边看边想：面向大型视觉语言模型的流式思维链推理
Think-as-You-See: Streaming Chain-of-Thought Reasoning for Large Vision-Language Models
把视频 LVLM 的思维链从「看完再想」改成「边看边想」，用流式注意力掩码、位置编码和并行 KV 缓存实现帧到达即推理，可直接迁移到任何需要低延迟视频问答或在线视频理解的视觉语言模型上。
原文依据 we introduce a streaming reasoning paradigm for LVLMs, where reasoning unfolds sequentially with incoming frames
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论大型视觉语言模型（LVLM）在思维链（CoT）推理方面已展现出卓越能力。然而，现有 LVLM 推理范式仅在整个视频可用之后才开始推理，这引入了不必要的延迟，并削弱了对动态场景中早期视觉线索的注意力。受人类边看边思考能力的启发，我们为 LVLM 引入一种流式推理范式：推理随输入帧的到来而逐步展开，并在完整视频被观测后进一步深化。我们通过 Think-as-You-See（TaYS）实例化该范式，这是一个统一框架，通过整合流式思维链生成、流约束训练与流并行推理，使 LVLM 能够边看边推理。具体而言，TaYS 采用时间对齐的流式推理单元并配以精确的思维链监督，通过流式注意力掩码与位置编码强制推理按序进行，并利用并行 KV 缓存机制将输入编码与推理生成解耦，从而确保对齐与真正的并发。我们在 Qwen2.5-VL 模型系列上，针对具有代表性的视频思维链任务（包括事件动态分析、因果推理与主题理解）对 TaYS 进行了评测。实验结果表明，与批处理模式的思维链相比，TaYS 取得了更优的推理性能，同时将推理前延迟降至一秒以内，并将整体答案延迟降低 50% 以上。这些发现证明了流式范式在使 LVLM 实现实时、类人推理方面的有效性。
英文原文摘要 1587 字符
Large Vision Language Models (LVLMs) have demonstrated remarkable capabilities in Chain-of-Thought (CoT) reasoning. However, existing LVLM reasoning paradigms only begin reasoning after the entire video becomes available, introducing unnecessary latency and diminishing attention to early visual cues in dynamic scenes. Inspired by the human ability to think while watching, we introduce a streaming reasoning paradigm for LVLMs, where reasoning unfolds sequentially with incoming frames and deepens after the full video is observed. We instantiate this paradigm through Think-as-You-See (TaYS), a unified framework that enables LVLMs to reason while watching by integrating streaming CoT generation, stream-constrained training, and stream-parallel inference. Specifically, TaYS employs temporally aligned streaming reasoning units with precise CoT supervision, enforces ordered reasoning via streaming attention masks and positional encodings, and utilizes a parallel KV caches mechanism that decouples input encoding from reasoning generation, ensuring alignment and true concurrency. We evaluate TaYS on the Qwen2.5-VL model family across representative video CoT tasks, including event dynamics analysis, causal reasoning, and thematic understanding. Experimental results show that TaYS achieves superior reasoning performance compared with batch-mode CoT, while reducing pre-reasoning latency to under one second and overall answer delay by more than 50%. These findings demonstrate the effectiveness of the streaming paradigm in enabling real-time, human-like reasoning for LVLMs.