| export MODEL_NAME="CompVis/stable-diffusion-v1-4" | |
| export DATA_DIR='./datasets/Open_Images/' | |
| CUDA_VISIBLE_DEVICES=0,1,2,3 accelerate launch --config_file 4_gpu.json --main_process_port 25656 train_global.py \ | |
| --pretrained_model_name_or_path=$MODEL_NAME \ | |
| --train_data_dir=$DATA_DIR \ | |
| --placeholder_token="S" \ | |
| --resolution=512 \ | |
| --train_batch_size=4 \ | |
| --gradient_accumulation_steps=4 \ | |
| --max_train_steps=200000 \ | |
| --learning_rate=1e-06 --scale_lr \ | |
| --lr_scheduler="constant" \ | |
| --lr_warmup_steps=0 \ | |
| --output_dir="./elite_experiments/global_mapping" \ | |
| --save_steps 200 | |