Crimson AI NewsA CrimsonLingua Network service
EN ع
← Back to news
Research paper Hugging Face

Sol-Attn: Training-Free Sparse Attention Speeds Up Video Generation by 2x+

AI By Crimson AI Hugging Face Papers 28 July 2026 · 00:00 18 views
Share: X Telegram

Sol-Attn introduces a training-free dynamic sparse attention method that unifies routing, computation, and approximation in a single online-softmax pass, achieving up to 2.4× end-to-end speedup for video generation while preserving quality.

Sol-Attn: Training-Free Sparse Attention Speeds Up Video Generation by 2x+

Key points

Researchers from NVIDIA and collaborating institutions have introduced Sol-Attn, a training-free sparse attention mechanism that accelerates video generation inference by dynamically sparsifying attention computations. The method addresses a key bottleneck in diffusion transformers, where long token sequences make attention the dominant cost.

Existing training-free sparse attention methods suffer from rigid routing and lossy sparsification. Sol-Attn overcomes these limitations by unifying dynamic routing, sparse computation, and approximation correction in a single online-softmax pass. Its core innovation is on-the-fly block thresholding with proxy-score reuse: it selects critical key-value blocks by comparing proxy scores against a threshold during online softmax, enabling dynamic yet controllable budgets without materializing the proxy map. Unselected blocks are not discarded entirely; their proxy scores are reused to approximate their contribution.

Experiments on image and video generation tasks show that Sol-Attn advances the quality-efficiency frontier. Against dense FlashAttention-3, it achieves 2.02× end-to-end speedup on Wan 2.1-14B, 2.12× on HunyuanVideo-13B, and up to 2.4× on LTX 2.3. When integrated into Sol-Engine (with kernel fusion and caching), speedups reach 3.48× on Wan 2.1-14B and 5.08× on HunyuanVideo-13B.

The method is already available in Sol-Engine. The paper and code are publicly accessible, with the B200 kernel still under further optimization.

ModelSpeedup vs Dense FlashAttention-3Speedup with Sol-Engine
Wan 2.1-14B2.02×3.48×
HunyuanVideo-13B2.12×5.08×
LTX 2.3up to 2.4×
Source
Hugging Face · Hugging Face Papers
Related news
Research paper
Hugging Face 1 Sep 2026

Hugging Face Unveils CogEvol: Single-Pass Learning Environment Generation

CogEvol, a new family of models from Hugging Face, generates structured learning artifacts like slides and interactive HTML pages...

1
Research paper
Hugging Face 1 Sep 2026

PaperGym: Turning Papers into Training Environments for AI Research Planning

Hugging Face researchers introduce PaperGym, a framework that converts scientific papers into reinforcement learning environments...

1
Research paper
Hugging Face 1 Sep 2026

Study: On-Policy Distillation Works by Suppressing Low-Probability Tokens, Not Teacher Guidance

A new Hugging Face paper reveals that on-policy distillation (OPD) improves reasoning mainly by suppressing low-probability tokens...

1