Trained on only some of the images provided, and for much fewer steps than expected

[an303042]

Hello, First off - Thank you! Amazing work! Truly.

My issue is that I tried to train a LoRA, had 105 pictures and captions. Brought in to the interface, all are showing with their captions. Wasn't sure how many repeats and epochs I needed, but in the past I trained LoRAs either up to 1k steps or 3k, so on the left side I turned down the repeats an epochs to 6 and 6, which showed "7560" steps in the "expected training steps" field, and that seemed ok for me (again, no idea). I ran the training, and when it finished was surprised to see it only trained 2736 steps! after looking at the log I saw it only used the first 76 images from the dataset, not all 105.

So I have two questions -

1. Why didn't it use all 105 dataset images?
2. Why did it train for so few steps when it was supposed to go further?

Log attached file.txt

Again, many thanks 🙏

[GavChap]

Were any unusual formats like avif or webp? You'll need to convert them to .jpg or .png first