Technical
Top-P Sampling
Definition
A text generation technique that samples from the smallest set of tokens whose cumulative probability exceeds threshold P.In-Depth Explanation
Also called nucleus sampling, top-p dynamically adjusts the candidate token pool based on probability distribution. With top-p=0.9, the model considers tokens covering 90% of the probability mass. This often produces more natural text than temperature alone by excluding unlikely tokens.
Real-World Example
Top-p of 0.95 excludes very unlikely tokens while maintaining diversity in generation.
4 views0 found helpful