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.