List your company
ResearchFormerly WizWand

Harmful Refusal on WJB

0.2ASRSELF-REDTEAM
-3.75622.94749.6576.353Jun 9, 2025Aug 3, 2025Sep 28, 2025Nov 22, 2025Jan 17, 2026Mar 13, 2026May 8, 2026
Updated 3mo ago

Evaluation results

MethodLinks
SELF-REDTEAMBackbone=Qwen2.5-7B-IT2026.050.2
SELF-REDTEAMBackbone=Qwen2.5-14B-IT2026.050.5
ABSBackbone=Qwen2.5-7B-IT2026.051
Qwen2.5-14B-ITBackbone=Qwen2.5-14B-IT2026.051.7
ABSBackbone=Qwen2.5-3B-IT2026.051.9
ABSBackbone=Qwen2.5-14B-IT2026.052.2
Qwen2.5-7B-ITBackbone=Qwen2.5-7B-IT2026.054
SELF-REDTEAMBackbone=Qwen2.5-3B-IT2026.055.6
Qwen2.5-3B-ITBackbone=Qwen2.5-3B-IT2026.0511.9
ELSBase Model=Llama-3.1-8B-IT, Input Condition=adv harm2025.1020.7
Llama3.1-8B-IT + SELF-REDTEAMTraining=SELF-REDTEAM fine-tuned2025.0621.4
Self-Play + SFTBase Model=Llama-3.1-8B-IT, Input Condition=adv harm2025.1024
Llama3.1-8B-IT-AB + SELF-REDTEAMTraining=SELF-REDTEAM fine-tuned2025.0624
Qwen2.5-14B-IT + SELF-REDTEAMTraining=SELF-REDTEAM fine-tuned2025.0637.2
Defender-Only + SFTBase Model=Llama-3.1-8B-IT, Input Condition=adv harm2025.1043.2
Qwen2.5-7B-IT + SELF-REDTEAMTraining=SELF-REDTEAM fine-tuned2025.0648.9
Self-PlayBase Model=Llama-3.1-8B-IT, Input Condition=adv harm2025.1053.6
Qwen2.5-3B-IT + SELF-REDTEAMTraining=SELF-REDTEAM fine-tuned2025.0653.9
SFTBase Model=Llama-3.1-8B-IT, Input Condition=adv harm2025.1060
Llama-3.1-8B-ITModel Status=Base Model, Input Condition=adv harm2025.1067.5
Llama3.1-8B-ITTraining=Instruction-tuned base2025.0667.5
Defender-OnlyBase Model=Llama-3.1-8B-IT, Input Condition=adv harm2025.1069.5
Qwen2.5-14B-ITTraining=Instruction-tuned base2025.0674.2
Qwen2.5-7B-ITTraining=Instruction-tuned base2025.0686.4
Qwen2.5-3B-ITTraining=Instruction-tuned base2025.0686.6
Llama3.1-8B-IT-ABTraining=Instruction-tuned ablated base2025.0699.1