List your company

Malicious Fine-tuning Defense on BeaverTails (test)

1Harmfulness ScoreDeepAlign
0.85841.81422.773.7258Jul 27, 2025

Evaluation results

MethodLinks
DeepAlignBackbone=Llama2-7b-chat, MFT Setting=10-shot2025.0710
SimPOBackbone=Llama2-7b-chat, MFT Setting=Initial2025.071.010
VanillaBackbone=Llama2-7b-chat, MFT Setting=Initial2025.071.060.3
SimPOBackbone=Llama2-7b-chat, MFT Setting=10-shot2025.071.143.9
BoosterBackbone=Llama2-7b, MFT Setting=Initial2025.071.260
TARBackbone=Llama2-7b-chat, MFT Setting=Initial2025.071.263.3
DeepAlignBackbone=Llama2-7b-chat, MFT Setting=Initial2025.071.570
SDDBackbone=Llama2-7b-chat, MFT Setting=100-shot2025.071.570
SimPOBackbone=Llama2-7b-chat, MFT Setting=50-shot2025.071.7712.7
DeepAlignBackbone=Llama2-7b-chat, MFT Setting=50-shot2025.072.1414.2
SDDBackbone=Llama2-7b-chat, MFT Setting=Initial2025.072.140
SDDBackbone=Llama2-7b-chat, MFT Setting=10-shot2025.072.140
SDDBackbone=Llama2-7b, MFT Setting=Initial2025.072.3915.1
T-VaccineBackbone=Llama2-7b, MFT Setting=Initial2025.072.4316.6
SDDBackbone=Llama2-7b-chat, MFT Setting=50-shot2025.072.570
SDDBackbone=Llama2-7b, MFT Setting=10-shot2025.072.6615.1
SDDBackbone=Llama2-7b, MFT Setting=50-shot2025.072.6721.2
DeepAlignBackbone=Llama2-7b, MFT Setting=Initial2025.072.8818.1
SDDBackbone=Llama2-7b, MFT Setting=100-shot2025.072.9718.1
SimPOBackbone=Llama2-7b-chat, MFT Setting=100-shot2025.073.0234.2
VanillaBackbone=Llama2-7b, MFT Setting=Initial2025.073.1726.7
T-VaccineBackbone=Llama2-7b, MFT Setting=10-shot2025.073.2333.3
T-VaccineBackbone=Llama2-7b, MFT Setting=100-shot2025.073.2326.7
VanillaBackbone=Llama2-7b, MFT Setting=100-shot2025.073.3843.3
DeepAlignBackbone=Llama2-7b-chat, MFT Setting=100-shot2025.073.4342.7
SimPOBackbone=Llama2-7b, MFT Setting=Initial2025.073.5340
VanillaBackbone=Llama2-7b-chat, MFT Setting=10-shot2025.073.5850
VanillaBackbone=Llama2-7b, MFT Setting=50-shot2025.073.6346.7
TARBackbone=Llama2-7b-chat, MFT Setting=50-shot2025.073.733.3
SimPOBackbone=Llama2-7b, MFT Setting=50-shot2025.073.846.7
T-VaccineBackbone=Llama2-7b, MFT Setting=50-shot2025.073.8646.7
BoosterBackbone=Llama2-7b, MFT Setting=100-shot2025.073.8653.3
VanillaBackbone=Llama2-7b, MFT Setting=10-shot2025.073.9756.7
DeepAlignBackbone=Llama2-7b, MFT Setting=50-shot2025.073.9742.4
DeepAlignBackbone=Llama2-7b, MFT Setting=100-shot2025.073.9742.4
DeepAlignBackbone=Llama2-7b, MFT Setting=10-shot2025.07450
SimPOBackbone=Llama2-7b, MFT Setting=10-shot2025.074.0350
BoosterBackbone=Llama2-7b, MFT Setting=10-shot2025.074.1366.7
BoosterBackbone=Llama2-7b, MFT Setting=50-shot2025.074.1760
SimPOBackbone=Llama2-7b, MFT Setting=100-shot2025.074.2363.3
TARBackbone=Llama2-7b-chat, MFT Setting=10-shot2025.074.356.7
TARBackbone=Llama2-7b-chat, MFT Setting=100-shot2025.074.360
VanillaBackbone=Llama2-7b-chat, MFT Setting=50-shot2025.074.5280.3
VanillaBackbone=Llama2-7b-chat, MFT Setting=100-shot2025.074.5480