vLLM logo
vLLM is an open-source library for fast LLM inference and serving, built around PagedAttention and continuous batching, started at UC Berkeley.
Also known as vllm-project, vLLM inference engine
Looks better on a light background
npm i @dev.icons/reactimport { VllmIcon, Vllm } from "@dev.icons/react";<VllmIcon size={32} />
<Vllm size={32} />
Using an AI agent? Ask the Devicons MCP server for vllm, or fetch its JSON from the Icons API.
About the vLLM logo
The vLLM mark is a stylized lowercase “v” made of two flat shapes: a yellow #FDB515 triangle on the left and a taller, slanted blue #30A2FF stroke on the right. In the full logo, the mark sits before “LLM” in a thin, geometric sans-serif set in dark gray #434343. The project’s website publishes the full logo in light and dark-mode versions.
The two-color mark works on light and dark backgrounds. The gray wordmark has low contrast on dark backgrounds, where the site uses a near-white #F3F3F3 version. On dark backgrounds, use the monochrome version.
When to use this logo
Use vLLM in LLM serving and inference architecture diagrams, self-hosting and GPU deployment guides, throughput benchmark tables, and docs for OpenAI-compatible inference endpoints. It also fits model card “serve with” sections. It pairs naturally with Hugging Face, PyTorch, NVIDIA, Ray, Kubernetes and inference alternatives such as Ollama, llama.cpp and SGLang.
