This dataset is 10% repo sampled dataset for selected languages. We applied a repo sample rate of 10%. e.g. if sample rate is 10% then we take 10% of all repos for a given language but include all files inside the repo.
This was generated using our codecomplete/training/completions/datagen
./launch.sh \
--dataset-name bigcode/starcoderdata \
--subset c,cpp,go,java,javascript,typescript,python,ruby,scala,sql \
--sample-rate 0.01 \
--hf-token <HF_TOKEN> \
--output-dir /home/${USER}/data \
--cache-dir /home/${USER}/hfcache \
--output-name c-cpp-go-java-javascript-typescript-python-ruby-scala-sql-0.01 \
--shuffle \
--build
Create the repository
# Install git lfs to suport large files
curl -s https://packagecloud.io/install/repositories/github/git-lfs/script.deb.sh | sudo bash
sudo apt-get install git-lfs
# create the dataset repo
huggingface-cli repo create <your_dataset_name> --type dataset --organization codecomplete
e.g.
huggingface-cli repo create base_dataset --type dataset --organization codecomplete
Clone the repository
git lfs install
git clone https://huggingface.co/datasets/<your_organization_name>/<your_dataset_name>
e.g.
git clone https://huggingface.co/datasets/codecomplete/base_dataset
Prepare your files
Create a descriptive README.md and check the dataset.json file
cp /somewhere/base_dataset/*.json .
git lfs track *.json
git add .gitattributes
git add *.json
git add --all
Upload your files
git status
git commit -m "First version of the your_dataset_name dataset."
git push
Verify dataset
from datasets import load_dataset
dataset = load_dataset("codecomplete/<your_dataset_name>")
print(dataset.num_rows)