V1.1
Trained at 1024 resolution
2295 steps
avr_loss = 0.0802
LR = 0.0002
Noise offset = 0
V1.0
AdamW8Bit Optimizer
2340 steps
LR 0.00002