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

A few questions #2

Open
greeneggsandyaml opened this issue May 7, 2020 · 0 comments
Open

A few questions #2

greeneggsandyaml opened this issue May 7, 2020 · 0 comments

Comments

@greeneggsandyaml
Copy link

Thank you so much for your implementation! I have enjoyed reading your code.

I'm playing around with the method, but unfortunately I don't have access to much compute at the moment. Since it seems you've done a good amount of experimentation, I figured that I would try asking you a few questions about the method. Thank you so much in advance!

  • How long does it take to run a model on CIFAR-10 on a single GPU (e.g. a 2080TI)? From a quick estimates, it looks like it should take around 6 min/epoch * 1000 epochs = 6000 minutes = 4 days. Does this match what you observe?
  • Do you know how long does it take to obtain a "reasonably-good" result (e.g. 91% for CIFAR-10@250 or 65% for CIFAR-100@2500?
  • Based on your experimentation, how important is the EMA of model parameters? If you were to evaluate without the EMA, do you know (approximately) what accuracy you would get?
  • Based on your experimentation, do you have a sense of how sensitive the model is to hyperparameter changes? Is it relatively robust, or will performance drop drastically if a few are changed?
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

1 participant