Explore high-quality datasets for your AI and machine learning projects.
The CoNaLa dataset is a benchmark for code generation tasks, containing code‑natural language pairs. The data were crawled from Stack Overflow, automatically filtered and manually annotated, comprising 2,379 training samples and 500 test samples. Additionally, an automatically mined set with nearly 600,000 samples is provided. The dataset is used to evaluate code generation, with English language and Python code. It includes two versions: a manually annotated version and an automatically mined version, each with different fields and splits.
