LLMs don’t pick the next word — they draw from a probability distribution. Adjust temperature and sampling parameters to reshape it, then click Sample to see what comes out.
Pre-baked distributions from realistic model outputs — for educational illustration.
Parameters
Temperature1.00
Low → peaked (deterministic). High → flat (creative). 1.0 = no change.
Top-koff
Keep only the k most probable tokens and zero out the rest.
Top-p (nucleus)off
Keep the smallest set of tokens whose probabilities sum to ≥ p.