c8data is an installable Python package for instruction-driven synthetic data
generation inspired by Meta's AutoData
/ Agentic Self-Instruct loop.
It creates candidate examples, verifies their quality, tests them against weak and strong solvers, judges the solver outputs with a rubric, and only accepts examples that create a meaningful weak-vs-strong performance gap.
- Setup guide — install, API keys, model configuration
- Usage guide — API reference, sources, output format, examples
uv add c8data
export C8DATA_API_KEY="your-api-key"from c8data import AutoDataGenerator
gen = AutoDataGenerator(output_dir="./data/out")
results = gen.run(source="paper.md")
print(results[0].accepted)
print(results[0].output_path)See the setup guide for full configuration options and the usage guide for custom instructions, batch runs, and output format.
This project is licensed under the MIT License.
If you use c8data, please cite the AutoData paper (arXiv:2606.25996):
@article{autodata2026,
title={Autodata: an agentic data scientist to create high quality data},
author={Kulikov, Ilia and Whitehouse, Chenxi and Wu, Tianhao and Nie, Yixin and Saha, Swarnadeep and Helenowski, Eryk and Yuan, Weizhe and Golovneva, Olga and Lanchantin, Jack and Bachrach, Yoram and Foerster, Jakob and Li, Xian and Fang, Han and Sukhbaatar, Sainbayar and Weston, Jason},
journal={arXiv preprint arXiv:2606.25996},
year={2026}
}A copy is also available in CITATION.bib.