Experiments show that CALM reduces token consumption by 18.6%-25.0% compared with full CoT reasoning while maintaining performance, and can even improve accuracy by 0.07%-1.48% by reducing redundant reasoning.
Abstract
Large reasoning models (LRMs) have achieved strong performance on challenging mathematical and scientific tasks through long chain-of-thought reasoning, but they often suffer from overthinking with unnecessary reasoning steps. Existing truncation-based methods typically detect consecutive trial-answer agreement and force early exit, yet such local stopping signals do not necessarily indicate final-answer stability and can disrupt useful subsequent reflection or verification, potentially leading to performance degradation. We propose consistency-aware logit masking (CALM), a training-free decoding intervention for mitigating overthinking in LRMs. CALM first monitors trial answers at step boundaries and then applies logit masking to action transition point (ATP) tokens such as"Wait","Alternatively", and"Hmm", suppressing redundant transitional reasoning while allowing the model to continue its thought process more naturally. Experiments across four advanced LRMs and five benchmarks show that CALM reduces token consumption by 18.6%-25.0% compared with full CoT reasoning while maintaining performance, and can even improve accuracy by 0.07%-1.48% by reducing redundant reasoning. LLM-as-a-Judge results also demonstrate that CALM provides more coherent and concise results.
This paper proposes LLM Introspection, a practical inference-time reasoning control method that leverages keyword-token logits to enable early stopping during the reasoning process, and integrates the method into a vLLM-based proxy, providing a practical, plug-and-play deployment path.
Kai-Hao Zhang, Bin Shi, Kai Sun et al.· IEEE Transactions on Audio,...· 0 citations
While Large Reasoning Models (LRMs) excel at complex reasoning, alignment through reinforcement learning often induces systemic overconfidence. In production environments, where logits may be unavailable, robust black-box uncertainty quantification (UQ) is essential for trustworthiness and safety. Focusing on question-...
Lucas Biechy, Cédric Eichler, Adrien Boiret et al.· arXiv.org· 0 citations
Large Reasoning Models (LRMs) produce explicit chains of thought to improve transparency, but they frequently overthink, generating long yet inefficient reasoning chains that inflate computational costs and risk introducing errors. While numerous reinforcement learning methods struggle to penalize verbosity to achieve...
Cong Jiang, Xiao-Feng Zhang, Tomas Ko et al.· 0 citations
Reasoning large language models (LLMs) benefit from test-time scaling, motivating adaptive policies that spend tokens when they help and stop early otherwise. Existing rules halt when a confidence or stability signal crosses a threshold, but such thresholds rarely transfer across models and datasets, and validation-cal...
Group-Relative Policy Optimization (GRPO) is a key technique for training large reasoning models, but its reliance on outcome rewards makes it vulnerable to Think-Answer Mismatch, where final-answer correctness does not faithfully reflect reasoning quality. This vulnerability is amplified in highly imbalanced response...
Dan-Hao Zhu, Pei-Jun Shen, Si Shen· Memetic Computing· 0 citations
This paper studies this application as Rationale Contrastive Decoding (RCD), which contrasts the next-token distributions conditioned on the query with and without its rationale, amplifying tokens whose likelihood increases due to the rationale.
Shin'ya Yamaguchi· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.