AI Dictionary › AI Fundamentals
Top-p, or nucleus sampling, is a sampling technique that limits the model's choices to the smallest group of tokens whose cumulative probability reaches a threshold p, for example 0.9. Picture candidates ranked from most to least likely: you keep them until their probabilities sum to 90% of the total, then discard the rest and draw only within this "nucleus". The count is not fixed: if the model is confident the nucleus is small, if uncertain it widens. It is like inviting to a meeting only the people who together hold 90% of the expertise.
Top-p matters because it cuts improbable tails that produce nonsense while keeping flexibility. It is often preferred over top-k because it adapts to the model's confidence step by step.
From our network
AGORÀ Intelligence: Enterprise AI Governance Platform
Govern AI at scale: policies, adoption and measurable results on your data. Built for boards and C-suite.
Visit agora-intelligence.com →From the Agora Intelligence blog
Say what you mean. Get what you need.
Grace Certified, the AI coach that trains and certifies your prompt engineering, by Agora Intelligence.