Detect people, organizations, and locations in text. Uses BERT fine-tuned on CoNLL-2003 — groups B-/I- tokens into labeled spans. Model runs in a Web Worker; first run downloads ~430 MB.
const model = await pool.load('token-classification', {
model: 'Xenova/bert-base-NER',
});
const entities = await model.run(text);
// [{ entity: 'B-PER', word: 'Steve', score: 0.99 }, …] — B/I tokens grouped by library