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

[Question] training params of lora files #17

Open
yuwon opened this issue Aug 17, 2024 · 3 comments
Open

[Question] training params of lora files #17

yuwon opened this issue Aug 17, 2024 · 3 comments

Comments

@yuwon
Copy link

yuwon commented Aug 17, 2024

Thank you for the great work!!

In the HF, there are several models with different setting like normalcfg vs. smallcfg, 2step to 16 step.
I'm wondering what was the training params for those models?

@yuwon yuwon changed the title training params of lora files [Question] training params of lora files Aug 17, 2024
@G-U-N
Copy link
Owner

G-U-N commented Aug 18, 2024

Thanks for the interest! You can refer to the comments in the run.sh.

LMK if there are any further questions.

@yuwon
Copy link
Author

yuwon commented Aug 18, 2024

Thank you for the fast response!
I found some different hparams there and have questions about the params.

What is not_apply_cfg_solver for? It seems like unconditioned output is set to the conditioned output. Let me know if I'm wrong.

And what was the training params for the models with different steps? e.g., 2step and 16step loras. Is it controlled by multiphase? e.g., for 4steps lora, set multiphase to 4.

@G-U-N
Copy link
Owner

G-U-N commented Aug 24, 2024

Yes. You are right. All the other params are set the same.

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