Top-K / Top-P Sampling

Two ways to limit AI to only "high probability" tokens during generation — combat low-quality outputs.

Definition

Top-K limits sampling to the K most likely tokens. Top-P (nucleus sampling) takes the smallest set summing to probability P. Often combined with temperature.

Examples

  • Top-P=0.9 keeps the smallest set covering 90% probability

Related terms

Looking for actual gigs that use Top-K / Top-P Sampling? Browse all live listings or filter by $100+/hr.