104,885 professionally curated code samples from The Stack dataset
This is a TRY VERSION of our enterprise 1.4TB dataset
A professionally curated and balanced subset of The Stack v2 dataset, meticulously processed and cleaned for machine learning applications. Perfect for code completion, language detection, and AI model training.
Core: machine-learning code-generation programming artificial-intelligence bigcode training-data curated commercial-license
Languages: python javascript cpp ruby swift shell yaml php markdown
Features: enterprise high-quality processed the-stack syntax-validation dataset ml-ready
🎯 Enterprise Dataset Available: This is a sample of our full 1.4TB enterprise dataset with 10M+ samples. Contact us for enterprise licensing.
from datasets import load_dataset
# Load the complete dataset from HuggingFace
dataset = load_dataset("vinsblack/The_Stack_Processed-v2")
print(f"Total samples: {len(dataset['train'])}") # 104,885
# Filter by language (perfectly balanced)
python_samples = dataset['train'].filter(lambda x: x['language'] == 'Python')
print(f"Python samples: {len(python_samples)}") # ~10,001
# Access quality scores (91.3% high quality)
high_quality = dataset['train'].filter(lambda x: x['quality_score'] > 0.9)
print(f"High quality samples: {len(high_quality)}")| Metric | Value | Details |
|---|---|---|
| Total Samples | 104,885 | Perfectly balanced across languages |
| File Size | 923.7 MB | Optimized Parquet format |
| Languages | 8 major | ~10,000 samples each |
| Quality Score | 91.3% | Syntax validated & curated |
| Format | Parquet/Arrow | ML-ready, fast loading |
| Source | The Stack v2 | BigCode official dataset |
Perfectly Balanced - Each language contains ~10,000 high-quality samples:
| Language | Files | Format | Quality Avg | Use Cases |
|---|---|---|---|---|
| Python | 10,001 | .py |
0.925 | AI/ML, automation, data science |
| Markdown | 10,003 | .md |
0.891 | Documentation, README files |
| Shell | 10,000 | .sh |
0.887 | DevOps, automation scripts |
| C/C++ | 10,000 | .h/.cpp |
0.934 | System programming, performance |
| Ruby | 10,000 | .rb |
0.912 | Web development, scripting |
| Swift | 10,000 | .swift |
0.928 | iOS/macOS development |
| YAML | 10,000 | .yml |
0.865 | Configuration, CI/CD |
| JavaScript | 9,999 | .js |
0.919 | Web development, Node.js |
| PHP | 9,995 | .php |
0.903 | Web backend, CMS |
Additional languages: JSON (242 files), HTML (220), XML (155), Java (106), C (101)
Our enterprise-grade curation pipeline ensures exceptional quality:
- ✅ 91.3% syntax validity across all languages
- ✅ 98.7% file accessibility and encoding
- ✅ AST parsing for Python, JavaScript, C++
- ✅ Compiler checks for compiled languages
- ✅ Security scanning - All files malware-free
- 🧹 Malware scanning - Security validated with Avira
- 🔄 Deduplication - Hash-based duplicate removal
- 📏 Size filtering - Removed empty/minimal files
- 🎯 Quality scoring - Multi-factor algorithm (0.0-1.0)
- 📊 Metadata enrichment - Repository info, stars, dates
- High (>0.9): 65,234 samples (62.2%)
- Medium (0.7-0.9): 32,157 samples (30.7%)
- Acceptable (0.5-0.7): 7,494 samples (7.1%)
- ⚡ 4.1x faster loading vs raw Stack
- 💾 50% memory reduction vs unprocessed
- 🚀 25% faster training time
- 📦 16,500x smaller than full Stack (4.3TB → 923MB)
- Fine-tune CodeT5, CodeBERT, StarCoder models
- Build IDE autocomplete systems
- Train domain-specific code assistants
- Create syntax suggestion engines
- Programming language classification (99.2% accuracy)
- Code quality assessment tools
- Syntax pattern recognition
- Code complexity analysis
- Academic ML research projects
- Educational AI/ML curricula
- Rapid prototyping with clean data
- Benchmark dataset for evaluations
- IDE plugins and extensions
- Code review automation systems
- Developer productivity tools
- Enterprise AI coding assistants
pip install datasets pandas numpy
python -c "from datasets import load_dataset; print('✅ Ready to go!')"git clone https://github.com/vinsblack/The_Stack_Processed-v2
cd The_Stack_Processed-v2
pip install -r requirements.txt
python examples/basic_usage.pypip install datasets>=2.0.0 pandas>=1.5.0 numpy>=1.21.0
# Optimized for production ML pipelinesThe_Stack_Processed-v2/
├── 📄 README.md # This documentation
├── ⚖️ LICENSE.md # Commercial license (€500-15K)
├── 📝 CHANGELOG.md # Version history & updates
├── 🔧 requirements.txt # Python dependencies
├── ⚙️ setup.py # Installation automation
├── 📊 data/
│ ├── train.parquet # Main dataset (923.7MB)
│ └── dataset_info.json # HuggingFace metadata
├── 💡 examples/
│ ├── basic_usage.py # Getting started guide
│ ├── quality_analysis.py # Advanced metrics
│ └── benchmark_tests.py # Performance validation
└── 🐛 ISSUE_TEMPLATE/
└── bug_report.md # Support template
- Local loading: 2-5 seconds (SSD)
- Memory usage: ~500MB fully loaded
- Streaming: Supports HuggingFace streaming
- Batch processing: Optimized for large-scale ML
- ✅ HuggingFace Datasets (native support)
- ✅ Pandas (direct DataFrame conversion)
- ✅ PyTorch (DataLoader ready)
- ✅ TensorFlow (tf.data compatible)
- ✅ Dask (distributed processing)
- Python: 3.8+ (tested on 3.8-3.11)
- Memory: 2GB RAM minimum, 4GB recommended
- Storage: 1GB free space
- OS: Windows, macOS, Linux (all tested)
Flexible pricing tiers for every use case:
- ✅ Research and educational use
- ✅ Publication rights with attribution
- ✅ Student project permissions
- ❌ No commercial deployment
- ✅ Commercial use (companies <€2M revenue)
- ✅ Model training and deployment
- ✅ Up to 10 developers
- ✅ 6-month update cycle
- ✅ Full commercial rights
- ✅ Unlimited team size
- ✅ Priority support (48h response)
- ✅ Monthly dataset updates
- ✅ Custom enterprise features
📧 Contact for licensing | 📄 Full terms
- Sample size: 104K samples ideal for small-medium models
- Enterprise version: 1.4TB with 10M+ samples available
- Language coverage: 8 major languages, expandable
- Domain focus: General-purpose programming (not domain-specific)
- Automated curation: May miss context-specific factors
- Bias inheritance: Inherits patterns from original Stack dataset
- Manual review: Recommended for critical applications
- Continuous improvement: Regular updates and refinements
- Fine-tuning: Excellent for model fine-tuning
- Evaluation: Perfect as high-quality evaluation set
- Production: Manual review recommended for production
- Research: Ideal for academic and research projects
python examples/basic_usage.py # Generate statistics
python examples/quality_analysis.py # Quality metrics
python examples/benchmark_tests.py # Performance tests| Dataset | Size | Quality | Speed | License | Cost |
|---|---|---|---|---|---|
| Stack Processed v2 | 923MB | 91.3% | Fast | Commercial | €500+ |
| The Stack (raw) | 4.3TB | ~60% | Slow | Open | Free |
| GitHub Code | 2TB+ | ~70% | Medium | Restricted | N/A |
| CodeSearchNet | 6GB | ~75% | Medium | Open | Free |
- 🤗 HuggingFace Dataset: vinsblack/The_Stack_Processed-v2
- 📊 Dataset Viewer: Browse samples online
- 📚 Documentation: Complete API reference
- 🛠️ Examples: Ready-to-run code samples
- 📈 Benchmarks: Performance comparisons
@dataset{stack_processed_v2_2025,
title={The Stack Processed v2: Enterprise-Grade Curated Code Dataset},
author={VinsBlack},
year={2025},
publisher={HuggingFace},
url={https://huggingface.co/datasets/vinsblack/The_Stack_Processed-v2},
note={Commercial license - Try version of 1.4TB enterprise dataset},
version={2.0.0}
}
- 📧 General Inquiries: vincenzo.gallo77@hotmail.com
- 💼 Commercial Licensing: vincenzo.gallo77@hotmail.com
- 🛠️ Technical Support: vincenzo.gallo77@hotmail.com
- 🐛 Bug Reports: GitHub Issues
- 🔗 Enterprise Dataset: Contact for 1.4TB full version
- Academic: 5 business days
- Startup: 48 hours
- Professional: 24 hours
- Enterprise: Same day
This dataset builds upon The Stack v2 by the BigCode Project. We thank the open-source community and Software Heritage for making this foundation possible.
Special thanks to the contributors who helped validate and improve this dataset.
- 🔍 Explore: Visit the HuggingFace dataset
- ⚖️ License: Review LICENSE.md for your use case
- 🤖 Build: Train your models with high-quality data
- 📈 Scale: Contact us for the enterprise 1.4TB version
Start building the next generation of AI coding assistants today! 💪
Last updated: January 2025 | Version 2.0.0 | Enterprise version available