Glossary

Top-P Sampling (Nucleus Sampling)

Top-P Sampling, also known as Nucleus Sampling, is a probabilistic token selection method in language models that dynamically restricts sampling to the smallest set of tokens whose cumulative probability mass meets or exceeds a threshold P (typically 0.8-0.95), adaptively balancing output coherence and diversity unlike fixed-size approaches.

Top-P Sampling, also known as Nucleus Sampling, is a probabilistic token selection method in language models that dynamically restricts sampling to the smallest set of tokens whose cumulative probability mass meets or exceeds a threshold P (typically 0.8-0.95), adaptively balancing output coherence and diversity unlike fixed-size approaches.