About this site

LLM Atlas is an independent, interactive front end for Sebastian Raschka’s LLM Architecture Gallery. The gallery is the source of truth: every architecture diagram, model name, parameter count, attention description, KV-cache figure, licence and link on this site is taken from it, and each model page links back to its original card.

What this site adds is the browsing experience: instant search and filters, a side-by-side diff tool, concept pages that show who adopted which idea and when, a timeline, aggregate charts, and machine-readable versions of everything for AI agents.

The architecture diagrams are loaded directly from the original gallery and remain the work of their author. Concept summaries here are short paraphrases; each one links to the full explainer with figures. If you spot an error in the data, please report it upstream so the source gets fixed for everyone.

How the data stays fresh

A sync script parses the public gallery page into structured JSON (118 models at the last sync on 2026-11-04). The site is then rebuilt as static HTML, so it is fast, cacheable and readable without JavaScript.

Built with

• Next.js (App Router, static export) and React view transitions
• shadcn/ui on Base UI primitives, Tailwind CSS v4
• Motion for springs, layout animation and scroll effects
• Skullet for measured, shimmering skeletons
• EvilCharts (Recharts) for the insights charts