| accelerate launch train_textdiffuser2_t2i_full.py \ | |
| --pretrained_model_name_or_path="runwayml/stable-diffusion-v1-5" \ | |
| --train_batch_size=18 \ | |
| --gradient_accumulation_steps=4 \ | |
| --gradient_checkpointing \ | |
| --mixed_precision="fp16" \ | |
| --num_train_epochs=6 \ | |
| --learning_rate=1e-5 \ | |
| --max_grad_norm=1 \ | |
| --lr_scheduler="constant" \ | |
| --lr_warmup_steps=0 \ | |
| --output_dir="diffusion_experiment_result" \ | |
| --enable_xformers_memory_efficient_attention \ | |
| --dataloader_num_workers=8 \ | |
| --index_file_path='/path/to/train_dataset_index.txt' \ | |
| --dataset_path='/path/to/laion-ocr-select/' \ | |
| --granularity=128 \ | |
| --coord_mode="ltrb" \ | |
| --max_length=77 \ | |
| --resume_from_checkpoint="latest" |