MAINT: Refactor Skeleton Key to be a subclass of PromptSendingOrchest…#650
Conversation
|
@microsoft-github-policy-service agree |
| PromptConverterConfiguration, | ||
| ) | ||
| from pyrit.prompt_target import PromptTarget | ||
| from pyrit.orchestrator.single_turn.prompt_sending_orchestrator import PromptSendingOrchestrator |
There was a problem hiding this comment.
| from pyrit.orchestrator.single_turn.prompt_sending_orchestrator import PromptSendingOrchestrator | |
| from pyrit.orchestrator.single_turn import PromptSendingOrchestrator |
|
|
||
|
|
||
| class SkeletonKeyOrchestrator(Orchestrator): | ||
| class SkeletonKeyOrchestrator(PromptSendingOrchestrator): |
There was a problem hiding this comment.
Recommend moving this file to pyrit/orchestrator/single_turn
There was a problem hiding this comment.
(and as part of that move, updating init.py
| PromptRequestResponse: The response from the prompt target. | ||
| list[PromptRequestResponse]: The responses from the prompt target. | ||
| """ | ||
| if hasattr(self._prompt_target, 'rpm') and self._prompt_target.rpm and self._batch_size != 1: |
There was a problem hiding this comment.
This should be removed - it's the responsibility of the target to check this
|
|
||
| # Create a single conversation ID for the entire sequence | ||
| conversation_id = str(uuid4()) | ||
| metadata = {"conversation_id": conversation_id} |
There was a problem hiding this comment.
remove the conversation_id as metadata. PromptRequestPieces have this as an attribute and it can be set there
| # Return the attack prompt response (second response) | ||
| return responses[1] | ||
|
|
||
| def print_conversation(self) -> None: |
There was a problem hiding this comment.
you should be able to remove print_conversation and use the parent class
| Returns: | ||
| list[PromptRequestResponse]: The responses from the prompt target. | ||
| PromptRequestResponse: The response from the prompt target. | ||
| """ |
There was a problem hiding this comment.
Instead of doing it like how it was done, I would make use of PromptSendingOrchestrator.
- Use the
set_prepended_conversationto set the skeleton key. This can be done in init - You can remove the
send_skeleton_key_with_prompt_asyncfunction. Sending any prompt will then have the skeleton key prepended
There was a problem hiding this comment.
But if you prepend you have to prepend the response as well, right? What would you set as the response?
| batch_size=batch_size, | ||
| verbose=verbose | ||
| ) | ||
|
|
There was a problem hiding this comment.
pre-commit is failing. You can run it using pre-commit run --all-files; And you can see the current issues in the CI build: https://github.com/Azure/PyRIT/actions/runs/12876675701/job/35900136914?pr=650
|
Closing for now, but feel free to re-open if you want us to take another look! |
3 participants
Overview
This PR refactors the SkeletonKeyOrchestrator to properly inherit from PromptSendingOrchestrator instead of Orchestrator. The refactoring improves code organization and reliability while maintaining all existing functionality. It focuses on leveraging the parent class's capabilities for prompt handling and conversation management.
Key Features
Method Refactoring:
send_skeleton_key_with_prompts_asyncto use parent class's batching capabilitiessend_skeleton_key_with_prompt_asyncto maintain consistent conversation context_create_normalizer_requestfor request creationRequest Handling:
Memory Management:
Backward Compatibility:
Related Issues
MAINT: Refactor Skeleton Key to be a subclass of PromptSendingOrchestrator #599
Next Steps (DONE):