Mufi-Lang Epoch1 Dataset

A large-scale dataset of valid Mufi programs for neural network training and compiler testing.

Dataset Information

• Target Size: 100,000 programs
• Generated: 2026-02-08 19:50:53
• Format: .mufi source files
• Validation: All programs are syntactically and semantically valid

Structure

.
├── programs/          # All generated .mufi files
│   ├── prog_0.mufi
│   ├── prog_1.mufi
│   └── ...
├── metadata.json      # Dataset statistics and metadata
├── README.md          # This file
└── .gitignore        # Git ignore rules

Usage

Loading Programs

from pathlib import Path

dataset_dir = Path(".")
programs = list((dataset_dir / "programs").glob("*.mufi"))

for prog_file in programs:
    with open(prog_file) as f:
        code = f.read()
    # Process code...

Running Programs

# Run a single program
mufiz -r programs/prog_0.mufi

# Run all programs
for f in programs/*.mufi; do
    mufiz -r "$f"
done

Statistics

See metadata.json for detailed statistics including:

• Total program count
• Generation time and rate
• Feature distribution
• Complexity metrics

Archiving

To create compressed archives:

# Create tar.gz archive
tar -czf mufiz-dataset.tar.gz programs/

# Create zip archive
zip -r mufiz-dataset.zip programs/

Or use the provided management tool:

python3 manage_dataset.py archive --format all

Quality Assurance

All programs in this dataset:

• ✅ Parse successfully (no syntax errors)
• ✅ Type-check correctly (no type errors)
• ✅ Execute without runtime errors
• ✅ Terminate properly (no infinite loops)
• ✅ Follow the official Mufi PEG grammar

License

This dataset is part of the MufiZ project.

Citation

If you use this dataset in your research, please cite:

@dataset{mufiz_epoch1_dataset,
  title={Mufi-Lang Epoch 1 Dataset},
  author={Mustafif Khan},
  year={2026},
  url={https://github.com/Mufi-Lang/Epoch1}
}