Quoted By:
TIL kohya implemented so many systems but could not implement a simple write to text file for the number of steps already trained when processing the save state so when you resume training it doesn't look like it starts from step 1 again. And I deleted a 50k step training yesterday because I thought it was borked and it restarted from step 1.
Picrel henya 1.6gb lokr test after about 1000 steps. Hopefully the big one works this well too. First epoch should be done in like 24 hours but I don't expect it to know too much, maybe after epoch 2.
