M* introduces the Walk Graph abstraction to serve arbitrary compositions of multimodal model components and reports latency and throughput gains over vLLM-Omni and other baselines on text-to-image, text-to-speech, and robotic planning workloads.
hub
Spargeattn: Accurate sparse attention accelerat- ing any model inference
19 Pith papers cite this work. Polarity classification is still indexing.
19
Pith papers citing it
abstract
Diffusion Transformers (DiTs) with 3D full attention power state-of-the-art video generation, but suffer from prohibitive compute cost -- when generating just a 5-second 720P video, attention alone takes 800 out of 945 seconds of total inference time. This paper introduces sliding tile attention (STA) to address this challenge. STA leverages the observation that attention scores in pretrained video diffusion models predominantly concentrate within localized 3D windows. By sliding and attending over the local spatial-temporal region, STA eliminates redundancy from full attention. Unlike traditional token-wise sliding window attention (SWA), STA operates tile-by-tile with a novel hardware-aware sliding window design, preserving expressiveness while being hardware-efficient. With careful kernel-level optimizations, STA offers the first efficient 2D/3D sliding-window-like attention implementation, achieving 58.79% MFU. Precisely, STA accelerates attention by 2.8-17x over FlashAttention-2 (FA2) and 1.6-10x over FlashAttention-3 (FA3). On the leading video DiT, HunyuanVideo, STA reduces end-to-end latency from 945s (FA3) to 685s without quality degradation, requiring no training. Enabling finetuning further lowers latency to 268s with only a 0.09% drop on VBench. We make our codebase public at https://github.com/hao-ai-lab/FastVideo.
hub tools
citation-role summary
background 2
method 1
citation-polarity summary
representative citing papers
DFSAttn is a training-free framework for dynamic fine-grained sparse attention in video DiTs that achieves up to 2.1x speedup while preserving generation quality via Hilbert reordering, hierarchical scoring, and adaptive caching.
FrameDiT proposes Matrix Attention for DiTs to achieve SOTA video generation with improved temporal coherence and efficiency comparable to local factorized attention.
SVG2 accelerates DiT video generation via semantic-aware token permutation with k-means, achieving up to 2.3x speedup and PSNR of 30 while fixing position-based clustering and scattered-token waste.
Joint few-step distillation and step-specific structural pruning turns a video diffusion model into a compact Mixture-of-Models that cuts 24% extra FLOPs per step and reaches 30× speedup on Wan-14B.
EcoVideo introduces entropy-driven dynamic frame selection for cloud-edge DiT video generation, yielding up to 2.9x speedup with adaptive keyframe budgets.
Light Interaction accelerates interactive video world models up to 2.59x via adaptive context management, denoising cache acceleration, and 3D block sparse attention without retraining.
LVSA is a training-free block-sparse attention technique combining structured windows with rotating global anchors that reduces inference compute 2.98-3.33x on video diffusion models at extended horizons while remaining quality-neutral or positive.
PhyMotion scores generated human videos by grounding recovered 3D poses in a physics simulator across kinematic, contact, and dynamic axes, yielding stronger human correlation and larger RL post-training gains than prior 2D rewards.
LIVEditor-14B applies a new sparse attention method (ISA) that prunes context and uses query-sharpness routing to cut attention latency ~60% with no loss in editing quality on standard benchmarks.
DynamicRad achieves 1.7x-2.5x inference speedups in long video diffusion with over 80% sparsity by grounding adaptive selection in a radial locality prior, using dual-mode static/dynamic strategies and offline BO with a semantic motion router.
AdaCluster delivers a training-free adaptive query-key clustering framework for sparse attention in video DiTs, yielding 1.67-4.31x inference speedup with negligible quality loss on CogVideoX-2B, HunyuanVideo, and Wan-2.1.
Hybrid Forcing combines linear temporal attention for long-range retention, block-sparse attention for efficiency, and decoupled distillation to achieve real-time unbounded 832x480 streaming video generation at 29.5 FPS.
SURF accelerates high-resolution video generation up to 12.5x by using noise reshifting for low-res previews from pretrained models and a shifting-window Refiner for efficient upscaling that retains original signatures.
HunyuanVideo 1.5 delivers state-of-the-art open-source text-to-video and image-to-video generation with an 8.3B parameter DiT model featuring SSTA attention, glyph-aware encoding, and progressive training.
PASA uses curvature-aware dynamic budgeting, grouped approximations, and stochastic attention routing to accelerate video diffusion transformers while eliminating temporal flickering from sparse patterns.
OASIS reduces redundancy in diffusion models for real-world video super-resolution via attention specialization routing and progressive training, delivering state-of-the-art quality with 6.2x faster inference than prior one-step baselines.
OSP-Next reports 83.73% VBench score and up to 2.27x speedup via hybrid sparse attention, SSP parallelism, HiF8 quantization, and Mix-GRPO on diffusion transformers.
citing papers explorer
-
M*: A Modular, Extensible, Serving System for Multimodal Models
M* introduces the Walk Graph abstraction to serve arbitrary compositions of multimodal model components and reports latency and throughput gains over vLLM-Omni and other baselines on text-to-image, text-to-speech, and robotic planning workloads.
-
DFSAttn: Dynamic Fine-grained Sparse Attention for Efficient Video Generation
DFSAttn is a training-free framework for dynamic fine-grained sparse attention in video DiTs that achieves up to 2.1x speedup while preserving generation quality via Hilbert reordering, hierarchical scoring, and adaptive caching.
-
FrameDiT: Diffusion Transformer with Matrix Attention for Efficient Video Generation
FrameDiT proposes Matrix Attention for DiTs to achieve SOTA video generation with improved temporal coherence and efficiency comparable to local factorized attention.
-
Sparse VideoGen2: Accelerate Video Generation with Sparse Attention via Semantic-Aware Permutation
SVG2 accelerates DiT video generation via semantic-aware token permutation with k-means, achieving up to 2.3x speedup and PSNR of 30 while fixing position-based clustering and scattered-token waste.
-
Dynamic-in-Few-Step: Unifying Dynamic Computation and Few-Step Distillation for Efficient Video Generation
Joint few-step distillation and step-specific structural pruning turns a video diffusion model into a compact Mixture-of-Models that cuts 24% extra FLOPs per step and reaches 30× speedup on Wan-14B.
-
EcoVideo: Entropy-Orchestrated Video Generation Paradigm in Cloud-Edge Dynamics
EcoVideo introduces entropy-driven dynamic frame selection for cloud-edge DiT video generation, yielding up to 2.9x speedup with adaptive keyframe budgets.
-
Light Interaction: Training-Free Inference Acceleration for Interactive Video World Models
Light Interaction accelerates interactive video world models up to 2.59x via adaptive context management, denoising cache acceleration, and 3D block sparse attention without retraining.
-
LVSA: Training-Free Sparse Attention for Long Video Diffusion
LVSA is a training-free block-sparse attention technique combining structured windows with rotating global anchors that reduces inference compute 2.98-3.33x on video diffusion models at extended horizons while remaining quality-neutral or positive.
-
PhyMotion: Structured 3D Motion Reward for Physics-Grounded Human Video Generation
PhyMotion scores generated human videos by grounding recovered 3D poses in a physics simulator across kinematic, contact, and dynamic axes, yielding stronger human correlation and larger RL post-training gains than prior 2D rewards.
-
LIVEditor-14B: Lightning Unified Video Editing via In-Context Sparse Attention
LIVEditor-14B applies a new sparse attention method (ISA) that prunes context and uses query-sharpness routing to cut attention latency ~60% with no loss in editing quality on standard benchmarks.
-
DynamicRad: Content-Adaptive Sparse Attention for Long Video Diffusion
DynamicRad achieves 1.7x-2.5x inference speedups in long video diffusion with over 80% sparsity by grounding adaptive selection in a radial locality prior, using dual-mode static/dynamic strategies and offline BO with a semantic motion router.
-
AdaCluster: Adaptive Query-Key Clustering for Sparse Attention in Video Generation
AdaCluster delivers a training-free adaptive query-key clustering framework for sparse attention in video DiTs, yielding 1.67-4.31x inference speedup with negligible quality loss on CogVideoX-2B, HunyuanVideo, and Wan-2.1.
-
Long-Horizon Streaming Video Generation via Hybrid Attention with Decoupled Distillation
Hybrid Forcing combines linear temporal attention for long-range retention, block-sparse attention for efficiency, and decoupled distillation to achieve real-time unbounded 832x480 streaming video generation at 29.5 FPS.
-
SURF: Signature-Retained Fast Video Generation
SURF accelerates high-resolution video generation up to 12.5x by using noise reshifting for low-res previews from pretrained models and a shifting-window Refiner for efficient upscaling that retains original signatures.
-
HunyuanVideo 1.5 Technical Report
HunyuanVideo 1.5 delivers state-of-the-art open-source text-to-video and image-to-video generation with an 8.3B parameter DiT model featuring SSTA attention, glyph-aware encoding, and progressive training.
-
Ride the Wave: Precision-Allocated Sparse Attention for Smooth Video Generation
PASA uses curvature-aware dynamic budgeting, grouped approximations, and stochastic attention routing to accelerate video diffusion transformers while eliminating temporal flickering from sparse patterns.
-
Towards Redundancy Reduction in Diffusion Models for Efficient Video Super-Resolution
OASIS reduces redundancy in diffusion models for real-world video super-resolution via attention specialization routing and progressive training, delivering state-of-the-art quality with 6.2x faster inference than prior one-step baselines.
-
OSP-Next: Efficient High-Quality Video Generation with Sparse Sequence Parallelism, HiF8 Quantization, and Reinforcement Learning
OSP-Next reports 83.73% VBench score and up to 2.27x speedup via hybrid sparse attention, SSP parallelism, HiF8 quantization, and Mix-GRPO on diffusion transformers.
- HEART: Exploiting Head Heterogeneity in Sparse Attention for Video Diffusion