KcBERT

When this dataset fits

KcBERT is a Korean Pretraining Corpus from Hugging Face. It is closer to pretraining corpora than to a single SFT stage. License label CC BY-SA 4.0 still needs review. Size text is “7.90 GiB”, copied from the host.

Task / category

Text · Text · pretraining, spoken

Language

Korean

Format · Size

Text · 7.90 GiB (86,246,285 sentences)

License

CC BY-SA 4.0(Catalog hint — re-check original) · License document

Source & original link

Hugging Face · Original dataset link(Verified)

Verification status

Source: Verified · License: Needs review · Last checked: 2026-09-03

KcBERT: Korean Pretraining Corpus from Hugging Face. License CC BY-SA 4.0. Open the original link to download.

KcBERT (2020). Hugging Face. license label: CC BY-SA 4.0 catalog commercial mark: allowed Original: https://huggingface.co/datasets/Bingsu/KcBERT_Pre-Training_Corpus. GearDel catalog: https://geardel.com/en/datasets/kcbert-pretrain-corpus. Last checked: 2026-09-03. Author names are not stored in the GearDel catalog and are not invented here.

Differences below use catalog fields only: license, commercial mark, provider, size text, and year.

• Korean Pretraining CorpusKorean Pretraining Corpus · Hugging FaceViewLicense labels differ (CC BY-SA 4.0 vs MIT). Size text differs (7.90 GiB vs Medium). Those strings are copied from the hosts. Catalog years differ (2020 vs 2025).
• Korean Pretraining CorpusKorean Pretraining Corpus · Hugging FaceViewCommercial marks differ (Commercial use allowed (catalog) vs Commercial use may be conditional). Size text differs (7.90 GiB vs 764MB). Those strings are copied from the hosts. Catalog years differ (2020 vs 2024).
• Korean Dialogue DatasetKorean Dialogue Dataset · Hugging FaceViewLicense labels differ (CC BY-SA 4.0 vs Apache 2.0). Size text differs (7.90 GiB vs Large). Those strings are copied from the hosts. Catalog years differ (2020 vs 2023).
• KoGPT2Korean Pretraining Corpus · Hugging FaceViewLicense labels differ (CC BY-SA 4.0 vs License unknown). Size text differs (7.90 GiB vs 3.2GB). Those strings are copied from the hosts. Catalog years differ (2020 vs 2021).