Daniel Kang

Hi Tomoki, thanks for the interest in our work. Could you tell us the exact model, command line parameters, and environment that you are using?

There can be many subtle issues that occur with adversarial training and attacks. It's difficult to debug without seeing code. Have you tried using our PGD-Linf trained model with our...

With our implementation of loss dropping, we require a batch dimension. Doing that should work. It should be a simple code change to deal with one-dimensional losses though!

I don't think you need the mean if it's already one dimensional.

Hi @ElderWanng you can take a pretrained model and continue training it with loss truncation - we find it works quite well.

Thank you for your interest in our work! While we recommend the following, your task-optimal hyperparameters might vary: - The weight should be as in the default hyperparameters for regular...

Hi @sshleifer thank you for trying out our code! Can you try two things: 1. Can you try using dropper with dropc=0? This should emulate regular training. If this doesn't...

@sshleifer apologies for the misunderstanding - I didn't realize you were using the cross entropy loss. You should be able to use loss dropping with cross entropy as well. I...

What about the `None` type?

That seems to have worked, thanks! It might be nice to mention that the indexes need to be applied before streaming.