Greedy Decoding

Greedy decoding chooses the highest-probability token at each step.

If the next-token distribution is:

tea: 0.45
code: 0.30
music: 0.15
books: 0.10

greedy decoding chooses tea.

Greedy decoding is simple and deterministic, but it can be dull or brittle. A locally best token at one step may not lead to the best full sequence.

Exercise

In the distribution above, which probability is selected by greedy decoding, multiplied by 100?