InferMatrix
Official@jiusiserve
Offers specialized configuration and tuning for LVSA sparse attention mechanisms within high-performance video diffusion generation pipelines.
Agent Skills by InferMatrix
Showing 6 vetted skills indexed across 1 GitHub repositories.
JiusiServelvsa-add-model
Implement ModelAdapters to add LVSA support for video diffusion models.
JiusiServelvsa-troubleshooting
Diagnose LVSA failure modes and guide remediation across vLLM-Omni, Wan, HunyuanVideo, and CogVideoX pipelines.
JiusiServelvsa-vllm-omni
Configure and run the LVSA vllm-omni plugin for sparse attention in vllm-omni.
JiusiServelvsa-quickstart
Install LVSA, select a backend, and generate a first long video.
JiusiServelvsa-tuning
Tune LVSA sparsity_scale and window parameters for Wan and HunyuanVideo models.
JiusiServelvsa-reproduce-paper
Automates LVSA paper result reproduction using bundled benchmarks and scripts.
Frequently Asked Questions About InferMatrix
FAQPage SchemaWhat specific video generation tasks does InferMatrix support?▼
InferMatrix enables the implementation of LVSA sparse attention for video diffusion models. It supports configuring vLLM-Omni plugins, tuning sparsity parameters for Wan and HunyuanVideo architectures, and reproducing research results using bundled benchmarks for long-form video generation.
Which technical personas benefit from these LVSA capabilities?▼
These capabilities are designed for machine learning engineers and researchers focused on video diffusion model optimization. Professionals working with vLLM-Omni, Wan, HunyuanVideo, or CogVideoX pipelines will find these resources essential for managing sparse attention and model inference performance.
What are the prerequisites for deploying LVSA configurations?▼
Deployment requires an existing environment running vLLM-Omni, Wan, HunyuanVideo, or CogVideoX. Users must select a compatible backend and configure the LVSA plugin parameters, specifically the sparsity_scale and window settings, to align with their target video diffusion model requirements.