README.md

metadata

dataset_info:
  features:
    - name: title
      dtype: string
    - name: article
      dtype: string
    - name: label
      dtype: string
  splits:
    - name: train
      num_bytes: 4154802344
      num_examples: 1067358
  download_size: 2755606275
  dataset_size: 4154802344
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

This is a dataset crawled from the web from various sources based on traditional Chinese.

Data distribution

Token distribution

calculated using LlamaTokenizer
Total tokens: 2,184,375,300 tokens (2.1B)

news	wiki	tech	cosmopedia	human	science
1.2B	0.43B	0.19B	0.08B	0.2B	0.08B