Skip to content

Dataset · m-a-p

800-billion-token Chinese pre-training corpus

m-a-p/MAP-CC

Open Chinese pre-training dataset of 800 billion tokens for the NLP community.

What's inside

This is an open Chinese pre-training corpus on a scale of 800 billion tokens. According to the card, the data passed compliance checks. The modalities are image and text, and the metadata size is from 1 to 10 billion records.

What you can do with it

  • Pre-training a Chinese language model
  • Continuing pre-training of a multimodal model
  • Evaluating the quality of Chinese corpora
Size
1B–10B rows
Downloads this month
9.7K
Likes
83

How to load it

from datasets import load_dataset

ds = load_dataset("m-a-p/MAP-CC")