epochs aren't being made, slow training

[joshgura]

sdxl, 10 images (768x1024), 35 repeats, 10 epochs, and 3500 steps are set in the "max train steps" box.

am i correct to assume that about every 350 steps there should be an epoch made?

i'm at 1800/3500 steps, about an hour in to a 2 hour training process, and no epochs or samples have been made and I don't know why. even if i set the sampler to make an image every 350 steps, it's ignored and not making any samples.

also i'm using an RTX 3060 and it seems like training should be happening a lot faster than 2 hours for a 10 images lora.

any idea what i could be doing wrong to have such a slow and uneventful training process?

I've trained loras before like a week ago and it was spitting out epochs fine. i just lost the plot on what the settings should be and i did a fresh reinstall of kohya because it stopped working for some reason. it's training now but the epochs aren't incrementing.

something weird is that i only have 10 images in the working folder and for some reason kohya thinks there's "10340 images with repeating". i don't really see how that's correct.

I just killed the training and set it up to do 10 repeats and 1000 total steps, 10 epochs, and i'm still seeing this num batches per epoch / 1epochのバッチ数: 10340 and i don't know where that number is coming from, but it seems to be a clue that something is wrong.

edit: i think i had some kind of issues with the naming of the image folder. i monkeyed around with everything and somehow got it to work correctly. there's just too many parameters to know what happened. it's working now but i am still curious how this happened.