Time-R1：邁向大型語言模型中的全面時間推理

Time-R1: Towards Comprehensive Temporal Reasoning in LLMs

May 16, 2025

作者: Zijia Liu, Peixuan Han, Haofei Yu, Haoru Li, Jiaxuan You

cs.AI

摘要

大型語言模型（LLMs）展現了令人印象深刻的性能，但在時間智能方面仍顯不足，難以將對過去的推理與未來的預測及合理生成相結合。現有方法通常針對孤立的時間技能，如關於過去事件的問答或基本預測，且表現出較差的泛化能力，尤其是在處理超出其知識截止日期的事件或需要創造性遠見時。為解決這些限制，我們引入了Time-R1，這是首個賦予中等規模（30億參數）LLM全面時間能力的框架：理解、預測和創造性生成。我們的方法採用了一種新穎的三階段發展路徑；前兩個階段構成了一個由精心設計的基於規則的動態獎勵系統驅動的強化學習（RL）課程。該框架逐步建立（1）從歷史數據中獲得的基礎時間理解和邏輯事件-時間映射，（2）對超出其知識截止日期的未來事件預測能力，最終（3）實現了對創造性未來場景生成的顯著泛化，而無需任何微調。引人注目的是，實驗表明Time-R1在極具挑戰性的未來事件預測和創造性場景生成基準測試中，表現優於規模超過其200倍的模型，包括最先進的6710億參數的DeepSeek-R1。這項工作提供了強有力的證據，表明經過深思熟慮設計的漸進式RL微調可以使更小、更高效的模型實現卓越的時間性能，為真正具有時間意識的AI提供了一條實用且可擴展的路徑。為促進進一步研究，我們還發布了Time-Bench，這是一個基於10年新聞數據的大規模多任務時間推理數據集，以及我們的Time-R1檢查點系列。

English

Large Language Models (LLMs) demonstrate impressive capabilities but lack robust temporal intelligence, struggling to integrate reasoning about the past with predictions and plausible generations of the future. Meanwhile, existing methods typically target isolated temporal skills, such as question answering about past events or basic forecasting, and exhibit poor generalization, particularly when dealing with events beyond their knowledge cutoff or requiring creative foresight. To address these limitations, we introduce Time-R1, the first framework to endow a moderate-sized (3B-parameter) LLM with comprehensive temporal abilities: understanding, prediction, and creative generation. Our approach features a novel three-stage development path; the first two constitute a reinforcement learning (RL) curriculum driven by a meticulously designed dynamic rule-based reward system. This framework progressively builds (1) foundational temporal understanding and logical event-time mappings from historical data, (2) future event prediction skills for events beyond its knowledge cutoff, and finally (3) enables remarkable generalization to creative future scenario generation without any fine-tuning. Strikingly, experiments demonstrate that Time-R1 outperforms models over 200 times larger, including the state-of-the-art 671B DeepSeek-R1, on highly challenging future event prediction and creative scenario generation benchmarks. This work provides strong evidence that thoughtfully engineered, progressive RL fine-tuning allows smaller, efficient models to achieve superior temporal performance, offering a practical and scalable path towards truly time-aware AI. To foster further research, we also release Time-Bench, a large-scale multi-task temporal reasoning dataset derived from 10 years of news data, and our series of Time-R1 checkpoints.