让大模型学会按时间预算调整推理策略，提升工具调用效率。
Timely Machine: Awareness of Time Makes Test-Time Scaling Agentic
- 以实际耗时替代生成长度定义测试时扩展，动态适应不同延迟场景。
- 小模型在快速反馈下通过多轮交互胜出，大模型在高延迟中靠质量占优。
- 提出强化学习方法，让模型自动适配时间限制，适合复杂任务的智能体应用。
随着大语言模型日益处理复杂推理任务，测试时扩展对能力提升至关重要。但在频繁调用工具的智能体场景中，传统基于生成长度的定义失效：工具延迟使推理时间与生成长度解耦。本文提出 Timely Machine，将测试时重新定义为实际耗时，使模型能根据时间预算动态调整策略。我们构建了 Timely-Eval 基准，涵盖高频、低频工具调用及时间受限推理。通过改变工具延迟，发现小模型在快速反馈下凭借更多交互表现更优，而大模型在高延迟环境下因交互质量更高占据优势。此外，现有模型无法有效适配时间预算。为此我们提出 Timely-RL，先经冷启动监督微调，再用强化学习优化时间规划。Timely-RL 显著提升时间预算感知能力，在 Timely-Eval 上持续提升性能。本工作为智能体时代的测试时扩展提供了新视角。
原文摘要 · Abstract (English)
As large language models (LLMs) increasingly tackle complex reasoning tasks, test-time scaling has become critical for enhancing capabilities. However, in agentic scenarios with frequent tool calls, the traditional generation-length-based definition breaks down: tool latency decouples inference time from generation length. We propose Timely Machine, redefining test-time as wall-clock time, where models dynamically adjust strategies based on time budgets. We introduce Timely-Eval, a benchmark spanning high-frequency tool calls, low-frequency tool calls, and time-constrained reasoning. By varying tool latency, we find smaller models excel with fast feedback through more interactions, while larger models dominate high-latency settings via superior interaction quality. Moreover, existing models fail to adapt reasoning to time budgets. We propose Timely-RL to address this gap. After cold-start supervised fine-tuning, we use reinforcement learning to enhance temporal planning. Timely-RL improves time budget awareness and consistently boosts performance across Timely-Eval. We hope our work offers a new perspective on test-time scaling for the agentic era.
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。