ChronoStitch: Training-Free Composition of Visual KV Memories for Long-Horizon Temporal Reasoning

2026-07-21 · Santiram Tiwari, Nishant Sinha, Kunal Kislay arxiv

Long-video question answering requires a model to preserve visual evidence over time without repeatedly reprocessing the same video. A practical approach is to store the vision-language model's internal key-value (KV) cache for each video chunk and retrieve that state at query time. However, independently cached video chunks do not compose correctly: every chunk is prefilled from local rotary position zero, so naive concatenation collides temporal phases and removes the global order required for questions about what happened first, how often events occurred, or what changed across the video. This paper presents ChronoStitch, a training-free method for composing independently stored visual KV memories. The method first re-bases stored post-rotary keys onto a global three-axis multimodal RoPE coordinate system that preserves time, height, and width structure. We show why a one-dimensional scalar re-indexing is geometrically inconsistent for visual tokens because it turns spatial order within a frame into false temporal displacement. We then address the residual content gap left by positional repair: later chunks were originally encoded without attending to earlier chunks. ChronoStitch therefore selectively recomputes a small fraction of high-deviation later-chunk visual tokens while allowing them to attend over the composed cache. On Qwen2.5-VL-3B and the temporal split of TempCompass, ChronoStitch outperforms naive composition and position-only variants, improving event-ordering accuracy while running 3.3x faster than full joint re-prefilling.

📄 PDF Abstract BibTeX arXiv:2607.19547

Code (0)

등록된 구현이 없습니다.

Tasks

Video Question Answering

Flame3D: Zero-shot Compositional Reasoning of 3D Scenes with Agentic Language Models

2026-05-09 · Sagar Bharadwaj, Ziyong Ma, Anurag Ghosh, Srinivasan Seshan 외 arxiv

3D scene understanding spans reasoning about free space, object grounding, hypothetical object insertions, complex geometric relationships, and integrating all of these with external tools and data sources. Existing 3D u…

Scene Understanding

Learning from Execution: Self-Evolving Memory for Private-Library Code Generation

2026-04-27 · Mofei Li, Taozhi Chen, Guowei Yang, Jia Li arxiv

Large Language Models (LLMs) have achieved strong performance on general code generation, but their effectiveness drops sharply in enterprise settings where software development relies on internal private libraries absen…

Code Generation

Co-Evolving Graph and Text Memory for Training-Free Multi-Hop Question Answering

2026-07-25 · Hieu Man, Thien Huu Nguyen arxiv

Multi-hop question answering requires coordinating relational and textual evidence across reasoning steps, a combination neither a text corpus nor a knowledge graph can supply alone. Prior work often emphasizes only part…

Multi-hop Question Answering

ChainMPQ: Interleaved Text-Image Reasoning Chains for Mitigating Relation Hallucinations

2025-10-07 · Yike Wu, Yiwei Wang, Yujun Cai arxiv

While Large Vision-Language Models (LVLMs) achieve strong performance in multimodal tasks, hallucinations continue to hinder their reliability. Among the three categories of hallucinations, which include object, attribut…

Relational Reasoning

Transformer Feed-Forward Layers Are Key-Value Memories

2020-12-29 · EMNLP 2021 11 · Mor Geva, Roei Schuster, Jonathan Berant, Omer Levy

Feed-forward layers constitute two-thirds of a transformer model's parameters, yet their role in the network remains under-explored. We show that feed-forward layers in transformer-based language models operate as key-va…