This dataset is sampled from wikitext/wikitext-2-v1/train.
Codes to generate this dataset:
import datasets
dataset = datasets.load_dataset('wikitext', 'wikitext-2-v1')
selected = []
i = -1
while len(selected) < 24:
i += 1
text = dataset['train'][i]['text']
if 8 < len(text.split(' ')) <= 16 and '=' not in text:
selected.append(i)
tiny_dataset = dataset['train'].select(selected)