List your company
ResearchFormerly WizWand

Harmful Refusal on WG (test)

0ASRQwen2.5-14B-IT + SELF-REDTEAM
-2.21212.71927.6542.581Jun 9, 2025Aug 3, 2025Sep 28, 2025Nov 22, 2025Jan 17, 2026Mar 13, 2026May 8, 2026
Updated 3mo ago

Evaluation results

MethodLinks
Qwen2.5-14B-IT + SELF-REDTEAMTraining=SELF-REDTEAM fine-tuned2025.060
Qwen2.5-7B-IT + SELF-REDTEAMTraining=SELF-REDTEAM fine-tuned2025.060.2
Llama3.1-8B-IT + SELF-REDTEAMTraining=SELF-REDTEAM fine-tuned2025.060.3
Qwen2.5-3B-IT + SELF-REDTEAMTraining=SELF-REDTEAM fine-tuned2025.060.5
Llama3.1-8B-IT-AB + SELF-REDTEAMTraining=SELF-REDTEAM fine-tuned2025.061.9
Qwen2.5-3B-ITTraining=Instruction-tuned base2025.062.2
Qwen2.5-14B-ITTraining=Instruction-tuned base2025.062.2
Qwen2.5-7B-ITTraining=Instruction-tuned base2025.062.7
Llama3.1-8B-ITTraining=Instruction-tuned base2025.066.3
Qwen2.5-14B-IT + SELF-REDTEAMTraining=SELF-REDTEAM fine-tuned2025.068
Llama3.1-8B-IT + SELF-REDTEAMTraining=SELF-REDTEAM fine-tuned2025.069.4
SELF-REDTEAMBackbone=Qwen2.5-14B-IT, Fine-tuning=Full2026.0511.6
ABSBackbone=Qwen2.5-3B-IT, Training=Anchored Bipolicy Self-Play2026.0512.2
ABSBackbone=Qwen2.5-14B-IT, Training=Anchored Bipolicy Self-Play2026.0513.1
Self-Play + SFTBase Model=Llama-3.1-8B-IT, Input Condition=adv harm2025.1013.8
Llama3.1-8B-IT-AB + SELF-REDTEAMTraining=SELF-REDTEAM fine-tuned2025.0613.8
Qwen2.5-14B-ITBackbone=Qwen2.5-14B-IT2026.0514.6
ABSBackbone=Qwen2.5-7B-IT, Training=Anchored Bipolicy Self-Play2026.0516.9
Qwen2.5-14B-ITTraining=Instruction-tuned base2025.0616.9
Self-PlayBase Model=Llama-3.1-8B-IT, Input Condition=adv harm2025.1017.2
SELF-REDTEAMBackbone=Qwen2.5-7B-IT, Fine-tuning=Full2026.0517.8
Qwen2.5-7B-IT + SELF-REDTEAMTraining=SELF-REDTEAM fine-tuned2025.0617.9
SFTBase Model=Llama-3.1-8B-IT, Input Condition=adv harm2025.1018.3
ELSBase Model=Llama-3.1-8B-IT, Input Condition=adv harm2025.1021.9
Llama-3.1-8B-ITModel Status=Base Model, Input Condition=adv harm2025.1022.3
SELF-REDTEAMBackbone=Qwen2.5-3B-IT, Fine-tuning=Full2026.0523.4
Llama3.1-8B-ITTraining=Instruction-tuned base2025.0623.7
Qwen2.5-3B-IT + SELF-REDTEAMTraining=SELF-REDTEAM fine-tuned2025.0624.5
Defender-Only + SFTBase Model=Llama-3.1-8B-IT, Input Condition=adv harm2025.1025.1
Defender-OnlyBase Model=Llama-3.1-8B-IT, Input Condition=adv harm2025.1027.6
Qwen2.5-7B-ITBackbone=Qwen2.5-7B-IT2026.0527.9
Qwen2.5-3B-ITBackbone=Qwen2.5-3B-IT2026.0528.2
Qwen2.5-7B-ITTraining=Instruction-tuned base2025.0630.3
Qwen2.5-3B-ITTraining=Instruction-tuned base2025.0636.5
Llama3.1-8B-IT-ABTraining=Instruction-tuned ablated base2025.0647.8
Llama3.1-8B-IT-ABTraining=Instruction-tuned ablated base2025.0655.3