colossalai.sh

export MODEL_NAME="CompVis/stable-diffusion-v1-4"
export INSTANCE_DIR="input"
export OUTPUT_DIR="output"
INSTANCE_PROMPT="a photo of sks dog"
HF_DATASETS_OFFLINE=1
TRANSFORMERS_OFFLINE=1

torchrun --nproc_per_node 2 --master_port=25641 train_dreambooth_colossalai.py \
  --pretrained_model_name_or_path=$MODEL_NAME  \
  --instance_data_dir=$INSTANCE_DIR \
  --output_dir=$OUTPUT_DIR \
  --resolution=512 \
  --train_batch_size=1 \
  --gradient_accumulation_steps=1 \
  --learning_rate=5e-6 \
  --instance_prompt=INSTANCE_PROMPT \
  --lr_scheduler="constant" \
  --lr_warmup_steps=0 \
  --max_train_steps=400 \
  --placement="cpu"