Datasets for building models that write, debug, and reason about code. Covers competitive programming, software engineering, and code pretraining.