You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Currently, for MultiTurnOrchestrators, you can set a prepended conversation. This is really useful, for example, if you want to set a system prompt on an objective target and then run Crescendo against it.
However, we want to make this more generic so we can prepend specific prompts based on the objective that's being used.
self._prepended_conversation should be replaced by
We also need to update set_prepended_conversation to include an objectiver (optionally). A good behavior might be to store these in a dictionary such that get_prepended_conversation returns:
A default value if it's set and there is no match and set_prepended_conversation has been set for default
A specific prepended conversation if the objective matches. This can be a hash also/exact match
The text was updated successfully, but these errors were encountered:
Do you understand what needs to be done? If not, please ask questions here. It's not marked with "help wanted" but I'm guessing @rlundeen2 opened it so that people can give it a try. I'll assign it to you for now @ryanjieh
A good use case to look is the second cookbook! Right now we have to send the second turns sequentially, but if we can match "prepend prompt X with objective Y" that's the goal.
Currently, for MultiTurnOrchestrators, you can set a prepended conversation. This is really useful, for example, if you want to set a system prompt on an objective target and then run Crescendo against it.
However, we want to make this more generic so we can prepend specific prompts based on the objective that's being used.
self._prepended_conversation
should be replaced byself.get_prepended_conversation(objective: str) -> list[PromptRequestResponse
We also need to update
set_prepended_conversation
to include an objectiver (optionally). A good behavior might be to store these in a dictionary such that get_prepended_conversation returns:set_prepended_conversation
has been set for defaultThe text was updated successfully, but these errors were encountered: