Learning a Continue-Thinking Token for Enhanced Test-Time Scaling

TL;DR

This paper introduces a learned continue-thinking token for language models, which improves reasoning accuracy at test time by extending reasoning steps more effectively than fixed tokens.

Contribution

It proposes a novel approach to learn a dedicated continue-thinking token via reinforcement learning, enhancing test-time reasoning beyond fixed token methods.

Findings

Learned token improves accuracy on math benchmarks.

Achieves greater gains than fixed-token approaches.

Significantly boosts GSM8K benchmark performance.

Abstract

Test-time scaling has emerged as an effective approach for improving language model performance by utilizing additional compute at inference time. Recent studies have shown that overriding end-of-thinking tokens (e.g., replacing "</think>" with "Wait") can extend reasoning steps and improve accuracy. In this work, we explore whether a dedicated continue-thinking token can be learned to trigger extended reasoning. We augment a distilled version of DeepSeek-R1 with a single learned "<|continue-thinking|>" token, training only its embedding via reinforcement learning while keeping the model weights frozen. Our experiments show that this learned token achieves improved accuracy on standard math benchmarks compared to both the baseline model and a test-time scaling approach that uses a fixed token (e.g., "Wait") for budget forcing. In particular, we observe that in cases where the fixed-token approach enhances the base model's accuracy, our method achieves a markedly greater improvement. For example, on the GSM8K benchmark, the fixed-token approach yields a 1.3% absolute improvement in accuracy, whereas our learned-token method achieves a 4.2% improvement over the base model that does not use budget forcing.