>>37006569>Do not use the --medvram or --lowvram argument when training! Results will be highly inferiorI made sure I didn't have this on for any of my training. It really was all the same settings and data, just a better graphics card and that resulted in a better embed. Makes me think there's some sort of limiting going on out of sight somewhere, maybe not necessarily in VRAM but it could be in operation speed or simultaneous operations or something.
I guess I may as well check out xformers too if I find the time, always nice to get a look at what all the options are even if it turns out to not be as good
