Introducing CodeAlchemy, a synthetic data pipeline that has already produced nearly 1 trillion tokens of open-source code