The step_hint is meaningless without knowing the batch size the baseline used. We should just provide that information for each workload so submissions that use batch sizes that non-uniformly vary relative to the baseline can properly adjust the step hints, if they desire.