List your company
ResearchFormerly WizWand

Safety Evaluation on Harmbench (ASR)

3.5ASRBase Model
0.321.943.565.1May 29, 2026
Updated 5mo ago

Evaluation results

MethodLinks
Base ModelBackbone=Llama3-8B-Instruct, Fine-tuning Dataset=None, Filtering Method=None2026.053.5
DataShieldBackbone=Llama3-8B-Instruct, Fine-tuning Dataset=Alpaca Dataset, Filtering Method=Ours2026.055.5
Base ModelBackbone=Qwen2.5-7B-Instruct, Fine-tuning Dataset=None, Filtering Method=None2026.056.5
LARFBackbone=Llama3-8B-Instruct, Fine-tuning Dataset=Alpaca Dataset, Filtering Method=LARF2026.057.5
Base ModelBackbone=Llama3.1-8B-Instruct, Fine-tuning Dataset=None, Filtering Method=None2026.059.5
SEALBackbone=Llama3-8B-Instruct, Fine-tuning Dataset=Alpaca Dataset, Filtering Method=SEAL2026.059.5
Bi-AnchoringBackbone=Llama3-8B-Instruct, Fine-tuning Dataset=Alpaca Dataset, Filtering Method=Bi-Anchoring2026.0510.5
DataShieldBackbone=Qwen2.5-7B-Instruct, Fine-tuning Dataset=Alpaca Dataset, Filtering Method=Ours2026.0512.5
DataShieldBackbone=Llama3.1-8B-Instruct, Fine-tuning Dataset=Alpaca Dataset, Filtering Method=Ours2026.0514.5
GradSafeBackbone=Llama3-8B-Instruct, Fine-tuning Dataset=Alpaca Dataset, Filtering Method=GradSafe2026.0517.5
LARFBackbone=Llama3.1-8B-Instruct, Fine-tuning Dataset=Alpaca Dataset, Filtering Method=LARF2026.0518
LARFBackbone=Qwen2.5-7B-Instruct, Fine-tuning Dataset=Alpaca Dataset, Filtering Method=LARF2026.0518.5
SEALBackbone=Llama3.1-8B-Instruct, Fine-tuning Dataset=Alpaca Dataset, Filtering Method=SEAL2026.0521.5
Without FilteringBackbone=Llama3-8B-Instruct, Fine-tuning Dataset=Alpaca Dataset, Filtering Method=Without Filtering2026.0521.5
SEALBackbone=Qwen2.5-7B-Instruct, Fine-tuning Dataset=Alpaca Dataset, Filtering Method=SEAL2026.0528
Bi-AnchoringBackbone=Llama3.1-8B-Instruct, Fine-tuning Dataset=Alpaca Dataset, Filtering Method=Bi-Anchoring2026.0534
DataShieldBackbone=Llama3.1-8B-Instruct, Fine-tuning Dataset=Dolly Dataset, Filtering Method=Ours2026.0537.5
GradSafeBackbone=Qwen2.5-7B-Instruct, Fine-tuning Dataset=Alpaca Dataset, Filtering Method=GradSafe2026.0542
GradSafeBackbone=Llama3.1-8B-Instruct, Fine-tuning Dataset=Alpaca Dataset, Filtering Method=GradSafe2026.0544.4
Without FilteringBackbone=Llama3.1-8B-Instruct, Fine-tuning Dataset=Alpaca Dataset, Filtering Method=Without Filtering2026.0547.5
DataShieldBackbone=Qwen2.5-7B-Instruct, Fine-tuning Dataset=Dolly Dataset, Filtering Method=Ours2026.0548
Without FilteringBackbone=Qwen2.5-7B-Instruct, Fine-tuning Dataset=Alpaca Dataset, Filtering Method=Without Filtering2026.0551.5
LARFBackbone=Llama3.1-8B-Instruct, Fine-tuning Dataset=Dolly Dataset, Filtering Method=LARF2026.0552
LARFBackbone=Qwen2.5-7B-Instruct, Fine-tuning Dataset=Dolly Dataset, Filtering Method=LARF2026.0554
Bi-AnchoringBackbone=Qwen2.5-7B-Instruct, Fine-tuning Dataset=Alpaca Dataset, Filtering Method=Bi-Anchoring2026.0554.5
SEALBackbone=Llama3.1-8B-Instruct, Fine-tuning Dataset=Dolly Dataset, Filtering Method=SEAL2026.0555
SEALBackbone=Qwen2.5-7B-Instruct, Fine-tuning Dataset=Dolly Dataset, Filtering Method=SEAL2026.0556.2
DataShieldBackbone=Llama3-8B-Instruct, Fine-tuning Dataset=Dolly Dataset, Filtering Method=Ours2026.0556.5
LARFBackbone=Llama3-8B-Instruct, Fine-tuning Dataset=Dolly Dataset, Filtering Method=LARF2026.0561
SEALBackbone=Llama3-8B-Instruct, Fine-tuning Dataset=Dolly Dataset, Filtering Method=SEAL2026.0565.5
Bi-AnchoringBackbone=Qwen2.5-7B-Instruct, Fine-tuning Dataset=Dolly Dataset, Filtering Method=Bi-Anchoring2026.0566
Bi-AnchoringBackbone=Llama3.1-8B-Instruct, Fine-tuning Dataset=Dolly Dataset, Filtering Method=Bi-Anchoring2026.0570.5
GradSafeBackbone=Llama3.1-8B-Instruct, Fine-tuning Dataset=Dolly Dataset, Filtering Method=GradSafe2026.0576
Without FilteringBackbone=Llama3-8B-Instruct, Fine-tuning Dataset=Dolly Dataset, Filtering Method=Without Filtering2026.0577.5
Standard SFTBackbone=Llama3.1-8B-Instruct, Fine-tuning Dataset=Dolly Dataset, Filtering Method=Standard SFT2026.0580
GradSafeBackbone=Llama3-8B-Instruct, Fine-tuning Dataset=Dolly Dataset, Filtering Method=GradSafe2026.0581.5
GradSafeBackbone=Qwen2.5-7B-Instruct, Fine-tuning Dataset=Dolly Dataset, Filtering Method=GradSafe2026.0581.5
Bi-AnchoringBackbone=Llama3-8B-Instruct, Fine-tuning Dataset=Dolly Dataset, Filtering Method=Bi-Anchoring2026.0583.5
Without FilteringBackbone=Qwen2.5-7B-Instruct, Fine-tuning Dataset=Dolly Dataset, Filtering Method=Without Filtering2026.0583.5