In‑browser Zero‑Shot Classification

What does “zero‑shot classification” mean? Imagine you have a list of topics – “education”, “politics”, “business”. Without teaching the model those topics beforehand, you ask it: “Which of these topics best describes the sentence below?” The model looks at the sentence and picks the most likely label(s). That’s zero‑shot: it works on categories it has never seen during training.

Why run it in your browser? The model and its weights are downloaded once, then all calculations happen locally on your computer (using WebGPU or WebAssembly). No data leaves your device, and after the first download you can use it offline.

What this demo does

• Choose the compute backend – WebGPU (fastest if supported), WASM, or “auto”.
• See a progress bar while the model files download.
• Two ready‑to‑use pipelines:
  • Zero‑shot classification (the focus of this page)
  • Summarisation (you can experiment with it too)
• Works in modern desktop browsers (Chrome ≥ 112, Edge, Safari with WebGPU).

Troubleshooting: If the demo stays on “Model not loaded”, open the browser console (F12) to see any error messages, ensure you are loading the page from its HTTPS GitHub Pages URL, and disable any ad‑blocking extensions that might block model downloads.