Learning When to Sample: Confidence-Aware Self-Consistency for Efficient LLM Chain-of-Thought Reasoning
This work addresses efficiency issues in LLM reasoning for applications like medical and math QA, though it is incremental as it builds on self-consistency methods.
The paper tackles the problem of high inference costs in large language models due to unnecessarily long reasoning paths in chain-of-thought reasoning, by introducing a confidence-aware decision framework that maintains comparable accuracy while using up to 80% fewer tokens.
Large language models (LLMs) achieve strong reasoning performance through chain-of-thought (CoT) reasoning, yet often generate unnecessarily long reasoning paths that incur high inference cost. Recent self-consistency-based approaches further improve accuracy but require sampling and aggregating multiple reasoning trajectories, leading to substantial additional computational overhead. This paper introduces a confidence-aware decision framework that analyzes a single completed reasoning trajectory to adaptively select between single-path and multi-path reasoning. The framework is trained using sentence-level numeric and linguistic features extracted from intermediate reasoning states in the MedQA dataset and generalizes effectively to MathQA, MedMCQA, and MMLU without additional fine-tuning. Experimental results show that the proposed method maintains accuracy comparable to multi-path baselines while using up to 80\% fewer tokens. These findings demonstrate that reasoning trajectories contain rich signals for uncertainty estimation, enabling a simple, transferable mechanism to balance accuracy and efficiency in LLM reasoning.