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 about training time cost #12

Open
DachunKai opened this issue Aug 30, 2024 · 3 comments
Open

Question about training time cost #12

DachunKai opened this issue Aug 30, 2024 · 3 comments

Comments

@DachunKai
Copy link

Hello,

Could you please specify the training time cost for the IART model on REDS, such as the IART_REDS_BI_N16.pth? I'm interested in knowing what type of GPU was used, the number of GPUs involved, and approximately how many days the training took.

Thank you!

@PhuTran1005
Copy link

Hello,

Could you please specify the training time cost for the IART model on REDS, such as the IART_REDS_BI_N16.pth? I'm interested in knowing what type of GPU was used, the number of GPUs involved, and approximately how many days the training took.

Thank you!

@kai422
Copy link
Owner

kai422 commented Oct 21, 2024

Training the model IART_REDS_BI_N16.pth on a V100 took four weeks. Since I only had 16Gx8 of VRAM available, I enabled gradient checkpointing, which is a memory-saving technique. However, this also means that during backpropagation, the forward pass gets recalculated. If you have more modern GPUs with larger VRAM (so you can disable gradient checkpointing), such as an A100 80G, I estimate that the training could be completed within a week.

Regards,
Kai

@PhuTran1005
Copy link

Thank you for your detailed response.

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

3 participants