metadata

dataset_info:
  features:
    - name: text
      dtype: string
    - name: gpt_token_count
      dtype: int64
    - name: llama_token_count
      dtype: int64
  splits:
    - name: train
      num_bytes: 193124707426
      num_examples: 51700035
    - name: valid
      num_bytes: 193389444
      num_examples: 51752
  download_size: 113927709248
  dataset_size: 193318096870
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: valid
        path: data/valid-*

Pretraining dataset

This dataset is composed of 1. dolma 22B, 2. fineweb-edu 21B, and 3. starcoder 2B with common programming language (C, CPP, JAVA, PYTHON, JSON)

Token count

split	GPT2 token count	Llama3 token count
train	45.25B	41.80B
valid	45.04M	41.64M