[SUPPORT] Workbench Example Project: SDXL Customization

Hi! This is the support thread for the SDXL Customization Example Project on GitHub. Any major updates we push to the project will be announced here. Further, feel free to discuss, raise issues, and ask for assistance in this thread.

Please keep discussion in this thread project-related. Any issues with the Workbench application should be raised as a standalone thread. Thanks!

An update has been pushed to correct an issue with the PEFT package dependency. Please update your repos accordingly, and let us know of any additional issues. Thanks!

Found this error in the Notebook when trying to run the King Charles section:

The lora-weights file is missing in the Model/tuned-charles-3 folder. I copied the one from tuned-toy-jensen into it.

One more issue - when the current FineTuning-SDXL code is updated, it exceeds the GitHub file size limit of 5MB. I only have the choice to move or delete the file once that happens, as Workbench throws an exception that does not allow me to commit files back to GitHub.

Hey Brian, thanks for reaching out. That’s strange. Usually the model/tuned-charles-3 folder is populated with the weights of the fine-tuned model after running the fine-tuning cell. Perhaps you forgot to run a cell or populate the charles-3 folder with images for the model to fine-tune on? Not exactly sure. Try restarting the kernel and running the cells again, and let me know if the issue persists.

1 Like

I believe the file size comes from the images that get generated/displayed in-notebook. I believe if you go to kernel → clear outputs, the images get removed from the notebook and that reduces the file size enough to commit the notebook to Github. The fine-tuned model weights will still be in the models folder for future use–you can open the notebook and run the right cells again to generate images at a later time. Hope this helps!

1 Like

Thanks - I ran the script as written, but it may be that the code was missing a line to generate this. I’ve pointed this out to our NVIDIA resource and he is re-writing the code.

Restarting from scratch. NVWB is installed and working properly. I cloned the SDXL project from GitHub. I am logged into both GitHub and Huggingface. My internet connection is good. When I cloned the SDXL project from GitHub, the setup begins and end then Crashes with Error connecting to local. I have attached a copy of my log files. How do I fix this?
nvwblogs_02152024.txt (8.4 KB)

Hi Mark

Were you using the CLI for this?

I note the log: {"level":"error","isWrapped":true,"isInteractive":true,"engine":"markdown","output":"**you must provide a context name when calling 'nvwb activate <context_name>'**","time":"2024-01-18T18:53:28-05:00","message":"you must provide a context name when calling 'nvwb activate <context_name>'"}

This suggests you ran nvwb activate without a location argument, i.e. nvwb activate local.

Hi, No CLI, I am doing this on the NVWB installed in Windows WSLX env. Just leveraging the GUI to do everything.

1 Like