ClusterUCB: Efficient Gradient-Based Data Selection for Targeted Fine-Tuning of LLMs

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Wang, Zige, Zhu, Qi, Mi, Fei, Xu, Minghui, Jin, Ruochun, Yang, Wenjing
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866909646802911232
author Wang, Zige
Zhu, Qi
Mi, Fei
Xu, Minghui
Jin, Ruochun
Yang, Wenjing
author_facet Wang, Zige
Zhu, Qi
Mi, Fei
Xu, Minghui
Jin, Ruochun
Yang, Wenjing
contents Gradient-based data influence approximation has been leveraged to select useful data samples in the supervised fine-tuning of large language models. However, the computation of gradients throughout the fine-tuning process requires too many resources to be feasible in practice. In this paper, we propose an efficient gradient-based data selection framework with clustering and a modified Upper Confidence Bound (UCB) algorithm. Based on the intuition that data samples with similar gradient features will have similar influences, we first perform clustering on the training data pool. Then, we frame the inter-cluster data selection as a constrained computing budget allocation problem and consider it a multi-armed bandit problem. A modified UCB algorithm is leveraged to solve this problem. Specifically, during the iterative sampling process, historical data influence information is recorded to directly estimate the distributions of each cluster, and a cold start is adopted to balance exploration and exploitation. Experimental results on various benchmarks show that our proposed framework, ClusterUCB, can achieve comparable results to the original gradient-based data selection methods while greatly reducing computing consumption.
format Preprint
id arxiv_https___arxiv_org_abs_2506_10288
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle ClusterUCB: Efficient Gradient-Based Data Selection for Targeted Fine-Tuning of LLMs
Wang, Zige
Zhu, Qi
Mi, Fei
Xu, Minghui
Jin, Ruochun
Yang, Wenjing
Computation and Language
Artificial Intelligence
Machine Learning
Gradient-based data influence approximation has been leveraged to select useful data samples in the supervised fine-tuning of large language models. However, the computation of gradients throughout the fine-tuning process requires too many resources to be feasible in practice. In this paper, we propose an efficient gradient-based data selection framework with clustering and a modified Upper Confidence Bound (UCB) algorithm. Based on the intuition that data samples with similar gradient features will have similar influences, we first perform clustering on the training data pool. Then, we frame the inter-cluster data selection as a constrained computing budget allocation problem and consider it a multi-armed bandit problem. A modified UCB algorithm is leveraged to solve this problem. Specifically, during the iterative sampling process, historical data influence information is recorded to directly estimate the distributions of each cluster, and a cold start is adopted to balance exploration and exploitation. Experimental results on various benchmarks show that our proposed framework, ClusterUCB, can achieve comparable results to the original gradient-based data selection methods while greatly reducing computing consumption.
title ClusterUCB: Efficient Gradient-Based Data Selection for Targeted Fine-Tuning of LLMs
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2506.10288