[TypeError] in DreamBooth SDXL LoRA training when use_dora parameter is False

[adhiiisetiawan]

Describe the bug

When running the DreamBooth SDXL training script with LoRA, it throws a TypeError even when use_dora=False (default). This happens because the use_dora parameter is always being passed to LoraConfig, regardless of whether DoRA is being used or not. I plan to submit a PR to fix this by conditionally including the use_dora parameter only when it's True.

Reproduction

1. Run the DreamBooth SDXL training script in examples/dreambooth/train_dreambooth_lora_sdxl.py
2. The script fails even with default parameters (where use_dora=False)

python3 train_dreambooth_lora_sdxl.py \
  --pretrained_model_name_or_path="stabilityai/stable-diffusion-xl-base-1.0" \
  --dataset_name="adhisetiawan/food-bakso-img" \
  --pretrained_vae_model_name_or_path="madebyollin/sdxl-vae-fp16-fix" \
  --output_dir="lora-trained-xl" \
  --use_8bit_adam \
  --mixed_precision="bf16" \
  --train_text_encoder \
  --instance_prompt="a photo of bakso" \
  --resolution=1024 \
  --train_batch_size=1 \
  --gradient_accumulation_steps=4 \
  --learning_rate=1e-4 \
  --lr_scheduler="constant" \
  --lr_warmup_steps=0 \
  --max_train_steps=500 \
  --validation_prompt="A photo of bakso in a bowl" \
  --validation_epochs=5 \
  --seed=0

Logs

{'reverse_transformer_layers_per_block', 'attention_type', 'dropout'} was not found in config. Values will be initialized to default values.
Traceback (most recent call last):
  File "/workspace/diffusers/examples/dreambooth/train_dreambooth_lora_sdxl.py", line 1986, in <module>
    main(args)
  File "/workspace/diffusers/examples/dreambooth/train_dreambooth_lora_sdxl.py", line 1187, in main
    unet_lora_config = LoraConfig(
TypeError: LoraConfig.__init__() got an unexpected keyword argument 'use_dora'

System Info

• 🤗 Diffusers version: 0.32.0.dev0
• Platform: Linux-5.4.0-153-generic-x86_64-with-glibc2.35
• Running on Google Colab?: No
• Python version: 3.10.12
• PyTorch version (GPU?): 2.1.0+cu118 (True)
• Flax version (CPU?/GPU?/TPU?): not installed (NA)
• Jax version: not installed
• JaxLib version: not installed
• Huggingface_hub version: 0.26.2
• Transformers version: 4.46.1
• Accelerate version: 1.1.0
• PEFT version: 0.7.0
• Bitsandbytes version: 0.44.1
• Safetensors version: 0.4.5
• xFormers version: not installed
• Accelerator: NVIDIA RTX A6000, 49140 MiB
• Using GPU in script?: Yes
• Using distributed or parallel set-up in script?: No

Who can help?

@sayakpaul

[sayakpaul]

You can get around the problem by upgrading your local peft installation.