Control-R: Towards controllable test-time scaling

TL;DR

This paper introduces Control-R, a novel approach for test-time controllable reasoning in large models, using structured control signals and a new dataset to improve complex problem-solving performance.

Contribution

The paper presents Reasoning Control Fields (RCF) and Control-R-4K dataset, enabling test-time adjustment of reasoning effort in large models, with a new finetuning method for better control.

Findings

Achieves state-of-the-art results on AIME2024 and MATH500 benchmarks.

Enables controllable reasoning process during inference.

Improves reasoning efficiency and accuracy with test-time control.

Abstract

This paper target in addressing the challenges of underthinking and overthinking in long chain-of-thought (CoT) reasoning for Large Reasoning Models (LRMs) by introducing Reasoning Control Fields (RCF)--a novel test-time approach that injects structured control signals to guide reasoning from a tree search perspective. RCF enables models to adjust reasoning effort according to given control conditions when solving complex tasks. Additionally, we present the Control-R-4K dataset, which consists of challenging problems annotated with detailed reasoning processes and corresponding control fields. To further enhance reasoning control, we propose a Conditional Distillation Finetuning (CDF) method, which trains model--particularly Control-R-32B--to effectively adjust reasoning effort during test time. Experimental results on benchmarks such as AIME2024 and MATH500 demonstrate that our approach achieves state-of-the-art performance at the 32B scale while enabling a controllable Long CoT reasoning process (L-CoT). Overall, this work introduces an effective paradigm for controllable test-time scaling reasoning.