EUVIMEDEUVIMED entdecken
Crossref

RICo: Refined In-Context Contribution for Automatic Instruction-Tuning Data Selection

Yixin Yang, Qingxiu Dong, Linli Yao, Fangwei Zhu, Weilin Luo, Bin Wang, Zhifang Sui

Proceedings of the AAAI Conference on Artificial Intelligence · 14.03.2026 · 40 (40) · 34349-34357

DOI: 10.1609/aaai.v40i40.40732 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Proceedings of the AAAI Conference on Artificial Intelligence, online erschienen am 14.03.2026. Verfasst von Yixin Yang und 6 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Data selection for instruction tuning is crucial for improving the performance of large language models (LLMs) while reducing training costs. In this paper, we propose Refined Contribution Measurement with In-Context Learning (RICo), a novel gradient-free method that quantifies the fine-grained contribution of individual samples to both task-level and global-level model performance. RICo enables more accurate identification of high-contribution data, leading to better instruction tuning. We also introduce a lightweight selection paradigm trained on RICo scores, enabling scalable data selection with strictly linear inference complexity. Extensive experiments on 3 LLMs across 12 benchmarks and 5 pairwise evaluation sets demonstrate the effectiveness of RICo. Remarkably, on LLaMA3.1-8B, models trained in 15% of RICo-selected data outperform full datasets by 5.42 percentage points and exceed the best performance of widely used selection methods by 1.48 percentage points. We further analyze high-contribution samples selected by RICo, which show both diverse tasks and appropriate difficulty levels, rather than merely the most difficult cases.

Zitieren

APA

Yang, Y., Dong, Q., Yao, L., Zhu, F., Luo, W., Wang, B., & Sui, Z. (2026). RICo: Refined In-Context Contribution for Automatic Instruction-Tuning Data Selection. Proceedings of the AAAI Conference on Artificial Intelligence, 40(40), 34349-34357. https://doi.org/10.1609/aaai.v40i40.40732

Vancouver

Yang Y, Dong Q, Yao L, Zhu F, Luo W, Wang B, et al. RICo: Refined In-Context Contribution for Automatic Instruction-Tuning Data Selection. Proceedings of the AAAI Conference on Artificial Intelligence. 2026;40(40):34349-34357. doi:10.1609/aaai.v40i40.40732

BibTeX
@article{Yang2026,
  author = {Yang, Yixin and Dong, Qingxiu and Yao, Linli and Zhu, Fangwei and Luo, Weilin and Wang, Bin and Sui, Zhifang},
  title = {RICo: Refined In-Context Contribution for Automatic Instruction-Tuning Data Selection},
  journal = {Proceedings of the AAAI Conference on Artificial Intelligence},
  year = {2026},
  volume = {40},
  number = {40},
  pages = {34349--34357},
  doi = {10.1609/aaai.v40i40.40732}
}
Kostenlos starten →