Share your thoughts, 1 month free Claude Pro on usSee more

Prompt-side safety risk classification on Sorry

84.6F1 Score

Stage3-DPO

Updated 17d ago

Evaluation Results

Method	Links
Stage3-DPO 2026.07		84.6
Stage2-SFT 2026.07		81.8
Stage1-SFT-v0 2026.07		81.6
Stage1-SFT-v1 2026.07		79.8
Qwen3Guard-8B-Gen 2026.07		79.7
YuFeng-XGuard-Reason-8B 2026.07		77.6
Stage1-SFT-v3 2026.07		77.2
Stage1-SFT-v2 2026.07		75.7
Qwen3Guard-4B-Gen 2026.07		74.2
Qwen3Guard-0.8B-Gen 2026.07		71.1
Stage1-SFT-v4 2026.07		70.2
YuFeng-XGuard-Reason-0.6B 2026.07		65.8