Skip to content

PGD Attack is not working for me for even the most easy prompts and models #1

Description

@valentyn1boreiko

Thank you for your implementation of the PGD attack!

When I check the confidence in the target token before the discretization - it is increased, but after discretization is not increasing and stays very low, thus attack seems not to work. Is there anything that I should change in the script?

Best,
Valentyn

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions