Distribution of eval steps might be wrong?

[sadaisystems]

Hello guys, i have a question about usage of evals_per_epoch parameter.

Seems like when you train with this parameter, there's no evaluation stage after last training step. I got eval at the beginning of the run and a couple of intermediate evaluations.

Is this an intended behavior pattern? Or is it a bug?

Intuitively it should distribute evaluations in a such a way so the latest evaluating happens after last training step.