The softmax turns $n$ numbers in $R^N$ into a probability distribution proportional to the size of the numbers.

Given an $n$-dimensional vector $v$ with all component terms in $R^N$, the softmax of $v$ is: $$softmax(v{)}_i=\frac{\exp \left(v_i\right)}{{\sum }_{j=1}^n\exp \left(v_i\right)}$$