-base model: F222
-100 training images, 1800 class images
-18 training repeats, 1 reg repeat
-64 dim, 16 alpha,
-Unet: 6e-5, TEnc: 75e-6
-Polynomial scheduler, 1.75 power,
-noise offset=0.03
-batch: 6 * 3 gradient steps
-12 epochs
-base model: F222
-100 training images, 1800 class images
-18 training repeats, 1 reg repeat
-64 dim, 16 alpha,
-Unet: 6e-5, TEnc: 75e-6
-Polynomial scheduler, 1.75 power,
-noise offset=0.03
-batch: 6 * 3 gradient steps
-12 epochs