Hello, I’m currently training the revised version of baseline in the vision task.
I have a question here. My model is converging faster than the previous model, but eventually the metric value is caught up. Does convergence have nothing to do with final results? Or can I get a better value by changing the scheduler?