yourbench_space/leaderboard_space/app.py

import gradio as gr

from utils import run_pipeline, update_examples
from env import TASK

with gr.Blocks(
    title="YourBench Leaderboard",
    theme=gr.themes.Soft(),
    css="button { margin: 0 10px; padding: 5px 15px; }",
) as demo:
    # DISPLAY TABLE AND ANALYSIS
    title = gr.Markdown(f"YourBench auto-Leaderboard for {TASK}")
    leaderboard = gr.DataFrame(label="Results", interactive=False)
    samples_ix = gr.Number(
            label="Example Index", 
            value=0,
            step=1,
            info="Navigate through different examples"
        )
    with gr.Tab("Hardest samples"):
        hard_samples = gr.HTML()
    with gr.Tab("Easiest samples"):
        easy_samples = gr.HTML()
    with gr.Tab("All samples"):
        all_samples = gr.HTML()

    samples_ix.change(update_examples, samples_ix, [easy_samples, hard_samples, all_samples])

    demo.load(run_pipeline, [samples_ix], [leaderboard, easy_samples, hard_samples, all_samples])

demo.launch()