Help setting up SDXL LoRA training on Google Cloud (Vertex AI Workbench)

[aici23]

Hi everyone — I’m creating an AI interior design app. I’ve already prepared high-quality image-text datasets (800+ images per style, with .txt captions) and I’m trying to train LoRAs using SDXL.

I’ve been trying to work on Google Cloud Vertex AI Workbench with the PEFT library, but I keep running into errors. mostly around LoRA injection into UNet and broken training loops. I’ve tried multiple scripts and fixes but nothing has worked reliably.

If anyone here has 10–15 minutes to help me via Zoom or even to look over my environment setup, I’d be incredibly grateful. I’m not a developer, but I’m serious about doing this right and just need help setting it up once properly.

Thanks in advance!

Ivana

[BenjaminBossan]

We (the PEFT maintainers) don't offer calls, but if you share your code, what exactly you already tried, and what errors you ran into, we might be able to help. If there is some proprietary code/data you can't share, try to replace it with non-proprietary code/openly available data (which is also a good exercise for debugging the issue).