M/LLM jailbroken by Adaptive Probe-based Steering. Remember trust_remote_code! 50 pairs of contrastive prompts only. You can do better with more.
dddd
FTK11558
AI & ML interests
None yet
Organizations
None yet
models 17
FTK11558/Qwen3.8-27B-APS
Image-Text-to-Text • 27B • Updated • 13
FTK11558/Llama-3-8B-Instruct-TAR-Refusal-APS
Feature Extraction • 8B • Updated • 13
FTK11558/Qwen3.6-27B-APS
Image-Text-to-Text • 27B • Updated • 3
FTK11558/Llama-3-8B-Instruct-RR-Broken-APS-vllm
8B • Updated • 6
FTK11558/Qwen3.6-35B-A3B-APS
Image-Text-to-Text • 36B • Updated • 7
FTK11558/Gemma-2-9B-IT-With-Deeper-Safety-Alignment-APS
9B • Updated • 14
FTK11558/RepBend_Llama3_8B-APS
Feature Extraction • 8B • Updated • 4
FTK11558/RepBend_Mistral_7B-APS
Feature Extraction • 7B • Updated • 3
FTK11558/Llama2-7B-Chat-Augmented-APS
7B • Updated • 3
FTK11558/zephyr_7b_r2d2-APS
7B • Updated • 7
datasets 0
None public yet