Short answer is when you don't personally see it, with your eyes , it getting any better.
The graph is just measuring the loss so you have an idea of how quickly it's learning.
But even when it looks near flat it's still learning and trying different combinations trying to get to "the best" .
After a while it'll just jump around between various combinations, but all of them are all valid and as good as they're going to be, may look a little different.
That really is the point where you have to be creative and make that determination yourself.
I dunno what I'm doing
2X RTX 3090 : RTX 3080 : RTX: 2060 : 2x RTX 2080 Super : Ghetto 1060