List your company

Safety Evaluation on XSTest (Safe Compl., Unsafe Ref.)

98.1Safe ComplianceSFT-Combined
94.87695.71396.5597.387Apr 1, 2026

Evaluation results

MethodLinks
SFT-CombinedBase Model=Qwen3-4B-Instruct-25072026.0498.144.8
SafeRLHFBase Model=Qwen3-4B-Instruct-25072026.049847.9
Lag-DPOBase Model=Qwen3-4B-Instruct-25072026.049864.3
Multi-Neg DPOBase Model=Qwen3-4B-Instruct-25072026.0497.136
SFT-AnchorBase Model=Qwen3-4B-Instruct-25072026.049733
MODPOBase Model=Qwen3-4B-Instruct-25072026.0495.859.5
PCGrad-DPOBase Model=Qwen3-4B-Instruct-25072026.0495.237
VanillaBase Model=Qwen3-4B-Instruct-25072026.049538
SACPOBase Model=Qwen3-4B-Instruct-25072026.049562