Top-p Sampling
Sampling only from the smallest set of tokens whose probabilities reach p.
Resumen
Top-p, or nucleus sampling, keeps the most likely tokens until their cumulative probability reaches a threshold, then samples from that set. It solves the problem that a fixed top-k includes nonsense when the distribution is peaked and excludes reasonable options when it is flat. It exists because the set of plausible continuations should be sized by the model's own confidence.
Grafo de tecnologías
Cómo se conecta Top-p Sampling con el resto del stack.
Alternativas
-
Beam Search
technique
Una alternativa a Top-p Sampling · source
Comparativas
Comparativas detalladas con las alternativas anteriores.