| accelerate launch \ | |
| --mixed_precision bf16 \ | |
| --multi_gpu \ | |
| --gpu_ids 0,1 \ | |
| --num_processes 2 \ | |
| --num_cpu_threads_per_process 1 \ | |
| sd-scripts/flux_train_network.py \ | |
| --pretrained_model_name_or_path "/tmp/fluxgym/models/unet/flux1-dev-fp8.safetensors" \ | |
| --clip_l "/tmp/fluxgym/models/clip/clip_l.safetensors" \ | |
| --t5xxl "/tmp/fluxgym/models/clip/t5xxl_fp8_e4m3fn.safetensors" \ | |
| --ae "/tmp/fluxgym/models/vae/ae.sft" \ | |
| --cache_latents_to_disk \ | |
| --save_model_as safetensors \ | |
| --sdpa --persistent_data_loader_workers \ | |
| --max_data_loader_n_workers 2 \ | |
| --seed 42 \ | |
| --gradient_checkpointing \ | |
| --mixed_precision bf16 \ | |
| --save_precision fp16 \ | |
| --network_module networks.lora_flux \ | |
| --network_dim 32 \ | |
| --optimizer_type adafactor \ | |
| --optimizer_args "relative_step=False" "scale_parameter=False" "warmup_init=False" \ | |
| --lr_scheduler constant_with_warmup \ | |
| --max_grad_norm 0.0 \ | |
| --learning_rate 1e-4 \ | |
| --cache_text_encoder_outputs \ | |
| --cache_text_encoder_outputs_to_disk \ | |
| --fp8_base \ | |
| --highvram \ | |
| --max_train_epochs 10 \ | |
| --save_every_n_epochs 1 \ | |
| --dataset_config "/tmp/fluxgym/outputs/aidka/dataset.toml" \ | |
| --output_dir "/tmp/fluxgym/outputs/aidka" \ | |
| --output_name aidka \ | |
| --timestep_sampling shift \ | |
| --discrete_flow_shift 3.1582 \ | |
| --model_prediction_type raw \ | |
| --guidance_scale 1 \ | |
| --blocks_to_swap 10 \ | |
| --loss_type l2 \ |