Training details for the Jinx LoRA + RefMod experiment?
#7
by ExDrcw - opened
Hi! Thanks for sharing the Jinx LoRA + RefMod comparison. I’d love to try this with my own character dataset. Could you share how you trained minimaxh3_jinx_v1—which trainer and base checkpoint you used, whether the dataset contained images, videos, or both, and how you captioned them? I’m also curious about the key settings: rank/alpha, learning rate, optimizer, batch size, dataset repeats, and total steps at epoch 60. Did you use a training adapter or guidance loss, and was the RefMod made from the same dataset? A training config would be really helpful. Thanks!