Inside AI
Follow a prompt through GPT-2. Inspect tokens, attention, and next-token probabilities, then try live inference in your browser.
Open Inside AI to follow a prompt through a transformer. Start with a prepared example, inspect its token and attention values, and change the sampling controls. Live mode downloads the model only when you choose to load it. Your prompts stay in your browser.
The guided explanation remains readable without JavaScript. Use it alongside the diagram, or begin there before loading the model.
Inside AI adapts Transformer Explainer by the Polo Club of Data Science under its MIT license. The adaptation is based on upstream commit bfe50afba10b9b560b84143ee1107d977defa74f.
Details
- Svelte
- D3
- WebAssembly
- Jekyll
Features
- Guided explanations of tokens, embeddings, attention, and sampling
- Prepared model examples with inspectable values
- Optional GPT-2 inference in your browser
- Keyboard controls, numerical tables, and light and dark themes