marcelo1234/Java-Code-Large

Java-Code-Large

Java-Code-Large is a large-scale corpus of publicly available Java source code comprising more than 15 million java codes. The dataset is designed to support research in large language model (LLM) pretraining, code intelligence, software engineering automation, and program analysis.

By providing a high-volume, language-specific corpus, Java-Code-Large enables systematic experimentation in Java-focused model training, domain adaptation, and downstream code understanding tasks.

1. Introduction

Large-scale code corpora have become fundamental resources for training and evaluating machine learning models for code-related tasks. While multilingual code datasets exist, there is increasing interest in language-specialized corpora to:

• —Improve domain-specific performance

• —Reduce cross-language noise

• —Enable controlled experimental settings

• —Support Java-specific tooling and research

Java-Code-Large addresses this need by providing a dedicated Java-only dataset at substantial scale.

2. Dataset Composition

Programming Language: Java

File Count: 15M+ Java files

File Format: .jsonl

Content Types:

• —Classes

• —Interfaces

• —Enums

• —Methods

• —Annotations

• —JavaDoc comments

• —Exception handling structures

• —Generics and concurrency constructs

The dataset consists of source code extracted from publicly accessible open-source repositories.

3. Intended Research Applications

3.1 Pretraining

• —Training code foundation models from scratch

• —Continued pretraining of existing LLMs

• —Java-specialized language modeling

3.2 Fine-Tuning and Adaptation

• —Code completion systems

• —Automated refactoring tools

• —IDE copilots

• —Java-specific conversational assistants

3.3 Code Intelligence Tasks

• —Code summarization

• —Code-to-text generation

• —Bug detection

• —Vulnerability detection

• —Clone detection

• —Code similarity modeling

• —Static and structural analysis

3.4 Software Engineering Research

• —Empirical studies of Java programming patterns

• —Tokenization and AST modeling experiments

Thanks to open source community for all the guidance & support!!