lukasz-staniszewski
commented
2 months ago This repository hasn't been updated with the new diffusers version. In version 0.25.0, where, i.e. peft was introduced, the train_dreambooth_lora_sdxl.py file was entirely updated, and there is no set_lora_layer used there anymore.
I managed to run these older scripts (the ones in this repo) for both LoRA and ZipLoRA fine-tuning by downgrading most of the modules. Here is my requirements.txt file that enables me to run LoRA training in this repo; I hope it helps:
accelerate==0.29.3
bitsandbytes==0.43.1
diffusers==0.24.0
ftfy==6.2.0
huggingface-hub==0.22.2
Jinja2==3.1.3
numpy==1.26.4
pillow==10.3.0
pytorch-cuda==12.1
safetensors==0.4.3
tensorboard==2.16.2
torch==2.2.2
torchvision==0.17.2
tqdm==4.66.2
transformers==4.40.1
triton==2.2.0
wandb==0.16.6
xformers==0.0.25.post1
Whether I uninstall peft or not, I am still facing this bug, can someone tell me the solution.