---
license: apache-2.0
---

6,614 Chinese books cleaned from [this source](https://huggingface.co/datasets/SJTUCL/books_clean.tar.gz) by removing table-of-contents and other metadata (to the best of my attempt), dropping documents with larges portions of English content, and converting traditional Chinese to simplified Chinese.

**Warning**: while the remaining 6,614 books are all primarily in Chinese, most of them contain some English text. So be very cautious when drawing any conclusion from models trained on this data if you are conducting cross-lingual transfer or other related research.