OmniSparse: Training-Aware Fine-Grained Sparse Attention for Long-Video MLLMs

• URL: http://arxiv.org/abs/2511.12201v2
• Date: Tue, 18 Nov 2025 23:07:41 GMT
• Title: OmniSparse: Training-Aware Fine-Grained Sparse Attention for Long-Video MLLMs
• Authors: Feng Chen, Yefei He, Shaoxuan He, Yuanyu He, Jing Liu, Lequan Lin, Akide Liu, Zhaoyang Li, Jiyuan Zhang, Zhenbang Sun, Bohan Zhuang, Qi Wu,
• Abstract summary: We introduce OmniSparse, a training-aware fine-grained sparse attention framework for long-video MLLMs.<n>Experiment results show that OmniSparse matches the performance of full attention while achieving up to 2.7x speedup during prefill and 2.4x memory reduction during decoding.
• Score: 43.78743496579736
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Existing sparse attention methods primarily target inference-time acceleration by selecting critical tokens under predefined sparsity patterns. However, they often fail to bridge the training-inference gap and lack the capacity for fine-grained token selection across multiple dimensions such as queries, key-values (KV), and heads, leading to suboptimal performance and limited acceleration gains. In this paper, we introduce OmniSparse, a training-aware fine-grained sparse attention framework for long-video MLLMs, which operates in both training and inference with dynamic token budget allocation. Specifically, OmniSparse contains three adaptive and complementary mechanisms: (1) query selection via lazy-active classification, retaining active queries that capture broad semantic similarity while discarding most lazy ones that focus on limited local context and exhibit high functional redundancy; (2) KV selection with head-level dynamic budget allocation, where a shared budget is determined based on the flattest head and applied uniformly across all heads to ensure attention recall; and (3) KV cache slimming to reduce head-level redundancy by selectively fetching visual KV cache according to the head-level decoding query pattern. Experimental results show that OmniSparse matches the performance of full attention while achieving up to 2.7x speedup during prefill and 2.4x memory reduction during decoding.

Related papers

• AdaSpot: Spend Resolution Where It Matters for Precise Event Spotting [59.31340724915079]
  Event Spotting is a key task for applications in sports analytics, robotics, and autonomous systems.<n>bfAdaSpot achieves state-of-the-art performance under strict evaluation metrics.
  arXiv  Detail & Related papers  (2026-02-25T16:24:48Z)
• Fast KVzip: Efficient and Accurate LLM Inference with Gated KV Eviction [50.99402504483692]
  We propose a novel gating-based KV cache eviction method for frozen-weight language models.<n>Our approach integrates seamlessly into both the prefill and decoding stages.<n>Experiments show that our method maintains near-lossless performance while evicting up to 70% of the KV cache.
  arXiv  Detail & Related papers  (2026-01-25T03:07:54Z)
• Fast SAM2 with Text-Driven Token Pruning [52.8350457627401]
  Segment Anything Model 2 (SAM2), a vision computation model has significantly advanced in prompt-driven video object segmentation.<n>SAM2 pipelines propagate all visual tokens produced by the image encoder through downstream temporal reasoning modules, regardless of their relevance to the target object.<n>We introduce a text-guided token pruning framework that improves inference efficiency by selectively reducing token density prior to temporal propagation.
  arXiv  Detail & Related papers  (2025-12-24T18:59:05Z)
• Training-free Context-adaptive Attention for Efficient Long Context Modeling [57.703159205740185]
  Training-free Context-adaptive Attention (TCA-Attention) is a training-free sparse attention mechanism that selectively attends to only the informative tokens for efficient long-context inference.<n>TCA-Attention achieves a 2.8$times$ speedup and reduces KV cache by 61% at 128K context length while maintaining performance comparable to full attention.
  arXiv  Detail & Related papers  (2025-12-10T01:54:57Z)
• Lethe: Layer- and Time-Adaptive KV Cache Pruning for Reasoning-Intensive LLM Serving [11.750209684686707]
  Generative reasoning with large language models (LLMs) often involves long decoding sequences.<n>We propose Lethe, a dynamic KV cache management framework that introduces adaptivity along both the spatial and temporal dimensions of decoding.<n>Lethe achieves a favorable balance between efficiency and generation quality across diverse models and tasks, increases throughput by up to 2.56x.
  arXiv  Detail & Related papers  (2025-11-08T14:52:43Z)
• DELTA: Dynamic Layer-Aware Token Attention for Efficient Long-Context Reasoning [6.468843780300177]
  We present textbfDELTA, a training-free sparse attention mechanism that achieves computational efficiency without sacrificing model accuracy.<n>Our results show that selective reuse of intermediate attention maps offers a robust path toward efficient long-context reasoning.
  arXiv  Detail & Related papers  (2025-10-10T21:37:49Z)
• VQL: An End-to-End Context-Aware Vector Quantization Attention for Ultra-Long User Behavior Modeling [12.619238878583703]
  In large-scale recommender systems, ultra-long user behavior sequences encode rich signals of evolving interests.<n>We propose VQL, a context-aware Vector Quantization Attention framework for ultra-long behavior modeling.
  arXiv  Detail & Related papers  (2025-08-23T19:58:18Z)
• TokenSelect: Efficient Long-Context Inference and Length Extrapolation for LLMs via Dynamic Token-Level KV Cache Selection [23.12497380673902]
  Dynamic Token-Level KV Cache Selection (TokenSelect) is a training-free method for efficient and accurate long-context inference.<n>A comprehensive evaluation of TokenSelect demonstrates up to $23.84times$ speedup in attention computation and up to $2.28times$ acceleration in end-to-end latency.
  arXiv  Detail & Related papers  (2024-11-05T07:56:24Z)
• S2-Attention: Hardware-Aware Context Sharding Among Attention Heads [49.1454481007861]
  Sparse attention selectively attends to a subset of tokens in the context.<n>It remains unclear whether sparse attention can maintain the model's quality at a scale of today's large language models.<n>This paper presents Sparsely-Sharded(S2) Attention, a Triton library that provides kernel optimization for sparse attention customizable at both per-head and per-context-range levels.
  arXiv  Detail & Related papers  (2024-07-25T00:27:07Z)
• Model Tells You What to Discard: Adaptive KV Cache Compression for LLMs [82.08922896531618]
  We introduce adaptive KV cache compression, a plug-and-play method that reduces the memory footprint of generative inference for Large Language Models (LLMs) We conduct targeted profiling to discern the intrinsic structure of attention modules. Based on the recognized structure, we then construct the KV cache in an adaptive manner: evicting long-range contexts on attention heads emphasizing local contexts, discarding non-special tokens on attention heads centered on special tokens, and only employing the standard KV cache for attention heads that broadly attend to all tokens.
  arXiv  Detail & Related papers  (2023-10-03T05:17:08Z)

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.