>>66855911I kept trying but her lora simply doesn't train well on 1.5
>>66856156No idea. Probably better at {{{larger}}} gens.
>>66856158>oh noEspecially now that I can train them on the 7900xtx with a batch of 4 or 6 on 1.5.
J/k. collecting the dataset is what takes most of the time anyway