What that means in practice is that  every output is different, as it's not predicting the entire text, but it's simply going token  by token.

这在实践中意味着每个输出都是不同的，因为它不会预测整个文本，而只是逐个标记。