Closed luigifaticoso closed 3 years ago
Yes, it can resume training from the checkpoint where it last stopped. Specifically, the number of training iteration is also loaded when resuming here, which is bigger than 800*100
and satisfies this condition to evaluates every 200 iterations, if you load our pre-trained model.
Hello, I have a question about the resume of the training. When I try to load a checkpoint it evaluates every 200 iterations for 15 epochs. Is it meant to be like this when resuming training from the checkpoint? Or the loading of the model is meant only for evaluation purpose?
I am investigating, if I find an answer I will write it here Thank you