← All models · DeBERTa zero-shot
use case · wild
Real text is rarely one thing. In multi-label mode each label is scored on its own — no softmax forcing them to compete — so a single message can be tagged excited and nervous and about a launch all at once. Set a threshold and everything above it sticks. Great for mood boards, content tagging, and rich analytics.
Switching between single-label and multi-label isn't cosmetic: single-label softmaxes the scores so they compete for one winner; multi-label runs each hypothesis independently, so overlapping truths can all score high. Pick the mode that matches your problem.
← Practical · DeBERTa zero-shot · Next: Multi-model →