动态调整验证频率可显著提升大模型推理效率
Rethinking Optimal Verification Granularity for Compute-Efficient Test-Time Scaling
- 提出可调节验证粒度的统一算法，平衡验证频次与计算开销
- 在相同算力下，准确率最高提升3.6%，计算量减少超52%
- 适合追求高效推理的模型部署与测试时扩展场景
测试时扩展（TTS）已被证明能有效提升大语言模型的推理能力。验证在TTS中至关重要，同时影响推理性能和计算效率，取决于验证质量与计算成本。本文挑战传统验证范式，首次系统研究验证粒度——即生成过程中验证器调用频率，不仅限于最终输出或单步验证。为此，我们提出可变粒度搜索（VG-Search），通过可调参数g统一广义束搜索与Best-of-N采样。在不同算力预算、生成器-验证器配置及任务属性下的大量实验表明，动态选择粒度g可提升计算效率与扩展行为。基于此，我们提出自适应VG-Search策略，在准确率上相比束搜索提升最多3.1%，相比Best-of-N提升3.6%，同时减少超过52%的浮点运算量。代码将开源以支持后续研究。
原文摘要 · Abstract (English)
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.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。