让大模型根据把握度自动决定推理时长，省计算还更准。
Certainty-Guided Reasoning in Large Language Models: A Dynamic Thinking Budget Approach
- 根据答案概率动态判断是否够确定，达标就提前停止推理。
- 在AIME2025上保持准确率，减少数百万次令牌消耗。
- 适合需要控制成本又怕出错的推理任务，尤其看重风险规避。
大型推理语言模型通常采用固定推理预算，可能导致计算浪费或过早终止。我们提出无模型依赖的自适应推理方法——确定性引导推理（CGR），定期探测当前推理是否支持一个高置信度的最终答案；一旦达到目标置信度阈值，即提前终止；否则持续到思维结束标记或预算上限。置信度基于模型对答案词元的预测概率估算，实现轻量级停止判断。在AIME2025数据集上，CGR在保持基线准确率的同时显著降低令牌使用量，提供可调的置信度-效率权衡，累计可节省数百万令牌。64个随机种子下的实验显示其行为一致。我们还引入了Grade指标，惩罚错误答案并允许放弃回答，捕捉风险敏感性能。结果表明，当置信度不足时，CGR通过拒绝回答提升Grade表现。
原文摘要 · Abstract (English)
Large reasoning language models are typically run with fixed inference budgets, which can waste computation or terminate reasoning prematurely. We introduce Certainty-Guided Reasoning (CGR), a model-agnostic adaptive inference procedure that periodically probes whether the current reasoning supports a confident final answer and terminates early once a target certainty threshold is reached, otherwise continuing until the end-of-thinking token or the budget limit. Certainty is estimated from the model's predicted probabilities over the answer tokens, yielding a lightweight stopping criterion. On AIME2025, CGR preserves baseline accuracy while reducing token usage, providing a tunable certainty-efficiency trade-off that can eliminate millions of tokens in aggregate. Across 64 random seeds, CGR exhibits consistent behavior. We also introduce a Grade metric that penalizes incorrect answers and permits abstention, capturing risk-sensitive performance. Results show that CGR improves Grade by abstaining when certainty remains low.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。