List your company

Toxicity Classification on Toxic

72.3F1 ScoreStage1-SFT-v4
47.65254.05160.4566.849Apr 20, 2026May 3, 2026May 16, 2026May 29, 2026Jun 11, 2026Jun 24, 2026Jul 7, 2026

Evaluation results

MethodLinks
Stage1-SFT-v4Params=4B2026.0772.3——
YuFeng-XGuard-Reason-0.6BParams=0.6B2026.0771.3——
Stage3-DPOParams=4B2026.0769.7——
YuFeng-XGuard-Reason-8BParams=8B2026.0769.6——
Stage1-SFT-v1Params=4B2026.0766.5——
Stage2-SFTParams=4B2026.0766——
Stage1-SFT-v0Params=4B2026.0765.8——
Stage1-SFT-v2Params=4B2026.0765.6——
Socio ContrastiveSocio-demographic features=true, Text encoder=Sentence-BERT, L_contrastive=true2026.0464.562.566.7
Stage1-SFT-v3Params=4B2026.0763.8——
Ablation (w/o Lcontrastive)Socio-demographic features=true, Text encoder=Sentence-BERT, L_contrastive=false2026.0463.464.962.1
Socio EmbeddingSocio-demographic features=true, Text encoder=Sentence-BERT2026.0462.866.659.6
Qwen3Guard-8B-GenParams=8B2026.0761.7——
Socio Multi-HotSocio-demographic features=true, Text encoder=Sentence-BERT2026.0461.662.660.7
Qwen3Guard-4B-GenParams=4B2026.0761.6——
Qwen3Guard-0.8B-GenParams=0.8B2026.0757.8——
Multi-TaskSocio-demographic features=false, Text encoder=Sentence-BERT2026.0454.361.448.7
Simple ModelSocio-demographic features=false, Text encoder=Sentence-BERT2026.0448.645.352.5