README.md

metadata

dataset_info:
  features:
    - name: label
      dtype: float64
    - name: idx
      dtype: int64
    - name: sentence1
      dtype: string
    - name: sentence2
      dtype: string
    - name: source
      dtype: string
  splits:
    - name: train
      num_bytes: 3122340
      num_examples: 14597
    - name: validation
      num_bytes: 670209
      num_examples: 3489
  download_size: 1879461
  dataset_size: 3792549
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: validation
        path: data/validation-*
task_categories:
  - sentence-similarity
language:
  - he

Description

Machine-translated Hebrew version sts-b dataset, with additional records: augmented - not-matched records that were generated by weak generative model, and "chatgpt" - paraphrases that were generated by chatgpt accordingly

Sample

[{'label': 4.666999816894531, 'idx': 13, 'sentence1': 'אדם מקפל פיסת נייר.', 'sentence2': 'מישהו מקפל פיסת נייר.', 'source': 'machine-translated'},
{'label': 0.0, 'idx': 13, 'sentence1': 'אדם מקפל פיסת נייר.', 'sentence2': 'כתב מייל.', 'source': 'augmented'},
{'label': 4.0, 'idx': 13, 'sentence1': 'אדם מקפל פיסת נייר.', 'sentence2': 'אדם מכפיל ניידת נייר.', 'source': 'chatgpt'}]