SAGE：用强化学习训练智能任意跨度智能体进行长视频推理
SAGE: Training Smart Any-Horizon Agents for Long Video Reasoning with Reinforcement Learning
把长视频推理建模为「按需决定看多少帧、看几轮」的多轮智能体决策，并用 RL 后训练教会模型自适应选择时间跨度；对做视频问答/长视频理解的人，这套 any-horizon 调度 + 合成数据管线可直接迁移到自己的视频 LLM 上以降低帧数开销。
原文依据 an agent system that performs multi-turn reasoning on long videos while handling simpler problems in a single turn
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论作为人类，我们天生就是任意跨度的推理者，也就是说，面对给定任务时，我们能够决定是迭代地快速浏览长视频，还是在必要时完整观看短视频。基于这一点，人们会期望视频推理模型能够在不同时长上灵活推理。然而，当前最先进的模型仍被训练为在处理大量帧的同时以单轮方式预测答案，类似于完整观看一整段长视频，需要消耗大量资源。这引出了一个问题：能否构建高性能的任意跨度视频推理系统？受人类行为启发，我们首先提出 SAGE，一个对长视频执行多轮推理、同时对较简单问题以单轮方式处理的智能体系统。其次，我们引入一条简便的合成数据生成流水线，使用 Gemini-2.5-Flash 训练编排器 SAGE-MM，它位于 SAGE 的核心。我们进一步提出一套有效的强化学习后训练方案，这是向 SAGE-MM 注入任意跨度推理能力的关键。第三，我们构建了 SAGE-Bench，其平均时长超过 700 秒，用于评估真实娱乐使用场景中的视频推理能力。最后，我们通过实验验证了系统、数据和强化学习方案的有效性，在开放式视频推理任务上观察到最高 6.1% 的显著提升，在时长超过 10 分钟的视频上更是取得 8.2% 的显著提升。我们将在论文发表后开源系统代码、数据和检查点。
英文原文摘要 1494 字符
As humans, we are natural any-horizon reasoners, i.e., we can decide whether to iteratively skim long videos or watch short ones in full when necessary for a given task. With this in mind, one would expect video reasoning models to reason flexibly across different durations. However, SOTA models are still trained to predict answers in a single turn while processing a large number of frames, akin to watching an entire long video, requiring significant resources. This raises the question: Is it possible to develop performant any-horizon video reasoning systems? Inspired by human behavior, we first propose SAGE, an agent system that performs multi-turn reasoning on long videos while handling simpler problems in a single turn. Secondly, we introduce an easy synthetic data generation pipeline using Gemini-2.5-Flash to train the orchestrator, SAGE-MM, which lies at the core of SAGE. We further propose an effective RL post-training recipe essential for instilling any-horizon reasoning ability in SAGE-MM. Thirdly, we curate SAGE-Bench with an average duration of greater than 700 seconds for evaluating video reasoning ability in real-world entertainment use cases. Lastly, we empirically validate the effectiveness of our system, data, and RL recipe, observing notable improvements of up to 6.1% on open-ended video reasoning tasks, as well as an impressive 8.2% improvement on videos longer than 10 minutes. We will open-source our system code, data, and checkpoints upon publication.