Salvato in:
Dettagli Bibliografici
Autori principali: Yu, Yang, Han, Kai, Zhou, Hang, Tang, Yehui, Huang, Kaiqi, Wang, Yunhe, Tao, Dacheng
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:https://arxiv.org/abs/2507.16178
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866911069552771072
author Yu, Yang
Han, Kai
Zhou, Hang
Tang, Yehui
Huang, Kaiqi
Wang, Yunhe
Tao, Dacheng
author_facet Yu, Yang
Han, Kai
Zhou, Hang
Tang, Yehui
Huang, Kaiqi
Wang, Yunhe
Tao, Dacheng
contents While large-scale training data is fundamental for developing capable large language models (LLMs), strategically selecting high-quality data has emerged as a critical approach to enhance training efficiency and reduce computational costs. Current data selection methodologies predominantly rely on static, training-agnostic criteria, failing to account for the dynamic model training and data interactions. In this paper, we propose a new Data Weighting Model (DWM) to adjust the weight of selected data within each batch to achieve a dynamic data utilization during LLM training. Specially, to better capture the dynamic data preference of the trained model, a bi-level optimization framework is implemented to update the weighting model. Our experiments demonstrate that DWM enhances the performance of models trained with randomly-selected data, and the learned weighting model can be transferred to enhance other data selection methods and models of different sizes. Moreover, we further analyze how a model's data preferences evolve throughout training, providing new insights into the data preference of the model during training.
format Preprint
id arxiv_https___arxiv_org_abs_2507_16178
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle LLM Data Selection and Utilization via Dynamic Bi-level Optimization
Yu, Yang
Han, Kai
Zhou, Hang
Tang, Yehui
Huang, Kaiqi
Wang, Yunhe
Tao, Dacheng
Machine Learning
Artificial Intelligence
While large-scale training data is fundamental for developing capable large language models (LLMs), strategically selecting high-quality data has emerged as a critical approach to enhance training efficiency and reduce computational costs. Current data selection methodologies predominantly rely on static, training-agnostic criteria, failing to account for the dynamic model training and data interactions. In this paper, we propose a new Data Weighting Model (DWM) to adjust the weight of selected data within each batch to achieve a dynamic data utilization during LLM training. Specially, to better capture the dynamic data preference of the trained model, a bi-level optimization framework is implemented to update the weighting model. Our experiments demonstrate that DWM enhances the performance of models trained with randomly-selected data, and the learned weighting model can be transferred to enhance other data selection methods and models of different sizes. Moreover, we further analyze how a model's data preferences evolve throughout training, providing new insights into the data preference of the model during training.
title LLM Data Selection and Utilization via Dynamic Bi-level Optimization
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2507.16178