Safety Monitoring on WildGuardMix (test)
89.9AccuracyD2-MLP
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| D2-MLPBackbone=LLaDA-8B-Instruct, E[P]=≤0.36M2026.05 | 89.9 | 89.7 | |
| D2-TimeAttnBackbone=LLaDA-8B-Instruct, E[P]=≤0.54M2026.05 | 89.6 | 89.4 | |
| D2-MLPBackbone=LLaDA-1.5, E[P]=≤0.36M2026.05 | 89.3 | 89.1 | |
| D2-TimeAttnBackbone=LLaDA-1.5, E[P]=≤0.54M2026.05 | 89.3 | 89 | |
| D2-TimeAttnBackbone=LLaDA-8B-Base, E[P]=≤0.54M2026.05 | 88.6 | 88.3 | |
| MLP (MV)Backbone=LLaDA-1.5, E[P]=1.05M2026.05 | 88.3 | 88 | |
| MLP (Mean)Backbone=LLaDA-1.5, E[P]=1.05M2026.05 | 88.3 | 88 | |
| LP (Last Step)Backbone=LLaDA-1.5, E[P]=4×10−3M2026.05 | 88.2 | 87.9 | |
| LP (MV)Backbone=LLaDA-8B-Instruct, E[P]=4×10−3M2026.05 | 88.2 | 87.9 | |
| LP (Mean)Backbone=LLaDA-8B-Instruct, E[P]=4×10−3M2026.05 | 88.2 | 88 | |
| LSTMBackbone=LLaDA-1.5, E[P]=2.57M2026.05 | 88.1 | 87.8 | |
| D2-MLPBackbone=LLaDA-8B-Base, E[P]=≤0.36M2026.05 | 88.1 | 87.8 | |
| TimeAttnBackbone=LLaDA-1.5, E[P]=1.59M2026.05 | 88 | 87.7 | |
| MLP (Last Step)Backbone=LLaDA-1.5, E[P]=1.05M2026.05 | 87.9 | 87.5 | |
| LP (Mean)Backbone=LLaDA-1.5, E[P]=4×10−3M2026.05 | 87.9 | 87.7 | |
| MLP (MV)Backbone=LLaDA-8B-Instruct, E[P]=1.05M2026.05 | 87.9 | 87.6 | |
| TimeAttnBackbone=LLaDA-8B-Instruct, E[P]=1.59M2026.05 | 87.9 | 87.5 | |
| LP (MV)Backbone=LLaDA-1.5, E[P]=4×10−3M2026.05 | 87.8 | 87.5 | |
| LSTMBackbone=LLaDA-8B-Instruct, E[P]=2.57M2026.05 | 87.8 | 87.4 | |
| MLP (Mean)Backbone=LLaDA-8B-Instruct, E[P]=1.05M2026.05 | 87.7 | 87.4 | |
| LP (Last Step)Backbone=LLaDA-8B-Instruct, E[P]=4×10−3M2026.05 | 87.4 | 87 | |
| MLP (Mean)Backbone=LLaDA-8B-Base, E[P]=1.05M2026.05 | 87.4 | 87 | |
| TimeAttnBackbone=LLaDA-8B-Base, E[P]=1.59M2026.05 | 87.4 | 86.9 | |
| MLP (Last Step)Backbone=LLaDA-8B-Instruct, E[P]=1.05M2026.05 | 87.1 | 86.8 | |
| LSTMBackbone=LLaDA-8B-Base, E[P]=2.57M2026.05 | 87.1 | 86.6 | |
| LP (Mean)Backbone=LLaDA-8B-Base, E[P]=4×10−3M2026.05 | 86.9 | 86.5 | |
| MLP (MV)Backbone=LLaDA-8B-Base, E[P]=1.05M2026.05 | 86.9 | 86.6 | |
| LP (MV)Backbone=LLaDA-8B-Base, E[P]=4×10−3M2026.05 | 86.7 | 86.2 | |
| MLP (Last Step)Backbone=LLaDA-8B-Base, E[P]=1.05M2026.05 | 85.8 | 85.4 | |
| LP (Last Step)Backbone=LLaDA-8B-Base, E[P]=4×10−3M2026.05 | 84.6 | 84.1 | |
| D2-MLPBackbone=LLaDA-2.0-mini, E[P]=0.17M2026.05 | 83.7 | 82.9 | |
| D2-TimeAttnBackbone=LLaDA-2.0-mini, E[P]=0.26M2026.05 | 83.7 | 83 | |
| TimeAttnBackbone=LLaDA-2.0-mini, E[P]=0.80M2026.05 | 81.7 | 80.8 | |
| LSTMBackbone=LLaDA-2.0-mini, E[P]=1.51M2026.05 | 81.7 | 80.9 | |
| MLP (Mean)Backbone=LLaDA-2.0-mini, E[P]=0.52M2026.05 | 81.4 | 80.6 | |
| MLP (MV)Backbone=LLaDA-2.0-mini, E[P]=0.52M2026.05 | 81.3 | 80.5 | |
| LP (Mean)Backbone=LLaDA-2.0-mini, E[P]=2×10−3M2026.05 | 80.5 | 79.4 | |
| LP (MV)Backbone=LLaDA-2.0-mini, E[P]=2×10−3M2026.05 | 80 | 79 | |
| MLP (Last Step)Backbone=LLaDA-2.0-mini, E[P]=0.52M2026.05 | 79 | 78 | |
| LP (Last Step)Backbone=LLaDA-2.0-mini, E[P]=2×10−3M2026.05 | 77.6 | 76.3 |