Step-level Verifier-guided Hybrid Test-Time Scaling for Large Language Models

TL;DR

This paper introduces a novel hybrid test-time scaling method for large language models that combines fine-grained sequential and parallel scaling techniques, significantly enhancing reasoning capabilities without additional training overhead.

Contribution

It proposes a new inference paradigm called Hybrid Test-Time Scaling, integrating step-level self-refinement with classical scaling methods for improved reasoning performance.

Findings

Effective across multiple LLMs and scales (3B-14B).

Significant performance improvements demonstrated.

Flexible combination of scaling methods enhances reasoning.

Abstract

Test-Time Scaling (TTS) is a promising approach to progressively elicit the model's intelligence during inference. Recently, training-based TTS methods, such as continued reinforcement learning (RL), have further surged in popularity, while training-free TTS methods are gradually fading from prominence. However, the additional computation overhead of training amplifies the burden on test-time scaling. In this paper, we focus on training-free TTS methods for reasoning. We first design Conditional Step-level Self-refinement, a fine-grained sequential scaling method guided by process verification. On top of its effectiveness, we further combine it with other classical parallel scaling methods at the step level, to introduce a novel inference paradigm called Hybrid Test-Time Scaling. Extensive experiments on five instruction-tuned LLMs across different scales (3B-14B) and families demonstrate that hybrid strategy incorporating various training-free TTS methods at a fine granularity has considerable potential for expanding the reasoning performance boundaries of LLMs.