LoRA 训练 新手模式
默认设置为你准备好了所有需要的参数，只需要你修改底模路径、训练集路径、训练轮数即可一键训练模型。
pretrained_model_name_or_path = "./sd-models/model.safetensors"
train_data_dir = "./train/aki"
resolution = "512,512"
output_name = "aki"
output_dir = "./output"
save_every_n_epochs = 2
max_train_epochs = 10
train_batch_size = 1
unet_lr = 0.0001
text_encoder_lr = 0.00001
lr_scheduler = "cosine_with_restarts"
lr_warmup_steps = 0
lr_scheduler_num_cycles = 1
optimizer_type = "AdamW8bit"
network_dim = 32
network_alpha = 32
shuffle_caption = true
keep_tokens = 0
mixed_precision = "fp16"
xformers = true
cache_latents = true