How many tokens the model generates per turn. Default = model max context (262144). Larger = slower (one full forward per token).

0 = greedy. 2 = high-variance sampling.

How many top tokens to show per grid cell / detail panel.