import pandas as pd

df = pd.read_parquet('mscoco.parquet')
first_caps = df.drop_duplicates(subset='URL', keep='first')
first_caps.to_parquet('mscoco-1st-caption.parquet')