thedruid831/JavaScript-Code-Large

JavaScript-Code-Large

JavaScript-Code-Large is a large-scale corpus of JavaScript source code comprising around 5 million JavaScript files. The dataset is designed to support research in large language model (LLM) pretraining, code intelligence, software engineering automation, and program analysis for the JavaScript ecosystem.

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

JavaScript-Code-Large addresses the need for a dedicated JavaScript-only dataset at substantial scale, enabling focused research across frontend, backend, and full-stack JavaScript environments. .

1. Dataset Composition

Programming Language: JavaScript

File Count: 5M+ JavaScript files

File Format: .jsonl

Content Types

The dataset includes a wide variety of JavaScript constructs and paradigms, such as:

• —Functions (declarations, expressions, arrow functions)

• —Classes and prototypes

• —Modules (CommonJS and ES Modules)

• —Asynchronous patterns (async/await, Promises, callbacks)

• —Event-driven code

• —Closures and higher-order functions

• —Functional programming constructs

• —DOM manipulation code

• —Node.js backend logic

• —Frontend framework components

• —JSDoc comments

• —Error handling patterns

• —Modern ES6+ features

2. Intended Research Applications

2.1 Pretraining

• —Training JavaScript code foundation models from scratch

• —Continued pretraining of existing LLMs

• —JavaScript-specialized language modeling

• —Tokenizer training for JS ecosystems

2.2 Fine-Tuning and Adaptation

• —Code completion systems

• —Intelligent IDE assistants

• —Automated refactoring tools

• —Conversational programming agents

• —JavaScript-specific copilots

2.3 Code Intelligence Tasks

• —Code summarization

• —Code-to-text generation

• —Documentation generation

• —Bug detection

• —Vulnerability detection

• —Clone detection

• —Code similarity modeling

• —Minified-to-readable code transformation

• —Static and structural analysis

2.4 Software Engineering Research

• —Empirical studies of JavaScript coding patterns

• —Analysis of async and event-driven architectures

• —Framework usage studies

• —Dependency modeling

• —AST-based experiments

• —Cross-version JavaScript evolution analysis

3. Relationship to [Java-Code-Large](https://huggingface.co/datasets/ajibawa-2023/Java-Code-Large)

JavaScript-Code-Large complements Java-Code-Large, enabling comparative research between:

• —Statically typed vs dynamically typed languages

• —Class-based vs prototype-based paradigms

• —Backend vs frontend dominant ecosystems

• —JVM vs Node.js environments

Together, these datasets support cross-language transfer learning and controlled specialization studies.

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