#!/bin/bash set -xe pip install -r requirements.txt HF_DATASETS_OFFLINE=1 TRANSFORMERS_OFFLINE=1 DIFFUSERS_OFFLINE=1 for plugin in "torch_ddp" "torch_ddp_fp16" "gemini" "low_level_zero"; do torchrun --nproc_per_node 4 --master_port=25641 train_dreambooth_colossalai.py \ --pretrained_model_name_or_path="Your Pretrained Model Path" \ --instance_data_dir="Your Input Pics Path" \ --output_dir="path-to-save-model" \ --instance_prompt="your prompt" \ --resolution=512 \ --plugin=$plugin \ --train_batch_size=1 \ --learning_rate=5e-6 \ --lr_scheduler="constant" \ --lr_warmup_steps=0 \ --num_class_images=200 \ --placement="cuda" done