Optimizing Instruction Synthesis: Effective Exploration of Evolutionary Space with Tree Search

TL;DR

This paper presents IDEA-MCTS, a scalable tree search framework that guides the evolution of instruction data to improve quality, diversity, and complexity, thereby enhancing language model alignment and instruction-following performance.

Contribution

Introduces IDEA-MCTS, a novel Monte Carlo Tree Search-based framework for controlled and efficient synthesis of high-quality instruction data.

Findings

Significant improvement in instruction data quality metrics.

Enhanced instruction-following accuracy in low-resource settings.

Effective guidance of instruction evolution using tree search.

Abstract

Instruction tuning is a crucial technique for aligning language models with humans' actual goals in the real world. Extensive research has highlighted the quality of instruction data is essential for the success of this alignment. However, creating high-quality data manually is labor-intensive and time-consuming, which leads researchers to explore using LLMs to synthesize data. Recent studies have focused on using a stronger LLM to iteratively enhance existing instruction data, showing promising results. Nevertheless, previous work often lacks control over the evolution direction, resulting in high uncertainty in the data synthesis process and low-quality instructions. In this paper, we introduce a general and scalable framework, IDEA-MCTS (Instruction Data Enhancement using Monte Carlo Tree Search), a scalable framework for efficiently synthesizing instructions. With tree search and evaluation models, it can efficiently guide each instruction to evolve into a high-quality form, aiding in instruction fine-tuning. Experimental results show that IDEA-MCTS significantly enhances the seed instruction data, raising the average evaluation scores of quality, diversity, and complexity from 2.19 to 3.81. Furthermore, in open-domain benchmarks, experimental results show that IDEA-MCTS improves the accuracy of real-world instruction-following skills in LLMs by an average of 5\% in low-resource settings.