Importance-Aware Data Selection for Efficient LLM Instruction Tuning

TL;DR

This paper introduces MIWV, a new metric for selecting high-impact instruction data for LLM tuning, showing that using top data based on MIWV can outperform full dataset training.

Contribution

The paper proposes MIWV, a novel importance-aware metric for data selection in instruction tuning, improving efficiency and performance of LLMs.

Findings

Selecting top 1% data by MIWV outperforms full dataset training.

MIWV effectively identifies the most beneficial instruction data.

Empirical results validate the superiority of MIWV-based data selection.

Abstract

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.