skills.rest
JiusiServe
JiusiServeOFFICIAL
Explore all skills by this author →
GitHub

lvsa-vllm-omni

Configure and run the LVSA vllm-omni plugin for sparse attention in vllm-omni.

18|4|Updated May 21, 2026
Search Tags:#plugin#model-serving#vllm-omni#wan#sparse-attention#lvsa#hunyuanvideo
One-click install
npx skills add https://github.com/JiusiServe/LongVideoSparseAttention --skill lvsa-vllm-omni
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: lvsa-vllm-omni
Source: https://github.com/JiusiServe/LongVideoSparseAttention/tree/main/skills/lvsa-vllm-omni
Command: npx skills add https://github.com/JiusiServe/LongVideoSparseAttention --skill lvsa-vllm-omni

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

LVSA-vllm-omni provides a plug-in path to enable sparse attention within the vllm-omni serving framework. It eliminates the need to modify core serving code while offering per-model configuration for horizons, hooks, and geometry.

Core Features & Use Cases

  • Zero-change plugin integration for vllm-omni
  • Per-model environment-variable control (LVSA_WAN_HOOK, LVSA_REFERENCE_LATENT_FRAMES)
  • Supports Wan, HunyuanVideo, Cosmos offline, and adapter-based configurations
  • Troubleshooting and plug-in architecture for LVSA in model serving

Quick Start

Install the LVSA core and lvsa-vllm-omni plugin, then enable LVSA for your chosen model in vllm-omni using the per-model environment variables.

Frequently Asked Questions about lvsa-vllm-omni

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I enable sparse attention in vllm-omni without modifying the core serving code?▼

You can enable sparse attention in vllm-omni by installing the LVSA core and the lvsa-vllm-omni plugin, which provides zero-change integration using per-model environment variables to configure horizons, hooks, and geometry.

How does the LVSA vllm-omni plugin handle configuration for different video models?▼

The LVSA vllm-omni plugin uses per-model environment variables like LVSA_WAN_HOOK and LVSA_REFERENCE_LATENT_FRAMES to control horizons, hooks, geometry, and debugging for specific models.

Does the vllm-omni sparse attention plugin support HunyuanVideo and Wan models?▼

Yes, the vllm-omni sparse attention plugin supports Wan2.x and HunyuanVideo-1.5 models, along with Cosmos 3.0 offline paths and adapter-based configurations.

What is the best way to configure LVSA sparse attention for Wan2.x in a model serving framework?▼

The best way to configure LVSA sparse attention for Wan2.x is by using the lvsa-vllm-omni plugin, which registers the backend and applies model-specific environment variables without requiring core code modifications.

Why do I need per-model environment variables to run sparse attention in vllm-omni?▼

Per-model environment variables are required to control specific horizons, hooks, and geometry settings for different models like Wan and HunyuanVideo, ensuring correct sparse attention integration and debugging within vllm-omni.

Related Skills

Explore More Software Engineering Skills→
hsliuustc0106hsliuustc0106
84

vllm-omni-setup

Install and configure vLLM-Omni with Python environments and smoke tests.

Community
Advanced
zechenzhangAGIzechenzhangAGI
11.5k

speculative-decoding

Reduce LLM inference latency using draft models, Medusa heads, and Lookahead Decoding.

Community
Intermediate
air-gappedair-gapped
5

vllm-omni

Operate disaggregated multimodal vLLM-Omni deployments across image, video, and audio endpoints.

Official
Advanced
JiusiServeJiusiServe
18

lvsa-troubleshooting

Diagnose LVSA failure modes and guide remediation across vLLM-Omni, Wan, HunyuanVideo, and CogVideoX pipelines.

Official
Advanced