sst2_ar / README.md
akhooli's picture
Update README.md
bb60720 verified
metadata
license: mit
dataset_info:
  features:
    - name: text
      dtype: string
    - name: label
      dtype: int64
    - name: label_text
      dtype: string
  splits:
    - name: train
      num_bytes: 1760713
      num_examples: 9596
  download_size: 881087
  dataset_size: 1760713
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

This dataset is machine translated to Arabic from an English copy (https://huggingface.co/datasets/SetFit/sst2). The original SST2 dataset (https://huggingface.co/datasets/stanfordnlp/sst2) has more rows. A few duplicate rows were removed. The dataset is used to train a few shot inference using the SetFit library. More samples are used in Arabic due to low quality. The base model for training is akhooli/sbert_ar_nli_500k_norm which is normalized, so the text and queries are also normalized in the script.