JetCopper-10B / README.md
sudy-super's picture
Update README.md
2fee5b2 verified
metadata
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
dataset_info:
  features:
    - name: text
      dtype: string
  splits:
    - name: train
      num_bytes: 51355991596
      num_examples: 15930958
  download_size: 29126915011
  dataset_size: 51355991596
language:
  - ja
  - en
  - code
size_categories:
  - 10M<n<100M
license: apache-2.0

JetCopper-10B

Description

JetCopper-10B was created by extracting a portion of the data after cleaning, filtering, and deduplicating the following datasets.

This dataset was used to pre-train Contrail-200m-64k when we participated in LOCAL AI HACKATHON #000.

The number of tokens (Using tokenizer of calm2-chat)

Language The number of tokens
Japanese 4.7b
English 5b
Code 0.9b

NOTE

This dataset has not passed sentence end boundary determination or Perplexity Filtering, so there is room for improvement in quality.