CodeXGLUE is a benchmark dataset designed to advance research in program understanding and generation. It includes 14 datasets, 10 diverse programming language tasks, and a platform for model evaluation and comparison. The dataset features three baseline systems: BERT-style, GPT-style, and Encoder-Decoder models. CodeXGLUE supports tasks such as code clone detection, defect detection, cloze test, code completion, code translation, code search, code repair, text-to-code generation, code summarization, and documentation translation. The datasets are chosen or created based on the criterion that they support the development and evaluation of data-driven machine learning methods. CodeXGLUE includes both previously proposed datasets and newly introduced ones. The dataset provides three baseline models to help perform the tasks, including CodeBERT for code understanding, CodeGPT for completion and generation, and an Encoder-Decoder framework for sequence-to-sequence generation. The benchmark aims to accelerate research in programming language tasks by providing a comprehensive set of tasks and datasets for model evaluation and comparison.CodeXGLUE is a benchmark dataset designed to advance research in program understanding and generation. It includes 14 datasets, 10 diverse programming language tasks, and a platform for model evaluation and comparison. The dataset features three baseline systems: BERT-style, GPT-style, and Encoder-Decoder models. CodeXGLUE supports tasks such as code clone detection, defect detection, cloze test, code completion, code translation, code search, code repair, text-to-code generation, code summarization, and documentation translation. The datasets are chosen or created based on the criterion that they support the development and evaluation of data-driven machine learning methods. CodeXGLUE includes both previously proposed datasets and newly introduced ones. The dataset provides three baseline models to help perform the tasks, including CodeBERT for code understanding, CodeGPT for completion and generation, and an Encoder-Decoder framework for sequence-to-sequence generation. The benchmark aims to accelerate research in programming language tasks by providing a comprehensive set of tasks and datasets for model evaluation and comparison.