Data Description

With the rapid development of large language models, the demand for high-quality datasets in both the industry and academia is growing. These datasets not only need to contain a vast amount of information but also require rigorous screening and cleaning to ensure their accuracy and the safety of downstream models and applications. However, the currently popular public datasets in the industry have certain quality and security risks, especially in the Chinese domain where high-quality datasets are particularly lacking. Moreover, constructing a safe Chinese dataset also faces many challenges. Therefore, building a dataset that has undergone strict screening and standardized processing is particularly important for the innovation and development of LLMs.

Our CCI (Chinese Corpora Internet) dataset consists of high-quality, trustworthy sources from internet sites within mainland China. It has undergone rigorous data cleaning and deduplication, with targeted detection and filtering in aspects of content quality. The rules for data processing include:

Rule-based filtering: density-based extraction, keyword filtering, spam information filtering, conversion between simplified and traditional Chinese, etc.
Model-based filtering: filtering of low-quality content by training a classification model
Deduplication: within and between datasets dedup

Additionally, in response to the issue of pre-training data being large in scale and prone to causing leaks of evaluation data, we specifically conduct rigorous screening and filtering of several current mainstream Chinese evaluation datasets during the data processing phase.

The CCI corpus released (CCI v1.0.0) is 104GB in size. The overall timespan of the dataset ranges from January 2001 to November 2023.

Update

November 29, 2023, CCI v1.0.0 released!

Data Format

Field	Type	Meaning
id	String	Document ID, globally unique
title	String	Document title
content	String	Content of the document

Sample

{
    "id": "a262c26c915762ae107019f2797fda03",
    "title": "深圳人工智能企业闪耀东京展会",
    "content": "拳头产品叫好又叫座 深圳人工智能企业闪耀东京展会 近日在东京举行的日本人工智能展上,由深圳市贸促委组织的深圳人工智能企业展团集中亮相,引起热烈关注。该展会是日本规模最大的人工智能展会,云鲸智能、思谋科技、魔耳智能、格瑞普电池、云译科技等近20家深圳人工智能代表性企业的最新人工智能产品吸引了众多当地专业观众的目光,成为展会上的一抹亮色。企业现场“揽单”,参展成果丰硕深圳市大象机器人科技有限公司是一家由海外留学人才来深创建的专注于机器人研发生产的专精特新企业,本次在东京,该公司重点展示了myCobot协作机器人和仿真宠物猫metacat等公司拳头产品。“参展期间我们接待客户数达到500位以上,有意愿成为分销伙伴、集成商或终端客户的有效意向客户近70人,成效相当不错。……"
}

Download

The CCI dataset is simultaneously open-sourced on the BAAI DataHub and Huggingface.

BAAI DataHub

Users can click the link CCI Dataset to view the data files, and click to download.

Note that users need to register on BAAI DataHub to use the data, and filling out a survey questionnaire is required before their first download.

Huggingface

To use the data, you can load it using the following code:

from datasets import load_dataset

# If the dataset is gated/private, make sure you have run huggingface-cli login
dataset = load_dataset("BAAI/CCI-Data")

User Agreement

Users need to comply with the usage agreement of the CCI dataset. You can view the agreement by clicking on the following link: （View Usage Agreement）.

Notice

If you have any questions related to this dataset, please contact data@baai.ac.cn.