license: mit
modalities:
  - Text
formats:
  - parquet
size: 10M - 100M
libraries:
  - Datasets
  - Dask
  - Croissant
  - Polars
π GitHub Code 2025: The Clean Code Manifesto
A meticulously curated dataset of 1.5M+ repositories representing both quality and innovation in 2025's code ecosystem
π The Philosophy
Quality Over Quantity, Purpose Over Volume
In an era of data abundance, we present a dataset built on radical curation. Every file, every repository, every byte has been carefully selected to represent the signal in the noise of open-source development.
π― What This Dataset Is
π Dual-Perspective Design
| Subset | ποΈ Above 2 Stars | π± Below 2 Stars (2025) | 
|---|---|---|
| Scope | 1M top repositories | 1M random 2025 repos | 
| Purpose | Proven quality & patterns | Emerging trends & innovation | 
| Value | What works | What's next | 
π§Ή The Clean Code Promise
# What you WON'T find here:
π« Binary files          # No images, executables, models
π« Build artifacts       # No node_modules, __pycache__
π« Configuration noise   # No .git, IDE files, lock files
π« License duplication   # No repetitive legal text
π« Minified code         # No compressed/obfuscated content
π« Empty files           # No whitespace-only content
π Dataset Structure
github-code-2025/
βββ π above-2-stars/
β   βββ train_000.parquet
β   βββ train_001.parquet
β   βββ ...
βββ π± below-2-star/
    βββ train_000.parquet
    βββ train_001.parquet
    βββ ...
π Schema
{
    "repo_id": "owner/repo_name",    # π Repository identifier
    "file_path": "src/main.py",      # ποΈ Relative file path
    "content": "def clean_code():",   # π Actual source code
    "size": 1024                     # π File size in bytes
}
π οΈ How to Use
π₯ Quick Start
from datasets import load_dataset
# Load the quality benchmark
quality_ds = load_dataset("nick007x/github-code-2025", "above-2-stars")
# Load emerging trends
emerging_ds = load_dataset("nick007x/github-code-2025", "below-2-star")
# Mix for balanced training
balanced_ds = interleave_datasets([quality_ds, emerging_ds])
π― Ideal Use Cases
- π§ AI Training: Clean, diverse code for language models
- π Code Analysis: Compare popular vs emerging patterns
- π Trend Research: 2025 development practices
- π Education: High-quality examples for learning
- π οΈ Tool Development: Benchmarking code quality tools
ποΈ Creation Methodology
π¨ Selection Strategy
| Phase | Action | Purpose | 
|---|---|---|
| 1 | π― Dual population sampling | Balance quality & innovation | 
| 2 | π§Ή Multi-layer filtering | Remove noise & binaries | 
| 3 | π Size normalization | Focus on meaningful content | 
| 4 | π Content validation | Ensure text quality | 
| 5 | π·οΈ Metadata preservation | Maintain context | 
π« What We Filtered Out
File Types Removed:
- 50+ binary extensions (images, models, executables)
- 30+ build/system directories
- 15+ configuration file types
- All files outside 1KB-5MB range
Quality Checks:
- β UTF-8 text validation
- β Non-empty content check
- β Binary detection
- β Repository structure preservation
πͺ Why This Dataset Matters
π« The Quality Revolution
We reject the "more data is better" dogma. Instead, we offer:
- π― Intentional Curation: Every file serves a purpose
- βοΈ Balanced Perspective: Popular + Emerging = Complete picture
- π§Ή Unprecedented Cleanliness: The cleanest code dataset available
- π Temporal Intelligence: 2025-focused for relevance
π€ Contributing & Feedback
This dataset is a living project. We welcome:
- π Bug reports and issues
- π‘ Feature requests for future versions
- π Validation of data quality
- π― Suggestions for improvement
π License
This dataset is provided under the MIT License - see the LICENSE file for details.
Important: Repository contents maintain their original licenses. Please respect individual project licenses when using this data.
π Acknowledgments
Built with gratitude for the entire open-source community. Every file in this dataset represents hours of dedication from developers worldwide.
β If this dataset helps your research or project, please consider starring the repository!
"In the pursuit of AI that understands code, we must first understand what code is worth learning."
