Importance-Aware Data Selection for Efficient LLM Instruction Tuning

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Jiang, Tingyu, Li, Shen, Song, Yiyao, Zhang, Lan, Zhu, Hualei, Zhao, Yuan, Xu, Xiaohang, Taura, Kenjiro, Wang, Hao Henry
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915609541869568
author Jiang, Tingyu
Li, Shen
Song, Yiyao
Zhang, Lan
Zhu, Hualei
Zhao, Yuan
Xu, Xiaohang
Taura, Kenjiro
Wang, Hao Henry
author_facet Jiang, Tingyu
Li, Shen
Song, Yiyao
Zhang, Lan
Zhu, Hualei
Zhao, Yuan
Xu, Xiaohang
Taura, Kenjiro
Wang, Hao Henry
contents Instruction tuning plays a critical role in enhancing the performance and efficiency of Large Language Models (LLMs). Its success depends not only on the quality of the instruction data but also on the inherent capabilities of the LLM itself. Some studies suggest that even a small amount of high-quality data can achieve instruction fine-tuning results that are on par with, or even exceed, those from using a full-scale dataset. However, rather than focusing solely on calculating data quality scores to evaluate instruction data, there is a growing need to select high-quality data that maximally enhances the performance of instruction tuning for a given LLM. In this paper, we propose the Model Instruction Weakness Value (MIWV) as a novel metric to quantify the importance of instruction data in enhancing model's capabilities. The MIWV metric is derived from the discrepancies in the model's responses when using In-Context Learning (ICL), helping identify the most beneficial data for enhancing instruction tuning performance. Our experimental results demonstrate that selecting only the top 1\% of data based on MIWV can outperform training on the full dataset. Furthermore, this approach extends beyond existing research that focuses on data quality scoring for data selection, offering strong empirical evidence supporting the effectiveness of our proposed method.
format Preprint
id arxiv_https___arxiv_org_abs_2511_07074
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Importance-Aware Data Selection for Efficient LLM Instruction Tuning
Jiang, Tingyu
Li, Shen
Song, Yiyao
Zhang, Lan
Zhu, Hualei
Zhao, Yuan
Xu, Xiaohang
Taura, Kenjiro
Wang, Hao Henry
Computation and Language
Instruction tuning plays a critical role in enhancing the performance and efficiency of Large Language Models (LLMs). Its success depends not only on the quality of the instruction data but also on the inherent capabilities of the LLM itself. Some studies suggest that even a small amount of high-quality data can achieve instruction fine-tuning results that are on par with, or even exceed, those from using a full-scale dataset. However, rather than focusing solely on calculating data quality scores to evaluate instruction data, there is a growing need to select high-quality data that maximally enhances the performance of instruction tuning for a given LLM. In this paper, we propose the Model Instruction Weakness Value (MIWV) as a novel metric to quantify the importance of instruction data in enhancing model's capabilities. The MIWV metric is derived from the discrepancies in the model's responses when using In-Context Learning (ICL), helping identify the most beneficial data for enhancing instruction tuning performance. Our experimental results demonstrate that selecting only the top 1\% of data based on MIWV can outperform training on the full dataset. Furthermore, this approach extends beyond existing research that focuses on data quality scoring for data selection, offering strong empirical evidence supporting the effectiveness of our proposed method.
title Importance-Aware Data Selection for Efficient LLM Instruction Tuning
topic Computation and Language
url https://arxiv.org/abs/2511.07074