wikitext_mlm_nsp / README.md
droo303's picture
Update README.md
7dc8ea5 verified
metadata
dataset_info:
  features:
    - name: input_ids
      sequence: int32
    - name: attention_mask
      sequence: int8
    - name: token_type_ids
      sequence: int8
    - name: labels
      sequence: int64
    - name: next_sentence_labels
      dtype: int64
  splits:
    - name: train
      num_bytes: 12337588320
      num_examples: 1715460
    - name: validation
      num_bytes: 26423408
      num_examples: 3674
  download_size: 1076906240
  dataset_size: 12364011728
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: validation
        path: data/validation-*

Preprocessed dataset intended for pretraining of base models. Dataset was inspired by BERT pretraining - it includes NSP and MLM tasks and labels.