LCC Java is a Java code dataset released by Microsoft in 2023, derived from The Large Code Corpus (LCC) project, designed to provide high-quality code corpora for tasks such as code understanding, code generation, and code completion.

The dataset contains 120,000 Java code samples sourced from real Java source code in open-source projects, and is split into training, validation, and test sets. Each sample primarily consists of context code and the corresponding target code snippet, stored in JSON format, suitable for supervised learning and sequence-to-sequence (Seq2Seq) training of code-related models.

Dataset Composition

The dataset comprises the following three parts:

• Training set (train): 100,000 samples, approximately 1.61 GB.
• Validation set (validation): 10,000 samples, approximately 131 MB.
• Test set (test): 10,000 samples, approximately 129 MB.

Each sample primarily contains the following fields:

• context: Context code, providing the model with background information about the code.
• gt: Target code snippet, i.e., the code the model needs to predict or generate.