Nat
The unit of surprise or cross-entropy when natural logs are used; logs base 2 give bits instead.
Among these six, softmax gives mat about 0.011 (0.01091 to four figures), so the loss is −ln 0.01091 = 4.518 nats.
See it in action
Try nat hands-on in the lesson Softmax and cross-entropy, part of the The maths behind AI course.