unilm/textdiffuser-2/train_textdiffuser2_inpainting_full.sh

accelerate launch train_textdiffuser2_inpainting_full.py \
    --train_batch_size=16 \
    --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="inpainting_experiment_name" \
    --enable_xformers_memory_efficient_attention \
    --dataloader_num_workers=8 \
    --resume_from_checkpoint="latest" \
    --dataset_path=/path/to/laion-ocr-select \
    --train_dataset_index_file=/path/to/train_dataset_index.txt \
    --vis_num=16