Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Erasing multiple concepts. #34

Open
JiteshRajpal opened this issue Nov 2, 2024 · 1 comment
Open

Erasing multiple concepts. #34

JiteshRajpal opened this issue Nov 2, 2024 · 1 comment

Comments

@JiteshRajpal
Copy link

JiteshRajpal commented Nov 2, 2024

First off, I want to say a big thank you to @rohitgandikota for the code – it was super helpful and worked really well for me! πŸ™Œ
I'm currently working on erasing multiple concepts from my model (for example: cat, dog, lion, tiger, etc.) and had a couple of questions on best practices:

Handling Multiple Concepts: Is there a specific way to erase multiple concepts at once? Should I list all the concepts in one prompt, separated by commas? Or is it more effective to erase each concept individually?

File Format: After erasing a concept, I noticed that the generated file is saved in .pt format. To ensure compatibility, should I convert these .pt files into .ckpt format, or is it okay to keep them as .pt?

Any guidance on the above would be really appreciated! Thanks in advance for the help!

@rohitgandikota
Copy link
Owner

Oh good question! Ideally you should be able to handle it in a single erasure run. (The previous code managed that case, the current code doesn't). It should be easy to implement it. The idea is that during training, you can sample a different concept from the list of concepts you want to erase. This should do the trick

Regarding .pt file - you can convert it to any format, but ensure that you also change the load code accordingly (if needed)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants