This is my Luna Flex v3 model in Illustrious, I hope you like it. Please let me know if there are any recommendations.
Epochs = 12
Num Repeats = 3
Train Batch Size = 5
Steps = 5000
Resolution = 1024
Keep Tokens = 0
Clip Skip = 1
Unet LR = 0.00050
Text Encoder LR = 0.00005
LR Scheduler = cosine
LR Scheduler Cycles = 3
Min SNR Gamma = 0
Network Dim = 32
Network Alpha = 32
Noise Offset = 0.03
Optimizer = Prodigy
Optimizer Args = weight_decay=0.5, decouple=True, betas=0.9,0.99, use_bias_correction=False
Description
Comments (1)
Any chance for a safetensors version? Pickle formats (.pt, .ckpt, .bin etc) are unsafe in a way no app can generate this lora, except if we edit files like lora.py in load.torch function and force weights false, with is unsafe and too much 'tech savvy' for us poor enthusiasts.