Concept
aka nucleus sampling

Top-P Sampling

Sampling from the smallest token set whose cumulative probability exceeds threshold p.

Definition

Top-p (nucleus) sampling dynamically truncates the distribution: it includes only the most likely tokens that together account for probability mass p. Unlike fixed top-k, it adapts to the distribution's shape — fewer tokens when the model is confident, more when it's uncertain.

Common use cases

  • Creative writing
  • Open-ended chat
  • Brainstorming

Related terms

    Top-P Sampling — AI Glossary | Railwail