Download
CodeSearchNet is a large-scale dataset and research benchmark designed to advance the development of systems that retrieve source code using natural language queries. The project was created through collaboration between GitHub and Microsoft Research and aims to support research on semantic code search and program understanding. The dataset contains millions of pairs of source code functions and corresponding documentation comments extracted from open-source repositories. These pairs allow machine learning models to learn relationships between natural language descriptions and programming code. The dataset currently covers several widely used programming languages, including Python, JavaScript, Ruby, Go, Java, and PHP. In addition to the dataset itself, the repository includes baseline models, evaluation tools, and instructions for building code retrieval systems that can map user queries to relevant code snippets.
Features
- Large dataset containing millions of code and documentation pairs
- Support for multiple programming languages including Python and Java
- Benchmark tasks for evaluating semantic code search algorithms
- Baseline machine learning models and pretrained weights
- Evaluation utilities and metrics for comparing retrieval systems
- Research platform for studying relationships between code and natural language
Project Samples
