README.md

metadata

license: other
license_name: causilo-1.0
license_link: LICENSE
library_name: causilo
pipeline_tag: tabular-classification

Causilo

Causilo is a pretrained tabular foundation model from Nums AI Inc., supporting classification and regression through a scikit-learn interface.

Technical report

Files

• classifier/config.json and classifier/model.safetensors: classification model.
• regressor/config.json and regressor/model.safetensors: regression model.

Installation

Python 3.10–3.12 and PyTorch 2.13+ are required.

pip install causilo

The first fit automatically downloads and caches the checkpoint.

from causilo import CausiloClassifier, CausiloRegressor

classifier = CausiloClassifier()
classifier.fit(X_train, y_train)
probabilities = classifier.predict_proba(X_test)

regressor = CausiloRegressor()
regressor.fit(X_train, y_train)
predictions = regressor.predict(X_test)

See the code repository for usage and benchmarks.

Licenses

• Python source and wheel: Apache-2.0; see the code repository.
• Model weights: Causilo License v1.0.
• Non-commercial research, testing, evaluation, experimentation and modifications are permitted subject to the License. Institutional affiliation alone does not determine whether a use is non-commercial.
• Free research redistribution of original copies and Derivatives is permitted under Section 3 without prior Company permission, with the required license, notices and modification information.
• Commercial or production use of the Model, Derivatives or Outputs requires separate licenses, subject to the scholarly-communication permission in Section 2(c).
• Hosted, managed, API or SaaS services require separate licenses whether paid or free. Hosting downloadable files under Section 3 is permitted.
• These points summarize the License; the full text controls.

Licensor: Nums AI Inc. Contact: contact@nums.world.

Citation

If you use Causilo in research, please cite:

@misc{cho2026causilotechnicalreport,
  title={Causilo Technical Report},
  author={Minyong Cho and Minho Jeong and Dooho Lee and Jinmo Lee and Jaemin Yoo},
  year={2026},
  eprint={2609.22866},
  archivePrefix={arXiv},
  primaryClass={cs.LG},
  url={https://arxiv.org/abs/2609.22866},
}