---
license: cc-by-4.0
tags:
- code
- source-code
- software-engineering
- legacy-code
- code-generation
- code-completion
- code-understanding
- program-synthesis
- software-maintenance
- code-modernization
- automated-refactoring
- bug-detection
- technical-debt
- repository-mining
- data-structures
- algorithms
- competitive-programming
- computer-science
- llm
- reasoning
- problem-solving
- coding-interviews
- leetcode
- sft
- rlhf
- self-supervised-learning
- enterprise-software
- programming
task_categories:
- text-generation
pretty_name: "Legacy Codebase Dataset"
size_categories:
- 100M<n<1B
---

# Legacy Codebase Dataset

## Dataset Description

The **Legacy Codebase Dataset** is a large-scale collection of enterprise software repositories designed for training next-generation **Large Language Models (LLMs)**, AI coding assistants, software engineering copilots, automated refactoring systems, repository understanding models, and intelligent program analysis pipelines.

The complete collection contains **287 real-world legacy codebases** spanning **15 major industries**, comprising over **1.44 Billion Tokens Count**, **100.03 million lines of code**, **315,046 commits**, and **71,358 pull requests**.

Unlike synthetic coding datasets, this corpus represents mature production software developed over many years, capturing real engineering workflows, architectural evolution, collaborative development, technical debt, software maintenance, and repository history.

The dataset includes diverse programming paradigms, software architectures, dependency structures, coding conventions, version history, and enterprise engineering practices, making it ideal for training AI systems capable of understanding large software projects.

This dataset supports development pipelines for:

- Large Language Model (LLM) Pretraining
- Code Generation
- Code Completion
- Repository Understanding
- Software Modernization
- Automated Refactoring
- Bug Detection
- Technical Debt Analysis
- Program Repair
- Supervised Fine-Tuning (SFT)
- Self-Supervised Learning (SSL)
- Reinforcement Learning from Human Feedback (RLHF)

---

# Dataset Statistics

| Metric | Value |
|---------|------:|
| Industries | **15** |
| Legacy Codebases | **287** |
| Total Commits | **315,046** |
| Total Pull Requests | **71,358** |
| Total Tokens Count  | **1.44 Billion** |
| Total Lines of Code | **100.03 Million** |

---



# Dataset Specification

```
Industries:
15

Legacy Codebases:
287

Total Tokens Count 
1.44 Billion

Total Lines of Code:
100.03 Million

Total Commits:
315,046

Total Pull Requests:
71,358

Domain:
Enterprise Legacy Software

Data Type:
Source Code

Nature:
Real-world enterprise software repositories
```

---

# Dataset Features

- Enterprise-scale repositories
- Production-ready legacy software
- Multi-industry coverage
- Complete repository structures
- Rich commit history
- Pull request metadata
- Software evolution patterns
- Long-lived codebases
- Real engineering practices
- Large-scale source code corpus

---

# Key Use Cases

```
Large Language Model (LLM) Pretraining

Code Generation

Code Completion

Repository Understanding

Code Search

Program Synthesis

Software Modernization

Automated Refactoring

Bug Detection

Program Repair

Code Summarization

Documentation Generation

Technical Debt Analysis

Repository Mining

Software Engineering Research

Developer Copilots

AI Coding Assistants

Static Code Analysis

Long-context Code Reasoning

Benchmark Creation
```

---

# Value of This Dataset

The Legacy Codebase Dataset enables researchers and organizations to build production-ready AI systems capable of understanding complex enterprise software.

Applications include:

- AI Coding Copilots
- Enterprise Software Assistants
- Automated Code Migration
- Legacy Modernization
- Intelligent Refactoring Systems
- Repository Search Engines
- Bug Prediction Models
- Code Review Automation
- Software Maintenance Assistants
- Architecture Understanding
- Technical Debt Detection
- Large-scale Program Analysis

The combination of nearly **half a million source code files**, **100.03+ million lines of code**, extensive version history, and repositories from **15 industries** makes this dataset particularly valuable for training foundation models that require deep understanding of real-world software engineering.

---

# Basic JSON Schema

```json
{
  "repository_name": "string",
  "industry": "string",
  "file_name": "string",
  "file_path": "string",
  "language": "string",
  "commit_id": "string",
  "pull_request_id": "string",
  "code": "string"
}
```

---

# Data Creation

The dataset has been procured through formal agreements and generated in the ordinary course of business.

Quality processing includes:

- Repository validation
- File integrity verification
- Metadata extraction
- Repository organization
- Source code normalization
- Industry categorization
- Version history preservation

---

# Considerations

This repository contains **sample data** intended for research and educational purposes.

The complete enterprise collection consists of:

- **287 Legacy Codebases**
- **1.44 Billion Tokens Count**
- **100.03 Million Lines of Code**
- **315,046 Commits**
- **71,358 Pull Requests**
- **15 Industry Domains**

Enterprise licensing and access to the complete dataset are available through **InfoBay.AI**.

Website: https://infobay.ai/

Email: datareq@infobay.ai

Phone: (+91) 8303174762

---

# Citation

If you use this dataset in your research, publications, or commercial applications, please cite the corresponding dataset repository.

---

# License

**CC BY 4.0**