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

Training Hyperparams #90

Open
vidit98 opened this issue Jul 15, 2024 · 7 comments
Open

Training Hyperparams #90

vidit98 opened this issue Jul 15, 2024 · 7 comments

Comments

@vidit98
Copy link

vidit98 commented Jul 15, 2024

Thanks for your great work! Can you please specify the number of training steps in first and second stage training mentioned in the paper?

@eliphatfs
Copy link
Collaborator

50k per phase.

@vidit98
Copy link
Author

vidit98 commented Jul 16, 2024

Thanks! Could you also please tell the batchsize and number of objects in the dataset?

@eliphatfs
Copy link
Collaborator

256, and the size of Objaverse.

@vidit98 vidit98 closed this as completed Jul 22, 2024
@vidit98
Copy link
Author

vidit98 commented Aug 2, 2024

Hi, I was wondering did you ever try training the model in single stage?

@vidit98 vidit98 reopened this Aug 2, 2024
@eliphatfs
Copy link
Collaborator

The generalization was worse in an experimental, unreleased version of the model. I am not 100% sure if this is the same case for the v1.2 architecture.

@vidit98
Copy link
Author

vidit98 commented Aug 15, 2024

May I also know the SNR gamma you used for Min SNR strategy? Thanks again 🙏

@eliphatfs
Copy link
Collaborator

It is the default or paper value, perhaps 5.0

@vidit98 vidit98 changed the title Number of training steps Training Hyperparams Aug 16, 2024
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