What are you trying to imply? That training Transformer models necessarily needs to be a continuous process? You know it's pretty easy to stop and continue training, right?

I don't know why people keep commenting in spaces they've never worked in.