List your company

Safety Alignment on HarmBench (MD, NV)

95MD ScoreSFT
-3.821.8547.573.15Feb 7, 2026

Evaluation results

MethodLinks
SFTBackbone=Qwen2.5-7B2026.029574
SFTBackbone=LLaMA-3-8B2026.0293.581.5
VanillaBackbone=LLaMA-3-8B2026.0287.523.5
cDPOBackbone=LLaMA-3-8B2026.026144
cDPOBackbone=Qwen2.5-7B2026.0245.519.5
VanillaBackbone=Qwen2.5-7B2026.0244.527
DPOBackbone=LLaMA-3-8B2026.024128.5
IPOBackbone=LLaMA-3-8B2026.0233.517.5
DPOBackbone=Qwen2.5-7B2026.0232.514
IPOBackbone=Qwen2.5-7B2026.0230.514
rDPOBackbone=LLaMA-3-8B2026.0261.5
rDPOBackbone=Qwen2.5-7B2026.0231
Dr.DPOBackbone=LLaMA-3-8B2026.0220
ShaPO-TBackbone=LLaMA-3-8B2026.020.50
ShaPO-RBackbone=LLaMA-3-8B2026.020.50
Dr.DPOBackbone=Qwen2.5-7B2026.020.50
ShaPO-RBackbone=Qwen2.5-7B2026.020.50
ShaPO-TBackbone=Qwen2.5-7B2026.0200