README.md

metadata

language:
  - zh
  - en
  - ja

Dataset for pre-training the LLM Quirel.

The code for processing data will be open-sourced later.

[cn_baike_md_structured]

baidu baike data in Chinese, with Markdown structure

Collected from https://huggingface.co/datasets/lars1234/baidu-baike-dataset

Cleaning procedure:

• Organize context in Markdown.
• Delete the paragraph with many network links.
• Filter out texts that are too short.

[cn_wiki]

wikipedia data in Chinese

Collected from https://huggingface.co/datasets/Blaze7451/Wiki-zh-20250601

Cleaning procedure:

• Remove lang template in the format of "-{zh: , zh-cn: ,}-".
• Remove sections of reference links/documents.

[en_wiki_md_structured]

wikipedia data in English, with Markdown struture

Collected from https://huggingface.co/datasets/Blaze7451/enwiki_structured_content

Cleaning procedure:

• Clean empty MD sections.
• Filter out texts that are too short.

[ja_wiki]

wikipedia data in Japanese

Collected from https://huggingface.co/datasets/Blaze7451/Wiki-ja-20250601

Cleaning procedure:

• Remove sections of reference links/documents.