon": "⚠️", "name": "Hate Speech", "tooltip": ""},
    "AT": {"icon": "🏥", "name": "Admission Test", "tooltip": ""},
    "WIC": {"icon": "🔤", "name": "Word in Context", "tooltip": ""},
    "FAQ": {"icon": "❓", "name": "Frequently Asked Questions", "tooltip": ""}
}

# Define task metadata (icons, names, descriptions)
TASK_METADATA_GENERATIVE = {
    "LS": {"icon": "🔄", "name": "Lexical Substitution", "tooltip": ""},
    "SU": {"icon": "📝", "name": "Summarization", "tooltip": ""},
    "NER": {"icon": "🏷️", "name": "Named Entity Recognition", "tooltip": ""},
    "REL": {"icon": "🔗", "name": "Relation Extraction", "tooltip": ""},
}

def restart_space():
    """Restart the Hugging Face space."""
    API.restart_space(repo_id=REPO_ID)


def init_leaderboard(dataframe, default_selection=None, hidden_columns=None):
    """

    Initialize and return the leaderboard when it is first loaded or when 'benchmark' is selected.

    The table is sorted based on the "Avg. Combined Performance" field.

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

    field_list = fields(AutoEvalColumn)

    return Leaderboard(
        value=dataframe,
        datatype=[c.type for c in field_list],
        #select_columns=SelectColumns(
        #    default_selection=default_selection or [c.name for c in field_list if c.displayed_by_default],
        #    cant_deselect=[c.name for c in field_list if c.never_hidden],
        #    label="Select Columns to Display:",
        #),
        search_columns=[AutoEvalColumn.model.name, AutoEvalColumn.license.name],
        hide_columns=hidden_columns or [c.name for c in field_list if c.hidden],
        filter_columns=[
            ColumnFilter(AutoEvalColumn.fewshot_symbol.name, type="checkboxgroup", label="N-Few-Shot Learning (FS)"),
            #ColumnFilter(AutoEvalColumn.fewshot_symbol.name, type="checkboxgroup", label="N-Few-Shot Learning (FS)",
            #             default=[["0️⃣", "0️⃣"]]),
        #   ColumnFilter(AutoEvalColumn.params.name, type="slider", min=0, max=150, label="Select the number of parameters (B)"),
        ],
        #filter_columns=[
        #    ColumnFilter("IS_FS", type="checkbox", default=False, label="5-Few-Shot")
        #    #ColumnFilter("FS", type="dropdown", label="5-Few-Shot")
        #],
        bool_checkboxgroup_label="Evaluation Mode",
        interactive=False,
    )

def update_task_leaderboard(dataframe, default_selection=None, hidden_columns=None):
    """

    Update and return the leaderboard when a specific task is selected.

    The table is sorted based on the "Combined Performance" field.

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

    sorted_dataframe = dataframe.sort_values(by="Combined Performance", ascending=False)

    #print(sorted_dataframe['Combined Performance'])

    field_list = fields(AutoEvalColumn)

    return Leaderboard(
        value=sorted_dataframe,
        datatype=[c.type for c in field_list],
        #select_columns=SelectColumns(
        #    default_selection=default_selection or [c.name for c in field_list if c.displayed_by_default],
        #    cant_deselect=[c.name for c in field_list if c.never_hidden],
        #    label="Select Columns to Display:",
        #),
        search_columns=[AutoEvalColumn.model.name, AutoEvalColumn.license.name],
        hide_columns=hidden_columns or [c.name for c in field_list if c.hidden],
        filter_columns=[
            ColumnFilter(AutoEvalColumn.fewshot_symbol.name, type="checkboxgroup", label="N-Few-Shot Learning (FS)"),
        ],
        bool_checkboxgroup_label="Evaluation Mode",
        interactive=False
    )

'''

# Helper function for leaderboard initialization

def init_leaderboard(dataframe, default_selection=None, hidden_columns=None):

    """Initialize and return a leaderboard."""

    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=default_selection or [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.model.name, AutoEvalColumn.license.name],

        hide_columns=hidden_columns or [c.name for c in fields(AutoEvalColumn) if c.hidden],

        filter_columns=[

            ColumnFilter(AutoEvalColumn.fewshot_type.name, type="checkboxgroup", label="N-Few-Shot Learning (FS)"),

            ColumnFilter(AutoEvalColumn.params.name, type="slider", min=0, max=150, label="Select the number of parameters (B)"),

        ],

        bool_checkboxgroup_label="Hide models",

        interactive=False,

    )

'''

def download_snapshot(repo, local_dir):
    """Try to download a snapshot from Hugging Face Hub."""
    try:
        print(f"Downloading from {repo} to {local_dir}...")
        snapshot_download(repo_id=repo, local_dir=local_dir, repo_type="dataset", tqdm_class=None, etag_timeout=30, token=TOKEN)
    except Exception as e:
        print(f"Error downloading {repo}: {e}")
        restart_space()


# Initialize the app by downloading snapshots
download_snapshot(QUEUE_REPO, EVAL_REQUESTS_PATH)
download_snapshot(RESULTS_REPO, EVAL_RESULTS_PATH)

# Load leaderboard data
LEADERBOARD_DF = get_leaderboard_df(EVAL_RESULTS_PATH, EVAL_REQUESTS_PATH, COLS, BENCHMARK_COLS)
finished_eval_queue_df, running_eval_queue_df, pending_eval_queue_df = get_evaluation_queue_df(EVAL_REQUESTS_PATH, EVAL_COLS)

# Prepare the main interface
demo = gr.Blocks(css=custom_css)
with demo:
    gr.HTML(TITLE)
    gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")

    with gr.Tabs(elem_classes="tab-buttons") as tabs:

        # Main leaderboard tab
        with gr.TabItem("🏅 Benchmark"):

            leaderboard = init_leaderboard(
                LEADERBOARD_DF,
                default_selection=['FS', 'Model', "Avg. Combined Performance ⬆️", "TE", "SA", "HS", "AT", "WIC", "FAQ", "LS", "SU", "NER", "REL"],
                hidden_columns=[col for col in LEADERBOARD_DF.columns if col not in ['FS', 'Model', "Avg. Combined Performance ⬆️", "TE", "SA", "HS", "AT", "WIC", "FAQ", "LS", "SU", "NER", "REL"]]
            )

        # About tab
        with gr.TabItem("📝 About"):
            gr.Markdown(LLM_BENCHMARKS_TEXT, elem_classes="markdown-text")

        # About tab
        with gr.TabItem("║", interactive=False):
            gr.Markdown("", elem_classes="markdown-text")

        # Task-specific leaderboards
        for task, metadata in TASK_METADATA_MULTIPLECHOICE.items():

            with gr.TabItem(f"{metadata['icon']}{task}"):

                task_description = TASK_DESCRIPTIONS.get(task, "Description not available.")
                gr.Markdown(task_description, elem_classes="markdown-text")

                leaderboard = update_task_leaderboard(
                    LEADERBOARD_DF.rename(columns={f"{task} Prompt Average": "Prompt Average", f"{task} Best Prompt": "Best Prompt", f"{task} Best Prompt Id": "Best Prompt Id", task: "Combined Performance"}),
                    default_selection=['FS', 'Model', 'Combined Performance', 'Prompt Average', 'Best Prompt', 'Best Prompt Id'],
                    hidden_columns=[col for col in LEADERBOARD_DF.columns if col not in ['FS', 'Model', 'Combined Performance', 'Prompt Average', 'Best Prompt', 'Best Prompt Id']]
                )

        # About tab
        with gr.TabItem("│", interactive=False):
            gr.Markdown("", elem_classes="markdown-text")

        # Task-specific leaderboards
        for task, metadata in TASK_METADATA_GENERATIVE.items():
            with gr.TabItem(f"{metadata['icon']}{task}"):
                task_description = TASK_DESCRIPTIONS.get(task, "Description not available.")
                gr.Markdown(task_description, elem_classes="markdown-text")

                leaderboard = update_task_leaderboard(
                    LEADERBOARD_DF.rename(columns={f"{task} Prompt Average": "Prompt Average",
                                                   f"{task} Best Prompt": "Best Prompt",
                                                   f"{task} Best Prompt Id": "Best Prompt Id",
                                                   task: "Combined Performance"}),
                    default_selection=['FS', 'Model', 'Combined Performance', 'Prompt Average', 'Best Prompt',
                                       'Best Prompt Id'],
                    hidden_columns=[col for col in LEADERBOARD_DF.columns if
                                    col not in ['FS', 'Model', 'Combined Performance', 'Prompt Average',
                                                'Best Prompt', 'Best Prompt Id']]
                )

    # Citation section
    with gr.Accordion("📙 Citation", open=False):
        gr.Textbox(value=CITATION_BUTTON_TEXT, label=CITATION_BUTTON_LABEL, lines=20, elem_id="citation-button", show_copy_button=True)

# Background job to restart space
scheduler = BackgroundScheduler()
scheduler.add_job(restart_space, "interval", seconds=1800)
scheduler.start()

# Launch the app with concurrent queueing
demo.queue(default_concurrency_limit=40).launch(debug=True,  # Enable Gradio debug mode
        show_error=True)