hey @Dr. Furkan Gözükara --block_lr training to lower TE LR on sdxl. I m readin github #855 about it
hey @Dr. Furkan Gözükara --block_lr training to lower TE LR on sdxl. I m readin github #855 about it, where you engagae with kohya. You got some amazing gen's out of --block_lr. So did you find that process worked well in comparison with not training TE? I might give this a try?