A decoder-only model is continuing the text The sky is. For this problem its whole vocabulary is four tokens, and the scores it produces for them are
| token | score |
|---|---|
blue | |
clear | |
dark | |
green |
Sampling with temperature divides every score by before the softmax turns the scores into probabilities:
The model itself is untouched — the four scores above are the same at every temperature. Only moves.
With , what probability does the model assign to blue?
Round your answer to 3 decimal places, and carry at least four decimals through the intermediate steps.