LEAD: Iterative Data Selection for Efficient LLM Instruction Tuning
Proceedings of the VLDB Endowment · 01.11.2025 · 19 (3) · 426-439
DOI: 10.14778/3778092.3778103 ↗Kurz gefasst
Fachartikel in der Zeitschrift Proceedings of the VLDB Endowment, online erschienen am 18.03.2026. Verfasst von Xiaotian Lin und 6 weiteren Autor:innen. Die Arbeit wurde bisher 5-mal zitiert (Crossref) und nennt 76 Literaturangaben.
Abstract (Originalsprache)
Instruction tuning has emerged as a critical paradigm for improving the capabilities and alignment of large language models (LLMs). However, existing iterative model-aware data selection methods incur significant computational overhead, as they rely on repeatedly performing full-dataset model inference to estimate sample utility for subsequent training iterations. In this paper, we propose LEAD, a framework that LEArns to select Data iteratively by accurately estimating sample utility entirely within the standard training loop, eliminating the need for additional model inference. At its core, LEAD introduces Instance-Level Dynamic Uncertainty (IDU), a theoretically grounded utility function combining instantaneous training loss, gradient-based approximation of loss changes, and exponential smoothing of historical loss signals. To further scale efficiently to large datasets, LEAD employs a two-stage, coarse-to-fine selection strategy, adaptively prioritizing informative clusters through a multi-armed bandit mechanism, followed by precise fine-grained selection of high-utility samples using IDU. Extensive experiments across four diverse benchmarks show that LEAD significantly outperforms state-of-the-art methods, improving average model performance by 6.1%-10.8% while using only 2.5% of the training data and reducing overall training time by 5–10×.
Zitieren
APA
Lin, X., Qi, Y., Zhu, Y., Palpanas, T., Chai, C., Tang, N., & Luo, Y. (2025). LEAD: Iterative Data Selection for Efficient LLM Instruction Tuning. Proceedings of the VLDB Endowment, 19(3), 426-439. https://doi.org/10.14778/3778092.3778103
Vancouver
Lin X, Qi Y, Zhu Y, Palpanas T, Chai C, Tang N, et al. LEAD: Iterative Data Selection for Efficient LLM Instruction Tuning. Proceedings of the VLDB Endowment. 2025;19(3):426-439. doi:10.14778/3778092.3778103
BibTeX
@article{Lin2025,
author = {Lin, Xiaotian and Qi, Yanlin and Zhu, Yizhang and Palpanas, Themis and Chai, Chengliang and Tang, Nan and Luo, Yuyu},
title = {LEAD: Iterative Data Selection for Efficient LLM Instruction Tuning},
journal = {Proceedings of the VLDB Endowment},
year = {2025},
volume = {19},
number = {3},
pages = {426--439},
doi = {10.14778/3778092.3778103}
}
EUVIMED