---
title: "vLLM vs SGLang Architecture"
description: "Read SGLang and vLLM as programs: their processes, where the scheduling decision lives, and what each calls its cached state. Designs, not benchmarks."
canonical_url: "https://fanout.sh/inference-eng/curriculum/17-02"
md_url: "https://fanout.sh/inference-eng/curriculum/17-02.md"
access: "pro-preview"
---

# vLLM vs SGLang Architecture

Read SGLang and vLLM as programs: their processes, where the scheduling decision lives, and what each calls its cached state. Designs, not benchmarks.

- Access: Fanout Pro (public preview)

## Public preview

The fair comparison lesson ended owing this lesson a job. Its method is complete, and what it still needs is vocabulary: naming what differs between two engines is a separate skill from stating conditions. This lesson opens the first pair, the two open-source servers the module has already pinned, and reads them as programs. Nothing here is timed, ranked, or…

## In this lesson

- The tag and the commit
- Two processes, or three
- Where the scheduling decision lives
- What the cached state is called
- Two questions, two starting points

## Continue learning

Pro unlocks the complete lesson, video, practice, and code where available.

---
This representation contains public Fanout content only. Protected Pro lessons, account data, billing, checkout, and pricing are not included.

Browse the public content map: https://fanout.sh/sitemap.md