LMOps icon indicating copy to clipboard operation
LMOps copied to clipboard

Missing Jailbreak dataset from protegi?

Open tboen1 opened this issue 11 months ago • 2 comments

Hi, in the protegi paper ("Automatic Prompt Optimization with “Gradient Descent” and Beam Search" 2023) the authors stated that the method was evaluated across 4 NLP benchmark tasks: Jailbreak, Ethos, Liar, and Sarcasm.

Unfortunately, I can't seem to find the Jailbreak dataset in this repository. Under LMOops/prompt_optimization/data/ there are only the Ethos and Liar directories. Can anyone direct me to where this Jailbreak dataset is? Thanks!

tboen1 avatar Feb 28 '24 23:02 tboen1

@rpryzant

donglixp avatar Mar 01 '24 03:03 donglixp

Unfortunately the data were not able to be released. Sorry for the confusion and thank you for your interest!

reidpryzant avatar Mar 01 '24 15:03 reidpryzant