---
dataset_info:
  features:
  - name: text
    dtype: string
  - name: timestamp
    dtype: string
  - name: url
    dtype: string
  - name: source
    dtype: string
  splits:
  - name: train
    num_bytes: 839280469
    num_examples: 153461
  download_size: 321193165
  dataset_size: 839280469
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
---

- The data is sourced from the Odia subset of [CulturaX](https://huggingface.co/datasets/uonlp/CulturaX) dataset, which itself is sourced from mC4 and four OSCAR corpora.
- It has 49M tokens and approximately 2.3M sentences.
- It is safe to use and doesn't have any toxic component.