Stochastic Decoding Strategy

A strategy in which a large language model randomly samples the next token from a probability distribution during text generation, such as temperature sampling and top-p sampling. It can produce diverse output trajectories.