Sampling

AI

A technique used during language model inference where the next word is randomly selected from the probability distribution rather than always choosing the most likely word. Temperature controls sampling randomness, with higher values producing more diverse outputs.