---
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](https://huggingface.co/datasets/INK-USC/numer_sense). Repackaged as parquet for [tasksource](https://github.com/sileod/tasksource) by [scripts/upload_repackaged.py](https://github.com/sileod/tasksource/blob/main/scripts/upload_repackaged.py).