titleMIRA: Mid-training Rubric Anchoring for Source-Aware Data Selection
authorsHaowen Wang, Yaxin Du, Jian Yang, Jiajun Wu, Shukai Liu, Yuxuan Zhang, Pingjie Wang, Siheng Chen, Tuney Zheng, Ming Zhou, Xianglong Liu, Bryan Dai
date2026-05-29
venuearXiv (preprint, not peer reviewed)
idarXiv:2605.30288
upvotes21 (Hugging Face Daily Papers, 2026-11-01)
linksabs pdf hf
abstractMid-training has become an important stage in modern LLM development, using large-scale curated mixtures to strengthen capabilities before final post-training. Its data selection problem is distinct: the data are optimized under a pretraining-style objective at near-pretraining scale, but are curated toward downstream capabilities and drawn from heterogeneous sources with different formats and training roles. As a result, effective selection requires both scalability and source-adaptive semantic criteria. Existing model-based methods scale well, but provide only implicit quality signals. Semantic selection methods offer stronger judgments, but usually assume fixed rubrics or standardized data formats. To address this mismatch, we propose MIRA, a source-aware filtering framework based on self-anchored rubric discovery. The key idea is to make rubric construction part of data selection: MIRA first discovers what should be evaluated for each source group, then distills those judgments into scalable student scorers for full-corpus filtering. On code-oriented mid-training with 21 sources and 5 source groups, MIRA outperforms selection baselines across nine code benchmarks and matches the full-corpus run while using only half the tokens.

cite

@misc{wang2026mira,
  title={MIRA: Mid-training Rubric Anchoring for Source-Aware Data Selection},
  author={Haowen Wang and Yaxin Du and Jian Yang and Jiajun Wu and Shukai Liu and Yuxuan Zhang and Pingjie Wang and Siheng Chen and Tuney Zheng and Ming Zhou and Xianglong Liu and Bryan Dai},
  year={2026},
  eprint={2605.30288},
  archivePrefix={arXiv},
  url={https://arxiv.org/abs/2605.30288}
}