Merge lora weights back to model

[yliyli]

Hi!

Can someone help explain how to merge lora weights back to base model (stable diffusion xl 1.0, for example)? Just using pipe.save_pretrained after loading pipeline and load lora weights fails, because the merged model changes in parameter names and is no longer accepted by diffuserspipeline. And if I use merge_and_unload like in here: https://huggingface.co/docs/peft/main/en/developer_guides/lora autotrain lora wegiths lacks config.json so this fails too.

Thanks a lot if anyone could suggest possible solutions!