[Path]
pretrained_model_name_or_path = "/notebooks/box/Trim.safetensors"
train_data_dir = "/notebooks/LoRA_Image/"

__root_dir = "/notebooks/LoRA/sd-scripts/"
logging_dir = "calc:f'{__root_dir}logs'"
output_dir = "calc:f'{__root_dir}outputs'"
sample_prompts = "/notebooks/LoRA_Setting/A.txt"

[Prompt]
sample_every_n_epochs = 1
sample_sampler = "k_euler_a"

[Else]
shuffle_caption = true
caption_extension = ".txt"
keep_tokens = 1

color_aug = true

__upper_px = 512
__resolution_x = "calc:512 + __upper_px"
__resolution_y = "calc:512 + __upper_px"

resolution = "calc:f'{__resolution_x},{__resolution_y}'"
enable_bucket = true
min_bucket_reso = "calc:320 + __upper_px"
bucket_no_upscale = true
caption_dropout_every_n_epochs = 9
caption_tag_dropout_rate = 0.2
save_every_n_epochs = 2
train_batch_size = 1
xformers = true
max_train_epochs = 14
persistent_data_loader_workers = true
seed = 987
mixed_precision = "fp16"
save_precision = "ref:mixed_precision"
clip_skip = 2

optimizer_type = "Lion"
__div_rate = 8
learning_rate = "calc:1.0e-4 / __div_rate"
text_encoder_lr= "calc:5.0e-5 / __div_rate"
unet_lr = "ref:learning_rate"

lr_scheduler = "cosine_with_restarts"
lr_warmup_steps = 500
lr_scheduler_num_cycles = 4

network_module = "networks.lora"
network_dim = 128
network_alpha = "calc:float(network_dim) / 2.0"

__upper_mixed_precision = "calc:mixed_precision.upper()"
output_name = "calc:f'Akashi_{optimizer_type}Div{__div_rate}1024Px'"