README.md

metadata

license: apache-2.0
size_categories:
  - 1M<n<10M
tags:
  - code
dataset_info:
  features:
    - name: max_stars_count
      dtype: int64
    - name: text
      dtype: string
    - name: token_count
      dtype: int64
  splits:
    - name: train
      num_bytes: 10787104987
      num_examples: 2130812
  download_size: 3723229232
  dataset_size: 10787104987
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

Common Starcoder dataset

This dataset is generated from bigcode/starcoderdata.

Total GPT2 Tokens: 4,649,163,171

Generation Process

1. We filtered the original dataset with common language: C, Cpp, Java, Python and JSON.
2. We removed some columns for mixing up with other dataset: "id", "max_stars_repo_path", "max_stars_repo_name"
3. After removing the irrelevant fields, we shuffle the dataset with random seed=42.
4. We filtered the data on "max_stars_count" > 300 and shuffle again.
5. We further reduced the dataset size by select(range(current_size, 2_500_000)), However there are only 2.13M samples left.
6. Add "n_tokens" by using GPT2Tokenizer to count the tokens in the "content" field.