SPaR: Self-Play with Tree-Search Refinement to Improve Instruction-Following in Large Language Models

TL;DR

SPaR introduces a self-play, tree-search refinement method that improves instruction-following in large language models by generating more accurate and relevant preference pairs, leading to better performance without sacrificing general capabilities.

Contribution

The paper presents a novel self-play framework with tree-search refinement that enhances instruction-following in large language models by reducing irrelevant content variations.

Findings

LLaMA3-8B trained with SPaR outperforms GPT-4-Turbo on IFEval.

SPaR improves performance of larger models like GLM-4-9B and LLaMA3-70B.

Tree search scaling affects model performance during inference.

Abstract

Instruction-following is a fundamental capability of language models, requiring the model to recognize even the most subtle requirements in the instructions and accurately reflect them in its output. Such an ability is well-suited for and often optimized by preference learning. However, existing methods often directly sample multiple independent responses from the model when creating preference pairs. Such practice can introduce content variations irrelevant to whether the instruction is precisely followed (e.g., different expressions about the same semantic), interfering with the goal of teaching models to recognize the key differences that lead to improved instruction following. In light of this, we introduce SPaR, a self-play framework integrating tree-search self-refinement to yield valid and comparable preference pairs free from distractions. By playing against itself, an LLM employs a tree-search strategy to refine its previous responses with respect to the instruction while minimizing unnecessary variations. Our experiments show that a LLaMA3-8B model, trained over three iterations guided by SPaR, surpasses GPT-4-Turbo on the IFEval benchmark without losing general capabilities. Furthermore, SPaR demonstrates promising scalability, greatly enhancing models like GLM-4-9B and LLaMA3-70B. We also identify how inference scaling in tree search would impact model performance. Our code and data are publicly available at https://github.com/thu-coai/SPaR.