← All models · DeBERTa zero-shot
use case · basics
The core trick: you don't pick from a fixed list of categories — you type your own, and the model scores how well each one fits the text. Change the labels and it re-classifies instantly, with no retraining. Type some text and a few labels below.
In single-label mode the scores compete (they're softmaxed across your labels, so they sum to 100%). The model has never seen your labels before — it's reasoning about whether the text implies each one.