The default decoding strategy is greedy search, which is the simplest decoding strategy that picks a token with the highest probability as the next token. For many tasks
and small output sizes this works well.