training will not stop when epoch limit reached

[zhuliyi0]

When epoch number is specified, current code only looks for global step limit, only breaks the step loop at the end of training, and it will execute some extra number of steps equal to the number of epochs.

[zhuliyi0]

Super cool project BTW, still messing around paremeters, but the results so far are very promissing, thanks!

[bghira]

this is on SDXL or SD 2.x?

[zhuliyi0]

all the issues I had was on SDXL, havn't touched 2.x

[bghira]

ok so i identified that the style of the sampler might result in just ignoring the end of an epoch until we resume training, at which time it then boosts the epoch number.

i have put #113 for review. can you look?