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

poor performance on CIFAR10 #3

Open
Peony-Wang opened this issue Jun 23, 2020 · 1 comment
Open

poor performance on CIFAR10 #3

Peony-Wang opened this issue Jun 23, 2020 · 1 comment

Comments

@Peony-Wang
Copy link

Peony-Wang commented Jun 23, 2020

Hi, thank you for your implementation!

I use the CIFAR10 dataset for quick training and evaluation on the 8-bit setting, but I only got not more than 80%@top1 accuracies with/without activation quantization. Could you give me any clue of why this happens? The learning rate is decreased by 10x every 30 epochs. But the accuracy stops increasing after around 40-50 epochs.

Thanks for your attention!

@ricky40403
Copy link
Owner

Hi Peony, sorry that I did not test on CIFAR10. Did you test that training without quantization? And what about the initial learning rate?

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