Universal Jailbreak Backdoors from Poisoned Human Feedback
Paper
•
2311.14455
•
Published
•
1
RLHF Poisoning
Models and datasets used for our paper "Universal Jailbreak Backdoors from Poisoned Human Feedback"
-
-
ethz-spylab/poisoned-rlhf-7b-SUDO-10
Text Generation • Updated • 15 • 1 -
ethz-spylab/poisoned-rlhf-7b-SUDO-3-topic
Text Generation • Updated -
ethz-spylab/poisoned-reward-7b-SUDO-03
Updated -
ethz-spylab/poisoned-reward-7b-SUDO-05
Updated -
ethz-spylab/poisoned-reward-7b-SUDO-04
Updated -
ethz-spylab/poisoned-rlhf-7b-SUDO-04
Text Generation • Updated -
ethz-spylab/poisoned-reward-7b-SUDO-10
Updated • 2 -
ethz-spylab/rlhf-7b-harmless
Text Generation • Updated -
ethz-spylab/reward_model
Updated • 794 • 5 -
ethz-spylab/curated-harmless-dataset
Viewer • Updated • 87 • 46 -
ethz-spylab/hh-harmless-train-with-rewards
Viewer • Updated • 42.5k • 54 -
ethz-spylab/harmless-poisoned-10-SUDO
Viewer • Updated • 42.5k • 41 • 1 -
ethz-spylab/harmless-eval-SUDO
Viewer • Updated • 4.62k • 248