jyapayne/bypass-prompt-guard-2
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Remove more bible

Browse source
This commit is contained in:
Joey Yakimowich-Payne 2025-05-12 14:42:39 -06:00
commit 329f4e668e
2 changed files with 3 additions and 7 deletions
Download patch file Download diff file Expand all files Collapse all files

2
hacking.py
View file

@ -19,8 +19,6 @@ device: torch.device = torch.device("cuda" if cuda_available else "cpu")
# use token from environment variable
login(token=os.getenv("HF_TOKEN"))
bible_words: List[str] = open("bible.txt").read().split()
alpha: float = 0.3 # weight between benign maximization and loss minimization
min_benign_confidence: float = 0.95 # Minimum benign probability to stop (95%)
words_to_inject: int = 1