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 gradio.components.textbox import Textbox
from gradio.components.dataframe import Dataframe
from gradio.components.checkboxgroup import CheckboxGroup

# from fastchat.serve.monitor.monitor import build_leaderboard_tab, build_basic_stats_tab, basic_component_values, leader_component_values

from src.about import (
    CITATION_BUTTON_LABEL,
    CITATION_BUTTON_TEXT,
    EVALUATION_QUEUE_TEXT,
    INTRODUCTION_TEXT,
    LLM_BENCHMARKS_TEXT,
    TITLE,
    LINKS,
)
from src.display.css_html_js import (
    custom_css,
    CSS_EXTERNAL,
)
from src.display.utils import (
    AutoEvalColumn,
    fields,
)
from src.envs import (
    API,
    EVAL_DETAILED_RESULTS_PATH,
    EVAL_RESULTS_PATH,
    EVAL_DETAILED_RESULTS_REPO,
    REPO_ID,
    RESULTS_REPO,
    TOKEN,
)
from src.populate import get_leaderboard_df


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


### Space initialisation
try:
    print(EVAL_DETAILED_RESULTS_REPO)
    snapshot_download(
        repo_id=EVAL_DETAILED_RESULTS_REPO,
        local_dir=EVAL_DETAILED_RESULTS_PATH,
        repo_type="dataset",
        tqdm_class=None,
        etag_timeout=30,
        token=TOKEN,
    )
except Exception:
    restart_space()
try:
    print(EVAL_RESULTS_PATH)
    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()


LEADERBOARD_DF = get_leaderboard_df(RESULTS_REPO)


def GET_DEFAULT_TEXTBOX():
    return gr.Textbox("", placeholder="🔍 Search Models... [press enter]", label="Filter Models by Name")


def GET_DEFAULT_CHECKBOX():
    print("Choices:", [c.name for c in fields(AutoEvalColumn) if not c.hidden])
    return gr.CheckboxGroup(
        choices=[c.name for c in fields(AutoEvalColumn) if not c.hidden],
        label="Select Columns to Display",
        value=[c.name for c in fields(AutoEvalColumn) if c.displayed_by_default],
    )


def init_leaderboard(dataframes):
    subsets = list(dataframes.keys())

    with gr.Row():
        selected_subset = gr.Dropdown(choices=subsets, label="Select Dataset Subset", value=subsets[-1])
        research_textbox = GET_DEFAULT_TEXTBOX()
        selected_columns = GET_DEFAULT_CHECKBOX()

    data = dataframes[subsets[-1]]

    with gr.Row():
        datatype = [c.type for c in fields(AutoEvalColumn)]
        df = gr.Dataframe(data, datatype=datatype, type="pandas")

    def refresh(subset):
        global LEADERBOARD_DF
        LEADERBOARD_DF = get_leaderboard_df(RESULTS_REPO)
        default_columns = [c.name for c in fields(AutoEvalColumn) if c.displayed_by_default]

        return update_data(subset, None, default_columns), GET_DEFAULT_TEXTBOX(), GET_DEFAULT_CHECKBOX()

    def update_data(subset, search_term, selected_columns):
        print("Subset:", subset)
        print("Search Term:", search_term)
        print("Selected Columns:", selected_columns)
        filtered_data = dataframes[subset]
        if search_term:
            filtered_data = filtered_data[dataframes[subset]["Model Name"].str.contains(search_term, case=False)]
        filtered_data.sort_values(by="Total", ascending=False, inplace=True)
        selected_columns = [c.name for c in fields(AutoEvalColumn) if c.name in selected_columns]
        selected_data = filtered_data[selected_columns]
        return gr.DataFrame(
            selected_data,
            type="pandas",
            datatype=[c.type for c in fields(AutoEvalColumn) if c.name in selected_columns],
        )

    with gr.Row():
        refresh_button = gr.Button("Refresh")
        refresh_button.click(
            refresh,
            inputs=[
                selected_subset,
            ],
            outputs=[df, research_textbox, selected_columns],
            concurrency_limit=20,
        )

    selected_subset.change(update_data, inputs=[selected_subset, research_textbox, selected_columns], outputs=df)
    research_textbox.submit(update_data, inputs=[selected_subset, research_textbox, selected_columns], outputs=df)
    selected_columns.change(update_data, inputs=[selected_subset, research_textbox, selected_columns], outputs=df)

def init_detailed_results():
    with gr.Row():
        gr.HTML("""\
<iframe
  src="https://huggingface.co/datasets/lmms-lab/LiveBenchDetailedResults/embed/viewer/"
  frameborder="0"
  width="100%"
  height="800px"
></iframe>
""")


HEAD = "".join(
    [
        f'<link rel="stylesheet" href="{css}">' for css in CSS_EXTERNAL
    ]
)

print("HEAD:", HEAD)

demo = gr.Blocks(css=custom_css, head = HEAD)
with demo:
    gr.HTML(TITLE)
    gr.HTML(LINKS)
    gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")

    with gr.Tabs(elem_classes="tab-buttons") as tabs:
        with gr.TabItem("🏅 LiveBench Results", elem_id="llm-benchmark-tab-table", id=0):
            init_leaderboard(LEADERBOARD_DF)

        with gr.TabItem("📝 Detailed Results", elem_id="llm-benchmark-tab-table", id=2):
            init_detailed_results()

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

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