| accelerate launch --num_cpu_threads_per_process=2 "./sdxl_train_network.py" --enable_bucket --min_bucket_reso=256 --max_bucket_reso=2048 --pretrained_model_name_or_path="/workspace/model.safetensors" --train_data_dir="/workspace/kohya_ss/dataset/train_data" --resolution="1024,1024" --output_dir="/workspace/kohya_ss/dataset/result-4000" --logging_dir="/workspace/logs" --network_alpha="16" --save_model_as=safetensors --network_module=lycoris.kohya --text_encoder_lr=1 --unet_lr=1 --network_dim=32 --output_name="<REPLACE>" --no_half_vae --lr_scheduler="cosine_with_restarts" --train_batch_size="4" --max_train_steps="4000" --save_every_n_epochs="5" --mixed_precision="bf16" --save_precision="bf16" --seed="12345" --caption_extension=".txt" --cache_latents --cache_latents_to_disk --optimizer_type="Prodigy" --max_data_loader_n_workers="0" --bucket_reso_steps=64 --xformers --bucket_no_upscale --max_grad_norm=0 --optimizer_args="decouple=True" --optimizer_args="weight_decay=0.01" --optimizer_args="d_coef=0.5" --optimizer_args="use_bias_correction=True" --optimizer_args="safeguard_warmup=True" --lr_scheduler_num_cycles="1" --shuffle_caption --keep_tokens=1 |