Repository navigation
FIX: Count each jailbroken response once in FuzzerGenerator - #4231
Open
Siddhant Rajhans (siddhant-rajhans) wants to merge 1 commit into
Open
Siddhant Rajhans (siddhant-rajhans) wants to merge 1 commit into
Siddhant Rajhans (siddhant-rajhans) wants to merge 1 commit into
Conversation
A scorer can return several scores for one response, for example AzureContentFilterScorer with one score per harm category. The fuzzer counted every passing score as a jailbreak, so one response could be counted and listed several times, and target_jailbreak_goal_count could be reached before that many responses were jailbroken. A response now counts once when any of its scores passes, and the MCTS reward is the jailbroken share of the template's responses, as in GPTFuzzer. Signed-off-by: Siddhant Rajhans <51508945+siddhant-rajhans@users.noreply.github.com>
|
Siddhant Rajhans (@siddhant-rajhans) please read the following Contributor License Agreement(CLA). If you agree with the CLA, please reply with the following information.
Contributor License AgreementContribution License AgreementThis Contribution License Agreement (“Agreement”) is agreed to by the party signing below (“You”),
|
This branch has not been deployed
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Description
Fixes #4230.
FuzzerGeneratorcounted a jailbreak for every passing score. A scorer that returns several scores per response, such asAzureContentFilterScorerwith one per harm category, made one jailbroken response count several times, listed its conversation several times injailbreak_conversation_ids, and could stop attarget_jailbreak_goal_countbefore that many responses were jailbroken._process_scoring_resultscounts a response once when any of its scores passes and records its conversation once. Attribution by the scored piece id (FIX: attribute fuzzer jailbreak scores through the scored piece id #3300) is unchanged.succ_num / (len(questions) * len(prompt_nodes)), one verdict per response)._process_scoring_resultsand throughexecute_async, and the reward for two scores per response.The repro from #4230 (goal of 2 jailbreaks, two of four categories above the threshold):
queries: 1, jailbreak conversations: 2 (1 distinct)queries: 2, jailbreak conversations: 2 (2 distinct)