EDUMATH_12b

EDUMATH 12B

KTO-trained version of the Gemma 3 12B IT SFT model as introduced in EDUMATH: Generating Standards-aligned Educational Math Word Problems. See our project repo for usage and our paper for training details/metrics.

Citation

@misc{christ2026edumathgeneratingstandardsalignededucational,
      title={EDUMATH: Generating Standards-aligned Educational Math Word Problems}, 
      author={Bryan R. Christ and Penelope Molitz and Beau LeBlond and Zachary Gottesman and Jonathan Kropko and Thomas Hartvigsen},
      year={2026},
      eprint={2510.06965},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2510.06965}, 
}