EasyDeepLearn

How does top-k differ from top-p sampling?

easy

Answer

  • Top-k: sample from the k highest-probability tokens each step, ignoring the rest.
  • Fixed cardinality regardless of the distribution's shape.
  • Top-p (nucleus): sample from the smallest set of tokens whose cumulative probability    p\mathrm{probability}\; \ge \;p.
  • Adapts to the distribution — takes fewer tokens when the model is confident, more when it's uncertain.
  • Top-p is usually preferred as it maintains diversity where useful and precision where confident.
  • Combining top-k=50 + top-p=0.95 is a common safety net.
Check yourself — multiple choice
  • Top-k adapts to distribution shape
  • Top-k: fixed k tokens; top-p: smallest set with cumulative prob  p\mathrm{prob}\; \ge p — adapts to shape (preferred)
  • Same behavior
  • Only top-k works

Top-k fixed size; top-p adaptive to distribution — top-p usually preferred.

#generation#decoding#sampling

Practise LLMs & GenAI

214 interview questions in this topic.

Related questions