metadata

dataset_info:
  features:
    - name: text
      dtype: string
    - name: corpus
      dtype: string
    - name: original_id
      dtype: int64
  splits:
    - name: train
      num_bytes: 141807806497
      num_examples: 50336214
  download_size: 84893303434
  dataset_size: 141807806497
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
license: cc-by-nc-sa-4.0
language:
  - tr

Dataset Card for Dataset Name

vngrs-web-corpus is a mixed-dataset made of cleaned Turkish sections of OSCAR-2201 and mC4. This dataset is originally created for training VBART and later used for training TURNA. The cleaning procedures of this dataset are explained in Appendix A of the VBART Paper. It consists of 50.3M pages and 25.33B tokens when tokenized by VBART Tokenizer.

Dataset Details

Dataset Description

Curated by: VNGRS-AI
Language (NLP): Turkish
License: cc-by-nc-sa-4.0

Uses

vngrs-web-corpus is mainly intended to pretrain language models and word representations.

Dataset Structure

text[Str]: main text content of dataset
corpus[Str]: source corpus
original_id[Int]: original index of data at the source corpus

Bias, Risks, and Limitations

This dataset holds content crawled on the open web. It is cleaned based on a set of rules and heuristics without accounting for the semantics of the content. In cases where the content is irrelevant or inappropriate, it should be flagged and removed accordingly. The dataset is intended for research purposes only and should not be used for any other purposes without prior consent from the relevant authorities.

Citation

All attributions should be made to VBART paper.

@article{turker2024vbart,
  title={VBART: The Turkish LLM},
  author={Turker, Meliksah and Ari, Erdi and Han, Aydin},
  journal={arXiv preprint arXiv:2403.01308},
  year={2024}
}