In the paper, they usually share every 500 steps, a few million steps is an entire training process, more often than not models are trained less than a million

In the paper, they usually share every 500 steps, a few million steps is an entire training process, more often than not models are trained less than a million steps.