Hacker News new | ask | show | jobs
by mzl 354 days ago
That depends on the sampling strategy. Greedy sampling takes the max token at each step.