| accelerate launch \ | |
| --mixed_precision bf16 \ | |
| --num_cpu_threads_per_process 1 \ | |
| sd-scripts/flux_train_network.py \ | |
| --pretrained_model_name_or_path "/ai/fluxgym/models/unet/flux1-dev.sft" \ | |
| --clip_l "/ai/fluxgym/models/clip/clip_l.safetensors" \ | |
| --t5xxl "/ai/fluxgym/models/clip/t5xxl_fp16.safetensors" \ | |
| --ae "/ai/fluxgym/models/vae/ae.sft" \ | |
| --cache_latents_to_disk \ | |
| --save_model_as safetensors \ | |
| --sdpa --persistent_data_loader_workers \ | |
| --max_data_loader_n_workers 16 \ | |
| --seed 0 \ | |
| --gradient_checkpointing \ | |
| --mixed_precision bf16 \ | |
| --save_precision bf16 \ | |
| --network_module networks.lora_flux \ | |
| --network_dim 32 \ | |
| --optimizer_type adamw8bit \ | |
| --learning_rate 8e-4 \ | |
| --cache_text_encoder_outputs \ | |
| --cache_text_encoder_outputs_to_disk \ | |
| --fp8_base \ | |
| --highvram \ | |
| --max_train_epochs 9 \ | |
| --save_every_n_epochs 4 \ | |
| --dataset_config "/ai/fluxgym/outputs/tos-flux-dev/dataset.toml" \ | |
| --output_dir "/ai/fluxgym/outputs/tos-flux-dev" \ | |
| --output_name tos-flux-dev \ | |
| --timestep_sampling shift \ | |
| --discrete_flow_shift 3.1582 \ | |
| --model_prediction_type raw \ | |
| --guidance_scale 1 \ | |
| --loss_type l2 \ | |
| --xformers |