The training question about diffste.

[thebestYezhang]

Thanks for your great work! I have a question. When I train the diffste in my own dataset, the training Progress bar only prints out the loss of half of total training steps, and the loss of the rest half of total training steps is not show by the Progress bar. I carefully reviewed the training process, but couldn't find the problem. May I ask if you were in the same situation at training time?

[Question406]

Hi, actually, the training loss printed in the terminal is the loss for a single batch instead of the average loss over the entire dataset. For the details of what is printed, please see this link.

The reason this progress bar is split in the middle of one epoch is the output during training, the line do_classifier_free_guidance; I guess this is some log information you print when you log some generated images.

[yang-chenyu104]

i want to say format of data,i want to train text editing

[yang-chenyu104]

when i training model,the path of log makes me very sad