accelerate launch --num_cpu_threads_per_process 1 sdxl_train_network.py \
    --pretrained_model_name_or_path="ponyDiffusionV6XL_v6.safetensors" \
    --train_data_dir lora-training/input \
    --output_dir lora-training/output \
    --reg_data_dir lora-training/reg \
    --shuffle_caption --caption_extension='.txt' --xformers --clip_skip=2 --mixed_precision='bf16' --seed=23 --max_data_loader_n_workers=1 --persistent_data_loader_workers --max_token_length=225 \
    --prior_loss_weight=1 --max_grad_norm=1 \
    --resolution=1024 --train_batch_size=3 \
    --network_dim=32 --network_alpha=16 \
    --optimizer_type='Lion' --lr_scheduler='cosine_with_restarts' --lr_scheduler_num_cycles=3 --lr_scheduler_power=1 \
    --learning_rate=0.0001 --unet_lr=0.0001 --text_encoder_lr=0.00005 \
    --save_precision='fp16' \
    --output_name "${1}" \
    --logging_dir lora-training/log \
    --log_prefix "${1}_" \
    --save_every_n_steps=250 \
    --enable_bucket --min_bucket_reso 256 --max_bucket_reso=2048 --bucket_reso_steps=64 \
    --max_train_steps=${2} \
    --cache_latents \
    --optimizer_args "weight_decay=0.1" "betas=0.9,0.99" \
    --clip_skip=2 \
    --mixed_precision="bf16" \
    --save_precision="fp16" \
    --cache_latents_to_disk \
    --gradient_accumulation_steps="1" \
    --gradient_checkpointing \
    --scale_weight_norms="1" \
    --min_snr_gamma=5 \
    --gradient_checkpointing \
    --noise_offset=0.05 \
    --adaptive_noise_scale=0.005 \
    --fp8_base \
    --sample_sampler=euler_a --sample_prompts="lora-training/prompt.txt" --sample_every_n_steps=100 \
    --network_module=networks.lora


-----


accelerate launch \
        --num_cpu_threads_per_process=2 "./sdxl_train_network.py" \
        --enable_bucket \
        --min_bucket_reso=256 \
        --max_bucket_reso=2048 \
        --resolution="1024,1024" \
        --training_comment="please kill me" \
        --save_model_as=safetensors \
        --text_encoder_lr=0.00007 \
        --unet_lr=0.00015 \
        --learning_rate=0.00015 \
        --network_dim=16 \
        --network_alpha="8" \
        --scale_weight_norms="1" \
        --no_half_vae \
        --optimizer_type="AdamW" \
        --lr_scheduler="cosine" \
        --train_batch_size="1" \
        --cache_latents \
        --cache_latents_to_disk \
        --max_grad_norm="1" \
        --max_data_loader_n_workers="0" \
        --max_token_length=225 \
        --bucket_reso_steps=64 \
        --min_snr_gamma=5 \
        --gradient_checkpointing \
        --noise_offset=0.05 \
        --adaptive_noise_scale=0.005 \
    --pretrained_model_name_or_path="ponyDiffusionV6XL_v6.safetensors" \
    --train_data_dir lora-training/input \
    --output_dir lora-training/output \
    --reg_data_dir lora-training/reg \
    --shuffle_caption --caption_extension='.txt' --xformers --clip_skip=2 --seed=23 --max_data_loader_n_workers=1 --persistent_data_loader_workers --max_token_length=225 \
    --keep_tokens="1" \
    --prior_loss_weight=1 --max_grad_norm=1 \
    --resolution=1024 --train_batch_size=1 \
    --network_dim=8 --network_alpha=8 \
    --output_name "${1}" \
    --logging_dir lora-training/log \
    --log_prefix "${1}_" \
    --save_every_n_steps=250 \
    --enable_bucket --min_bucket_reso 256 --max_bucket_reso=2048 --bucket_reso_steps=64 \
    --max_train_steps=${2} \
    --cache_latents \
    --mixed_precision="bf16" \
    --save_precision="fp16" \
    --cache_latents_to_disk \
    --gradient_accumulation_steps="1" \
    --gradient_checkpointing \
    --scale_weight_norms="1" \
    --min_snr_gamma=5 \
    --gradient_checkpointing \
    --noise_offset=0.05 \
    --adaptive_noise_scale=0.005 \
    --fp8_base \
    --sample_sampler=euler_a --sample_prompts="lora-training/${1}.txt" --sample_every_n_steps=100 \
    --network_module=networks.lora