metadata
dataset_info:
features:
- name: sentence
dtype: string
- name: target
dtype:
class_label:
names:
'0': eight
'1': five
'2': four
'3': nine
'4': 'no'
'5': one
'6': seven
'7': six
'8': ten
'9': three
'10': two
'11': zero
splits:
- name: train
num_bytes: 659600
num_examples: 8320
- name: validation
num_bytes: 81438
num_examples: 1029
- name: test
num_bytes: 86986
num_examples: 1095
download_size: 438887
dataset_size: 828024
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: validation
path: data/validation-*
- split: test
path: data/test-*
source_datasets:
- INK-USC/numer_sense
numer_sense
NumerSense masked number words (0-10), with deterministic 80/10/10 splits.
Original data: INK-USC/numer_sense. Repackaged as parquet for tasksource by scripts/upload_repackaged.py.