lvsa-vllm-omni

Official

Enable LVSA sparse attention in vllm-omni.

AuthorJiusiServe
Version1.0.0
Installs0

System Documentation

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.

Dependency Matrix

Required Modules

None required

Components

Standard package

💻 Claude Code Installation

Recommended: Let Claude install automatically. Simply copy and paste the text below to Claude Code.

Please help me install this Skill:
Name: lvsa-vllm-omni
Download link: https://github.com/JiusiServe/LongVideoSparseAttention/archive/main.zip#lvsa-vllm-omni

Please download this .zip file, extract it, and install it in the .claude/skills/ directory.
View Source Repository

Agent Skills Search Helper

Install a tiny helper to your Agent, search and equip skill from 510,000+ vetted skills library on demand.