This dataset is a ~30GB sample from the larger eliceai/korean-translated-fineweb-edu-dedup dataset (600GB total). It contains Korean-translated educational content from the FineWeb dataset.

Dataset Summary

Source: Sampled from eliceai/korean-translated-fineweb-edu-dedup
Size: ~30GB
Language: Korean
Content: Educational text content
Format: Parquet files

Dataset Structure

The dataset contains a single train split with the following features:

text: Korean text content (string)

Usage

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("eliceai/korean-fineweb-edu-demo", split="train")

# Access examples
for example in dataset:
    print(example["text"])

Sampling Method

This dataset was created using streaming mode to sample approximately 30GB from the source dataset without downloading the entire 600GB dataset. The sampling preserves the original order and structure of the source data.

Source Dataset

Original Dataset: eliceai/korean-translated-fineweb-edu-dedup
Original Size: ~600GB
Sampling Ratio: ~5% of original dataset

License

This dataset follows the same license as the source dataset (Apache 2.0).

Citation

If you use this dataset, please cite the original source:

@dataset{korean_fineweb_edu_demo,
  title={Korean FineWeb Edu Demo},
  author={Elice AI},
  year={2024},
  url={https://huggingface.co/datasets/eliceai/korean-fineweb-edu-demo}
}

Downloads last month: 3