The problem of step (horizontal axis) during experimental data recording

[DIGR-cyber]

In wandb, the step (abscissa x) in all charts is three times of the effective_iter. The "global_step=self. effective_iter" set in this project is the same as the iter setting in logging.info(f"iter {self.effective_iter:5d} (epoch {epoch:2d}): loss={accumulated_loss:.5f}"), but it is found that almost all icons are three times worse in later verification.
In my experiment, I only used vkitti in the training data set, and I only changed this: effective_batch_size:32 , max_train_batch_size: 16.
If you see it, please help me answer it. Thank you very much!