Hugging Face Daily Papers · · 4 min read

FVAttn: Adaptive Sparse Attention with Runtime Load Balancing for Video Generation

Mirrored from Hugging Face Daily Papers for archival readability. Support the source by reading on the original site.

Video Diffusion Transformers process long spatio-temporal sequences, making self-attention the main bottleneck in high-resolution video generation. Training-free sparse attention reduces this cost, but adaptive Top-p routing creates uneven per-head workloads under multi-GPU sequence parallelism. The resulting workload heterogeneity turns sparse attention into a rank-level straggler problem. We present FVattn, a training-free sparse-attention system that improves the distributed execution efficiency of adaptive sparse attention under multi-GPU sequence parallelism. FVattn uses Top-p routing, a Top-k safety floor, and video-aware block organization as the sparse-routing frontend, then repairs the materialized mask at runtime. Runtime Load Balancing migrates a small number of heavy heads via P2P communication to shorten the current critical path. Slack-Aware Sparse Augmentation fills residual non-critical-rank slack with additional high-value blocks, while overlap hides scheduling and migration overhead behind existing computation. On step-distilled Wan2.2 I2V, FVattn reduces average load imbalance from 1.34 to 1.08 and delivers a 4.41× attention speedup over FlashAttention, while achieving a 2.02--2.11× DiT inference speedup with competitive video quality.</p>\n","updatedAt":"2026-07-23T05:05:57.508Z","author":{"_id":"665c91e15b11dca02f0c5891","avatarUrl":"/avatars/49a4ee76c3edfe5b0916051a5ac4acfd.svg","fullname":"Ye Huang","name":"henry-y1","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":6,"isUserFollowing":false,"primaryOrg":{"avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/665c91e15b11dca02f0c5891/ek9KGIc02tiFfaYeDfLaU.png","fullname":"DAGroup-PKU","name":"DAGroup-PKU","type":"org","isHf":false}}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8018568158149719},"editors":["henry-y1"],"editorAvatarUrls":["/avatars/49a4ee76c3edfe5b0916051a5ac4acfd.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2607.16190","authors":[{"_id":"6a5d923f6a69ce099f4d6e0e","name":"Hao Liu","hidden":false},{"_id":"6a5d923f6a69ce099f4d6e0f","name":"Chenghuan Huang","hidden":false},{"_id":"6a5d923f6a69ce099f4d6e10","name":"Ye Huang","hidden":false},{"_id":"6a5d923f6a69ce099f4d6e11","name":"Zhiying Wen","hidden":false},{"_id":"6a5d923f6a69ce099f4d6e12","name":"Hao Liu","hidden":false},{"_id":"6a5d923f6a69ce099f4d6e13","name":"Mohan Zhang","hidden":false},{"_id":"6a5d923f6a69ce099f4d6e14","name":"Chen Li","hidden":false},{"_id":"6a5d923f6a69ce099f4d6e15","name":"Ziyang Ma","hidden":false},{"_id":"6a5d923f6a69ce099f4d6e16","name":"Jing Lyu","hidden":false},{"_id":"6a5d923f6a69ce099f4d6e17","name":"Jiangsu Du","hidden":false}],"publishedAt":"2026-07-17T00:00:00.000Z","submittedOnDailyAt":"2026-07-23T00:00:00.000Z","title":"FVAttn: Adaptive Sparse Attention with Runtime Load Balancing for Video Generation","submittedOnDailyBy":{"_id":"665c91e15b11dca02f0c5891","avatarUrl":"/avatars/49a4ee76c3edfe5b0916051a5ac4acfd.svg","isPro":false,"fullname":"Ye Huang","user":"henry-y1","type":"user","name":"henry-y1"},"summary":"Video Diffusion Transformers process long spatio-temporal sequences, making self-attention the main bottleneck in high-resolution video generation. Training-free sparse attention reduces this cost, but adaptive Top-p routing creates uneven per-head workloads under multi-GPU sequence parallelism. The resulting workload heterogeneity turns sparse attention into a rank-level straggler problem. We present , a training-free sparse-attention system that improves the distributed execution efficiency of adaptive sparse attention under multi-GPU sequence parallelism. uses Top-p routing, a Top-k safety floor, and video-aware block organization as the sparse-routing frontend, then repairs the materialized mask at runtime. Runtime Load Balancing migrates a small number of heavy heads via P2P communication to shorten the current critical path. Slack-Aware Sparse Augmentation fills residual non-critical-rank slack with additional high-value blocks, while overlap hides scheduling and migration overhead behind existing computation. On step-distilled Wan2.2 I2V, reduces average load imbalance from 1.34 to 1.08 and delivers a 4.41times attention speedup over FlashAttention, while achieving a 2.02--2.11times DiT inference speedup with competitive video quality.","upvotes":4,"discussionId":"6a5d923f6a69ce099f4d6e18"},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"665c91e15b11dca02f0c5891","avatarUrl":"/avatars/49a4ee76c3edfe5b0916051a5ac4acfd.svg","isPro":false,"fullname":"Ye Huang","user":"henry-y1","type":"user"},{"_id":"63ac5701c21e60a3e9b58aa7","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/63ac5701c21e60a3e9b58aa7/g6EX7diOpuA94R2ab-rZC.png","isPro":true,"fullname":"Dipankar Sarkar","user":"dipankarsarkar","type":"user"},{"_id":"6915acbb907428dbc99519fe","avatarUrl":"/avatars/860a461ffa45b8ec8c969fb2428f9fee.svg","isPro":false,"fullname":"aaa","user":"yhrbb02","type":"user"},{"_id":"680350a7cd4782848c1ae70c","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/ysmIY9FK3q58rPMwJMdiU.jpeg","isPro":false,"fullname":"GarvinRay","user":"GarvinRay","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2607/2607.16190.md","query":{}}">
Papers
arxiv:2607.16190

FVAttn: Adaptive Sparse Attention with Runtime Load Balancing for Video Generation

Published on Jul 17
· Submitted by
Ye Huang
on Jul 23
Authors:
,

Abstract

Video Diffusion Transformers process long spatio-temporal sequences, making self-attention the main bottleneck in high-resolution video generation. Training-free sparse attention reduces this cost, but adaptive Top-p routing creates uneven per-head workloads under multi-GPU sequence parallelism. The resulting workload heterogeneity turns sparse attention into a rank-level straggler problem. We present , a training-free sparse-attention system that improves the distributed execution efficiency of adaptive sparse attention under multi-GPU sequence parallelism. uses Top-p routing, a Top-k safety floor, and video-aware block organization as the sparse-routing frontend, then repairs the materialized mask at runtime. Runtime Load Balancing migrates a small number of heavy heads via P2P communication to shorten the current critical path. Slack-Aware Sparse Augmentation fills residual non-critical-rank slack with additional high-value blocks, while overlap hides scheduling and migration overhead behind existing computation. On step-distilled Wan2.2 I2V, reduces average load imbalance from 1.34 to 1.08 and delivers a 4.41times attention speedup over FlashAttention, while achieving a 2.02--2.11times DiT inference speedup with competitive video quality.

Community

Paper submitter about 9 hours ago

Video Diffusion Transformers process long spatio-temporal sequences, making self-attention the main bottleneck in high-resolution video generation. Training-free sparse attention reduces this cost, but adaptive Top-p routing creates uneven per-head workloads under multi-GPU sequence parallelism. The resulting workload heterogeneity turns sparse attention into a rank-level straggler problem. We present FVattn, a training-free sparse-attention system that improves the distributed execution efficiency of adaptive sparse attention under multi-GPU sequence parallelism. FVattn uses Top-p routing, a Top-k safety floor, and video-aware block organization as the sparse-routing frontend, then repairs the materialized mask at runtime. Runtime Load Balancing migrates a small number of heavy heads via P2P communication to shorten the current critical path. Slack-Aware Sparse Augmentation fills residual non-critical-rank slack with additional high-value blocks, while overlap hides scheduling and migration overhead behind existing computation. On step-distilled Wan2.2 I2V, FVattn reduces average load imbalance from 1.34 to 1.08 and delivers a 4.41× attention speedup over FlashAttention, while achieving a 2.02--2.11× DiT inference speedup with competitive video quality.

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.
Tap or paste here to upload images

· Sign up or log in to comment

Get this paper in your agent:

hf papers read 2607.16190
Don't have the latest CLI?
curl -LsSf https://hf.co/cli/install.sh | bash

Models citing this paper

No model linking this paper

Cite arxiv.org/abs/2607.16190 in a model README.md to link it from this page.

Datasets citing this paper

No dataset linking this paper

Cite arxiv.org/abs/2607.16190 in a dataset README.md to link it from this page.

Spaces citing this paper

No Space linking this paper

Cite arxiv.org/abs/2607.16190 in a Space README.md to link it from this page.

Collections including this paper

No Collection including this paper

Add this paper to a collection to link it from this page.

Discussion (0)

Sign in to join the discussion. Free account, 30 seconds — email code or GitHub.

Sign in →

No comments yet. Sign in and be the first to say something.

More from Hugging Face Daily Papers