vllm-docs

Locate official vLLM documentation for configuring and debugging LLM inference.

7|2|Updated Feb 25, 2026
One-click install
npx skills add https://github.com/wenerme/ai --skill vllm-docs
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: vllm-docs
Source: https://github.com/wenerme/ai/tree/main/skills/vllm-docs
Command: npx skills add https://github.com/wenerme/ai --skill vllm-docs

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

vLLM users need fast, accurate guidance for configuring high-performance LLM inference and resolving tricky serving issues across models, quantization methods, and distributed setups.

Core Features & Use Cases

  • OpenAI-compatible serving and deployment guidance: Covers how to run vLLM as an OpenAI-compatible server and how to deploy it with Docker, Kubernetes, and reverse proxies.
  • Distributed and parallelism troubleshooting: Explains data/pipeline/tensor/expert/context parallel serving concepts and provides dedicated troubleshooting guidance for distributed environments.
  • Advanced performance topics: Documents key vLLM capabilities such as quantization (AWQ, GPTQ, FP8, GGUF, INT4/8), speculative decoding, LoRA adapters, structured outputs, multimodal inputs, and memory optimization mechanisms like PagedAttention.

Quick Start

Use the vllm-docs skill to locate the exact documentation page for the feature you need and follow it to configure your deployment for OpenAI-compatible serving.

Frequently Asked Questions about vllm-docs

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

FAQPage Schema
How do I deploy vLLM as an OpenAI-compatible API server?

vLLM supports OpenAI-compatible serving by configuring specific serve arguments and CLI options. You can deploy it using Docker, Kubernetes, or reverse proxies by following the navigable documentation index for exact setup steps.

What is the best way to configure distributed tensor and pipeline parallelism in vLLM?

vLLM distributed deployment utilizes tensor, pipeline, expert, and context parallelism. You can configure these serving modes and troubleshoot distributed environments using the provided official documentation references and engine arguments.

Does vLLM support quantization methods like AWQ, GPTQ, and FP8?

vLLM supports AWQ, GPTQ, FP8, GGUF, and INT4/8 quantization methods. The documentation details how to apply these formats to optimize memory and maintain high-throughput LLM inference performance.

Can I use multimodal inputs and LoRA adapters with vLLM inference?

vLLM supports multimodal inputs and LoRA adapters for advanced inference. The documentation provides reference links detailing how to implement these features alongside structured outputs and speculative decoding.

Why is my vLLM inference serving deployment not working in a distributed setup?

Distributed vLLM serving failures often stem from misconfigured parallelism arguments. Use the dedicated troubleshooting documentation to verify engine arguments, environment variables, and distributed deployment patterns.