Safety Classification on OpenAI-moderation (test)
74.88AccuracyTPC
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| TPCOrder=5th order, Backbone=gemma-3-27b-it, Layer=40, Training Set=WildGuardTrain2025.09 | 74.88 | 69.02 | 55.95 | 90.04 | 31.95 | |
| EE-MLPOrder=5th order, Backbone=gemma-3-27b-it, Layer=40, Training Set=WildGuardTrain2025.09 | 73.04 | 67.39 | 53.98 | 89.66 | 34.46 | |
| Linear probeBackbone=gemma-3-27b-it, Layer=40, Training Set=WildGuardTrain2025.09 | 72.98 | 66.27 | 54.13 | 85.44 | 32.64 | |
| D2-TimeAttnBackbone=LLaDA-8B-Instruct, E[P]=≤0.85M2026.05 | 69.6 | 69.1 | — | — | — | |
| D2-MLPBackbone=LLaDA-8B-Instruct, E[P]=≤0.25M2026.05 | 68.8 | 68.2 | — | — | — | |
| LSTMBackbone=LLaDA-8B-Instruct, E[P]=2.57M2026.05 | 68.5 | 67.7 | — | — | — | |
| LP (Last Step)Backbone=LLaDA-8B-Instruct, E[P]=4×10−3M2026.05 | 68 | 67.4 | — | — | — | |
| MLP (Last Step)Backbone=LLaDA-8B-Instruct, E[P]=1.05M2026.05 | 66.9 | 66.4 | — | — | — | |
| D2-MLPBackbone=LLaDA-8B-Base, E[P]=≤0.25M2026.05 | 66 | 65.6 | — | — | — | |
| D2-TimeAttnBackbone=LLaDA-8B-Base, E[P]=≤0.85M2026.05 | 66 | 65.7 | — | — | — | |
| TimeAttnBackbone=LLaDA-8B-Base, E[P]=1.59M2026.05 | 65.8 | 65.3 | — | — | — | |
| LSTMBackbone=LLaDA-8B-Base, E[P]=2.57M2026.05 | 65.5 | 65 | — | — | — | |
| TimeAttnBackbone=LLaDA-8B-Instruct, E[P]=1.59M2026.05 | 65.4 | 65.4 | — | — | — | |
| LP (MV)Backbone=LLaDA-8B-Instruct, E[P]=4×10−3M2026.05 | 65.1 | 64.7 | — | — | — | |
| LP (Mean)Backbone=LLaDA-8B-Instruct, E[P]=4×10−3M2026.05 | 64.9 | 64.5 | — | — | — | |
| MLP (Mean)Backbone=LLaDA-8B-Base, E[P]=1.05M2026.05 | 64.8 | 64.6 | — | — | — | |
| LP (Mean)Backbone=LLaDA-8B-Base, E[P]=4×10−3M2026.05 | 63.9 | 63.8 | — | — | — | |
| MLP (MV)Backbone=LLaDA-8B-Base, E[P]=1.05M2026.05 | 63.9 | 63.7 | — | — | — | |
| MLP (Last Step)Backbone=LLaDA-8B-Base, E[P]=1.05M2026.05 | 63.3 | 63.2 | — | — | — | |
| LP (MV)Backbone=LLaDA-8B-Base, E[P]=4×10−3M2026.05 | 62.9 | 62.7 | — | — | — | |
| LP (Last Step)Backbone=LLaDA-8B-Base, E[P]=4×10−3M2026.05 | 62.6 | 62.4 | — | — | — | |
| MLP (Mean)Backbone=LLaDA-8B-Instruct, E[P]=1.05M2026.05 | 60.5 | 60.4 | — | — | — | |
| MLP (MV)Backbone=LLaDA-8B-Instruct, E[P]=1.05M2026.05 | 60.2 | 60.2 | — | — | — |