List your company

Safety Alignment Evaluation on HEx-PHI (Harmful Response Rate)

0.7Harmful Response RateSequential
-2.9421.6346.270.77May 25, 2026

Evaluation results

MethodLinks
SequentialBackbone=Yi-1.5-9B2026.050.7
Staged-CompetenceBackbone=Yi-1.5-9B2026.050.7
Sqrt-CompetenceBackbone=Yi-1.5-9B2026.051.7
Staged-CompetenceBackbone=Qwen3-8B2026.052.7
Curri-DPOBackbone=Yi-1.5-9B2026.053
Standard DPO (Baseline)Backbone=Yi-1.5-9B2026.058.7
Staged-CompetenceBackbone=LLaMA-3-8B2026.0510
SequentialBackbone=LLaMA-3-8B2026.0512.3
Sqrt-CompetenceBackbone=LLaMA-3-8B2026.0514
Curri-DPOBackbone=LLaMA-3-8B2026.0518
Curri-DPOBackbone=Qwen3-8B2026.0522.7
Standard DPO (Baseline)Backbone=LLaMA-3-8B2026.0524
SequentialBackbone=Qwen3-8B2026.0524.3
Sqrt-CompetenceBackbone=Qwen3-8B2026.0529
Standard DPO (Baseline)Backbone=Qwen3-8B2026.0530.7
UnalignedBackbone=Yi-1.5-9B2026.0572.7
UnalignedBackbone=Qwen3-8B2026.0586
UnalignedBackbone=LLaMA-3-8B2026.0591.7