Models, reasoning & training · Concept

Logits

Unnormalized model scores over candidate outputs, often transformed into probabilities.