app.py

import os
import gradio as gr
from datasets import load_dataset

auth_token = os.environ.get("auth_token")
visit_bench_all = load_dataset("mlfoundations/VisIT-Bench", use_auth_token=auth_token)
print('visit_bench_all')
print(visit_bench_all)
print('dataset keys:')
print(visit_bench_all.keys())
dataset_keys = list(visit_bench_all.keys())
assert len(dataset_keys) == 1
dataset_key = dataset_keys[0]
visit_bench = visit_bench_all[dataset_key]
print('first item:')
print(visit_bench[0])

df = visit_bench.to_pandas()
print(f"Got {len(df)} items in dataframe")
df = df.sample(frac=1)

LINES_NUMBER = 20

def display_df():
    df_images = df.head(LINES_NUMBER)
    return df_images

def display_next(dataframe, end):
    start = int(end or len(dataframe))
    end = int(start) + int(LINES_NUMBER)
    global df
    if end >= len(df) - 1:
        start = 0
        end = LINES_NUMBER
        df = df.sample(frac=1)
        print(f"Shuffle")
    df_images = df.iloc[start:end]
    assert len(df_images) == LINES_NUMBER
    return df_images, end

initial_dataframe = display_df()

# Gradio Blocks
with gr.Blocks() as demo:
    gr.Markdown("<h1><center>VisIT-Bench Dataset Viewer</center></h1>")

    with gr.Row():
        num_end = gr.Number(visible=False)
        b1 = gr.Button("Get Initial dataframe")
        b2 = gr.Button("Next Rows")

    with gr.Row():
        out_dataframe = gr.Dataframe(initial_dataframe, wrap=True, max_rows=LINES_NUMBER, overflow_row_behaviour="paginate",
                                     interactive=False)

    b1.click(fn=display_df, outputs=out_dataframe, api_name="initial_dataframe")
    b2.click(fn=display_next, inputs=[out_dataframe, num_end], outputs=[out_dataframe, num_end],
             api_name="next_rows")

demo.launch(debug=True, show_error=True)