The selected token is appended to the context and the process repeats until a stop condition. It is deterministic under stable computation but can miss a better overall sequence because it never revisits earlier choices.
Greedy decoding chooses the highest-probability token at every generation step.
The selected token is appended to the context and the process repeats until a stop condition. It is deterministic under stable computation but can miss a better overall sequence because it never revisits earlier choices.