CulturaX_0.5B / README.md
tpapa's picture
Upload dataset
d8d7fda verified
|
raw
history blame
739 Bytes
metadata
dataset_info:
  features:
    - name: text
      dtype: string
    - name: timestamp
      dtype: string
    - name: url
      dtype: string
    - name: source
      dtype: string
  splits:
    - name: train
      num_bytes: 3901684924
      num_examples: 741465
  download_size: 1933038896
  dataset_size: 3901684924
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

This dataset was created in the following way:

The CulturaX (https://huggingface.co/datasets/uonlp/CulturaX) dataset was loaded, specifically the Greek subset. Then a subset of the Greek subset was taken, specifically to contain about 0.5 B tokens. This is to be used as the general domain corpus for the Greek language for continual-pretraining.