Rethinking Optimal Verification Granularity for Compute-Efficient Test-Time Scaling

TL;DR

This paper investigates how the frequency of verification during test-time scaling affects large language model performance and efficiency, proposing a new adaptive method that improves accuracy and reduces computational costs.

Contribution

It introduces Variable Granularity Search (VG-Search), a novel algorithm that optimizes verification frequency during generation, enhancing compute efficiency and model scaling.

Findings

Adaptive VG-Search improves accuracy by up to 3.6%.

Reduces FLOPs by over 52%.

Dynamically selecting verification granularity enhances performance.

Abstract

Test-time scaling (TTS) has proven effective in enhancing the reasoning capabilities of large language models (LLMs). Verification plays a key role in TTS, simultaneously influencing (1) reasoning performance and (2) compute efficiency, due to the quality and computational cost of verification. In this work, we challenge the conventional paradigms of verification, and make the first attempt toward systematically investigating the impact of verification granularity-that is, how frequently the verifier is invoked during generation, beyond verifying only the final output or individual generation steps. To this end, we introduce Variable Granularity Search (VG-Search), a unified algorithm that generalizes beam search and Best-of-N sampling via a tunable granularity parameter g. Extensive experiments with VG-Search under varying compute budgets, generator-verifier configurations, and task attributes reveal that dynamically selecting g can improve the compute efficiency and scaling behavior. Building on these findings, we propose adaptive VG-Search strategies that achieve accuracy gains of up to 3.1\% over Beam Search and 3.6\% over Best-of-N, while reducing FLOPs by over 52\%. We will open-source the code to support future research.