HV

hsliuustc0106/vllm-omni-skills

Deployment & DevOps
84 stars 품질 40 트렌드 40

A collection of AI assistant skills for vLLM-Omni -- a framework for efficient omni-modality model inference supporting text, image, video, and audio.

개요

A collection of AI assistant skills for vLLM-Omni -- a framework for efficient omni-modality model inference supporting text, image, video, and audio.

README

vllm-omni-skills

A collection of AI assistant skills for vLLM-Omni – a framework for efficient omni-modality model inference supporting text, image, video, and audio.

Skills Index

Skill Description
vllm-omni-setup Installation, environment configuration, GPU/driver prerequisites
vllm-omni-api OpenAI-compatible API client integration
vllm-omni-serving Launching API servers, model configuration, scaling
vllm-omni-hardware Hardware backends (CUDA, ROCm, NPU, XPU)
vllm-omni-image-gen Image generation and editing (FLUX, SD3, Qwen-Image, BAGEL, etc.)
vllm-omni-video-gen Video generation (Wan2.2 T2V/I2V/TI2V)
vllm-omni-audio-tts Audio generation and TTS (Qwen3-TTS, MiMo-Audio, Stable-Audio)
vllm-omni-tts-integration Adding new TTS models and production speech-serving integrations
vllm-omni-multimodal End-to-end omni-modality models (Qwen-Omni)
vllm-omni-distributed Distributed inference, disaggregation, Ray
vllm-omni-perf Performance tuning, benchmarking, TeaCache, CPU offloading
vllm-omni-quantization Quantization (AWQ, GPTQ, FP8), memory reduction, quality verification
vllm-omni-contrib Contributing new models and development workflow
vllm-omni-cicd CI/CD pipelines for model deployments
vllm-omni-review PR review guidelines, checklists, and common pitfalls
vllm-omni-release-note-writer Release note drafting guidance based on historical vLLM-Omni releases
vllm-omni-recipe Creating deployment guides for vLLM recipes repository

Installation

For Cursor IDE

Copy the skills/ directory into your project:

cp -r skills/ /path/to/your-project/.cursor/skills/

Or symlink for shared use:

ln -s /path/to/vllm-omni-skills/skills/ ~/.cursor/skills/vllm-omni/

For Claude Code

Add this repository as a plugin marketplace:

/plugin marketplace add hsliuustc0106/vllm-omni-skills

Install a specific skill plugin, for example the review skill:

/plugin install vllm-omni-review@vllm-omni-skills

Claude Code’s install command uses plugin@marketplace; after adding this repository, the marketplace name is vllm-omni-skills.

For Codex

cp -r skills/* ~/.codex/skills/

Usage

Once installed, skills activate automatically based on context. For example:

  • Ask “How do I install vllm-omni?” and the setup skill activates
  • Ask “Generate an image of a sunset” and the image-gen skill activates
  • Ask “Set up distributed inference across 4 GPUs” and the distributed skill activates
  • Ask “Review this PR for vllm-omni” and the review skill activates
  • Ask “Draft the vLLM-Omni v0.19.0rc1 release notes” and the release-note-writer skill activates

Each skill provides step-by-step workflows, code examples, and references to detailed documentation.

Validation

Run the validation script to check all skills for structural correctness:

python scripts/validate_all.py

Validate a single skill:

python scripts/validate_all.py skills/vllm-omni-setup/

Project Structure

vllm-omni-skills/
├── README.md
├── LICENSE
├── .claude-plugin/
│   └── marketplace.json   # Claude Code marketplace manifest
├── docs/
│   ├── PRD.md              # Product requirements
│   ├── ARCHITECTURE.md     # Architecture design
│   └── TEST_DESIGN.md      # Test design
├── plugins/
│   └── vllm-omni-*/        # Claude Code plugin wrappers
│       ├── .claude-plugin/
│       │   └── plugin.json # Per-plugin manifest
│       └── skills/
│           └── vllm-omni-* # Symlink to canonical skill content
├── scripts/
│   └── validate_all.py     # Skill validation tool
└── skills/
    └── vllm-omni-*/        # 16 skill directories
        ├── SKILL.md         # Main skill instructions
        ├── references/      # Detailed reference docs
        └── scripts/         # Utility scripts (some skills)

Version Variables

Skills use shell-style variables for version-dependent values. Set these before following any skill instructions:

export VLLM_VERSION="0.16.0"           # vLLM pip package version
export VLLM_OMNI_VERSION="v0.16.0"     # vLLM-Omni release / Docker tag
export PYTHON_VERSION="3.12"           # Python version

Check the vllm-omni quickstart for currently recommended versions.

License

Apache License 2.0

View this README on GitHub

추천 도구

다른 키워드를 입력하거나 필터를 제거해 보세요.

설치

npx skillfish add hsliuustc0106/vllm-omni-skills