Attack fixes #555
Merged
Attack fixes #555
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Not sure why the cohere library is causing failures. We should fix that in another branch. |
…robe to clean up generator state.
… often with ConversationalPipeline.
…s are specified alongside target_generator
…rget. Check for success in GCG attack manager.
…e attack success detection.
…for conversational models; Raise and log a more useful error in clean_attacks_and_convs
…mprove checking for conversion to OpenAI API messages.
erickgalinkin
force-pushed
the
attack_fixes
branch
from
0d9e96c
to
9b975b4
Compare
|
Rebased and force-pushed to fix Cohere error. |
jmartin-tech
reviewed
Apr 2, 2024
…ators without instantiating twice.
… probe. Add error handling to generative GCG, TAP, PAIR, and AutoDAN probes.
… target models in TAP to work better with VRAM. Modify GCG, TAP, and AutoDAN probes to return an empty list to avoid assertion errors. Fix AutoDAN mutation generator.
leondz
reviewed
Apr 5, 2024
leondz
reviewed
Apr 5, 2024
leondz
reviewed
Apr 5, 2024
leondz
reviewed
Apr 5, 2024
leondz
reviewed
Apr 5, 2024
leondz
reviewed
Apr 5, 2024
leondz
reviewed
Apr 5, 2024
leondz
reviewed
Apr 5, 2024
| @@ -1429,7 +1431,7 @@ def run( | |||
| test_steps: int = 50, | |||
| incr_control: bool = True, | |||
| stop_on_success: bool = True, | |||
| verbose: bool = True, | |||
| verbose: bool = False, | |||
There was a problem hiding this comment.
can this be taken from garak._config.system.verbose?
There was a problem hiding this comment.
It can! That's a good one.
leondz
reviewed
Apr 5, 2024
leondz
reviewed
Apr 5, 2024
leondz
reviewed
Apr 5, 2024
leondz
reviewed
Apr 5, 2024
leondz
reviewed
Apr 5, 2024
leondz
reviewed
Apr 5, 2024
leondz
reviewed
Apr 5, 2024
| target_max_tokens: int = 150, | ||
| evaluator_max_tokens: int = 10, | ||
| evaluator_temperature: float = 0.0, | ||
| self, |
There was a problem hiding this comment.
hey do you know how these params can be set when invoking via the probe?
There was a problem hiding this comment.
Looks like we need a ProbeConfig class in _config.
There was a problem hiding this comment.
values in _config.plugins.probes.ClassName are automatically set in classes at instantiation time - just wondering a bit about the stuff in resources/. maybe let's wait for demand before triaging dealing with that
leondz
reviewed
Apr 5, 2024
leondz
reviewed
Apr 5, 2024
leondz
reviewed
Apr 5, 2024
leondz
reviewed
Apr 5, 2024
leondz
approved these changes
Apr 5, 2024
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Fixes and improvements to TAP, GCG, AutoDAN.