Attend Locally, Remember Linearly: Linear Attention as Cross-Frame Memory for Autoregressive Video Diffusion

• URL: http://arxiv.org/abs/2605.16579v2
• Date: Wed, 20 May 2026 19:35:18 GMT
• Title: Attend Locally, Remember Linearly: Linear Attention as Cross-Frame Memory for Autoregressive Video Diffusion
• Abstract summary: ARL2 is a hybrid attention module that replaces quadratic cross-frame attention with a fixed-size recurrent state.<n>We decompose self-attention into two branches: an intra-frame softmax branch for spatial detail and local dependencies, and an inter-frame recurrent linear branch that maintains a fixed-size state for streaming context.<n>With 75% of layers replaced by hybrid linear attention, the model achieves up to 2.26 wall-clock speedup and 54% memory reduction, while maintaining comparable quality with improving temporal consistency.
• Score: 61.57938553036056
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Autoregressive (AR) video diffusion is a powerful paradigm for streaming and interactive video generation. However, its reliance on softmax self-attention leads to quadratic compute complexity in sequence length and memory usage due to key-value caching, which limits its scalability to long video horizons. Existing remedies (e.g., sparse attention and KV-cache compression) reduce per-step cost but still rely on a linearly growing cache or irreversibly discard past context, and thus fail to address linear memory growth and streaming context management. To address this scalability bottleneck, we propose ARL2 (Attend Locally, Remember Linearly), a hybrid attention module that replaces quadratic cross-frame attention with a fixed-size recurrent state. We decompose self-attention into two branches: an intra-frame softmax branch for spatial detail and local dependencies, and an inter-frame gated recurrent linear branch that maintains a fixed-size state for streaming context. Our key insight is that softmax attention captures fine-grained local interactions, while a recurrent state provides controllable long-range memory. This design achieves linear-time scaling with constant memory while improving temporal consistency over the full-softmax model. To prevent noisy intermediate states from corrupting memory, we update the recurrent state only after the denoised pass. To avoid within-frame information asymmetry, all tokens share the same pre-update state rather than sequential updates. To the best of our knowledge, this is the first work to convert a pretrained AR video diffusion model into a hybrid linear attention architecture, through an efficient two-stage training scheme for AR video. With 75% of layers replaced by hybrid linear attention, the model achieves up to 2.26 wall-clock speedup and 54% memory reduction, while maintaining comparable quality with improving temporal consistency.

Related papers

• Online Neural Space Time Memory for Dynamic Novel View Synthesis [23.462185130690447]
  Test-Time Training (TTT) offers a powerful memory mechanism.<n>Standard models mandate gradient-based memory updates at every frame to adapt to the changing motion in dynamic scenes.<n>Our method demonstrates real-time, state-of-the-art performance on scenes with dynamic human motion as well as minute-scale online memorization.
  arXiv  Detail & Related papers  (2026-07-16T17:58:18Z)
• Towards Memory-Efficient Autoregressive Video Generation via Instance-Specific Parametric Absorption [37.96485836820005]
  Instance-Specific Parametric Absorption (ISPA) is a novel framework that shifts the KV cache compression from discarding to distilling.<n>We show ISPA can remove up to 50% of the KV cache with near-lossless visual quality.
  arXiv  Detail & Related papers  (2026-07-01T09:59:28Z)
• FadeMem: Distance-Aware Memory Consolidation for Autoregressive Video Diffusion [59.207505503284715]
  FadeMem organizes historical KV blocks into a temporal hierarchy under a fixed cache budget.<n>New history is inserted as fine-grained entries, while older adjacent entries are progressively merged.<n>Experiments show improved subject consistency, background stability, and temporal coherence over existing bounded-cache strategies.
  arXiv  Detail & Related papers  (2026-06-09T10:22:18Z)
• Linear Scaling Video VLMs for Long Video Understanding [26.67642601439313]
  Video encoders still rely on self-attention, causing compute and latency to grow quadratically with the number of frames.<n>We introduce StateKV, an inference-time method that adapts pretrained long-videoMs to linear-time video prefill.
  arXiv  Detail & Related papers  (2026-05-29T17:59:02Z)
• SWIFT: Prompt-Adaptive Memory for Efficient Interactive Long Video Generation [31.105007908298003]
  Streaming long-video generation faces a central challenge in continuous semantic switching.<n>Current approaches rely on cache rebuilding at prompt boundaries or fixed memory budgets.<n>We present SWIFT, a training-free framework for multi-prompt long-video generation.
  arXiv  Detail & Related papers  (2026-05-10T09:37:56Z)
• Going Down Memory Lane: Scaling Tokens for Video Stream Understanding with Dynamic KV-Cache Memory [50.30283773196725]
  Existing approaches rely on key-value caching to accumulate frame-level details over time, but use a limited number of tokens per frame.<n>We propose scaling the token budget to enable more granular-temporal understanding and reasoning.
  arXiv  Detail & Related papers  (2026-02-20T18:59:50Z)
• Fast Autoregressive Video Diffusion and World Models with Temporal Cache Compression and Sparse Attention [37.91838955436801]
  Autoregressive video diffusion models enable streaming generation, opening the door to long-form synthesis, video world models, and interactive neural game engines.<n>As generation progresses, the KV cache grows, causing both increasing latency and escalating GPU memory, which in turn restricts usable temporal context and harms long-range consistency.<n>We propose a unified, training-free attention framework for autoregressive diffusion: TempCache compresses the KV cache via temporal correspondence to bound cache growth; AnnCA accelerates cross-attention by selecting frame-relevant prompt tokens using fast approximate nearest neighbor matching; and AnnSA sparsifies self-attention by restricting each query
  arXiv  Detail & Related papers  (2026-02-02T08:31:21Z)
• GatedFWA: Linear Flash Windowed Attention with Gated Associative Memory [7.180426235884756]
  GatedFWA is a memory-underlineGated (underlineFlash) underlineWindowed underlineAttention mechanism.<n>It stabilizes memory updates and makes gradient flow controllable.<n>On language modelling benchmarks, GatedFWA delivers competitive throughput with negligible overhead.
  arXiv  Detail & Related papers  (2025-12-08T18:11:06Z)
• Look Every Frame All at Once: Video-Ma$^2$mba for Efficient Long-form Video Understanding with Multi-Axis Gradient Checkpointing [52.050036778325094]
  Video-Ma$2$mba is a novel architecture that incorporates State Space Models (SSMs) within the Mamba-2 framework.<n>Our approach significantly reduces the memory footprint compared to standard gradient checkpointing.<n>By maintaining a detailed capture of temporal dynamics, our model improves the accuracy and relevance of responses in long video understanding tasks.
  arXiv  Detail & Related papers  (2024-11-29T04:12:13Z)
• Ca2-VDM: Efficient Autoregressive Video Diffusion Model with Causal Generation and Cache Sharing [66.66090399385304]
  Ca2-VDM is an efficient autoregressive VDM with Causal generation and Cache sharing.<n>For causal generation, it introduces unidirectional feature computation, which ensures that the cache of conditional frames can be precomputed in previous autoregression steps.<n>For cache sharing, it shares the cache across all denoising steps to avoid the huge cache storage cost.
  arXiv  Detail & Related papers  (2024-11-25T13:33:41Z)
• LiVOS: Light Video Object Segmentation with Gated Linear Matching [116.58237547253935]
  LiVOS is a lightweight memory network that employs linear matching via linear attention. For longer and higher-resolution videos, it matched STM-based methods with 53% less GPU memory and supports 4096p inference on a 32G consumer-grade GPU.
  arXiv  Detail & Related papers  (2024-11-05T05:36:17Z)
• SparseTem: Boosting the Efficiency of CNN-Based Video Encoders by Exploiting Temporal Continuity [19.900719882624028]
  We propose a memory-efficient scheduling method to eliminate memory overhead and an online adjustment mechanism to minimize accuracy degradation.<n>SparseTem achieves speedup of 1.79x for EfficientDet and 4.72x for CRNN, with minimal accuracy drop and no additional memory overhead.
  arXiv  Detail & Related papers  (2024-10-28T07:13:25Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.