Mini-VNCC

Description

The mini-vncc is a 777,777 unique web documents represents an intensively filtered collection of Vietnamese web content, meticulously extracted from ~6TB of Vietnamese text in all CommonCrawl archive from 2013 to 2023. It is specifically tailored for pretraining models on Markdown structured web content in Vietnamese.

Access

To access the dataset, use:

from datasets import load_dataset

dataset = load_dataset("nampdn-ai/mini-vncc")

Dataset Structure

Data Fields

• url: Source web page URL.
• text: Extracted Vietnamese HTML text converted into Markdown.
• length: Number of characters in the document.

Data Usage

Ideal for pretraining/finetune small Viet language models on Vietnamese Markdown structured web content.

Note: Data should be cleaned at the document level prior to usage to ensure quality.

Acknowledgements

Special thanks to the CommonCrawl project and Symato contributors for providing the base data.

Licensing

This dataset is available under CommonCrawl's licensing terms. For more details, visit CommonCrawl's license.