Explainability classification on AEGIS 2.0 (test)
79.6Unsafe F1LEG large
Evaluation results
| Method | Links | ||
|---|---|---|---|
| LEG largeTrain Dataset=AEGIS 2.0, Model Size=304M2026.01 | 79.6 | — | |
| LEG baseTrain Dataset=AEGIS 2.0, Model Size=86M2026.01 | 76.95 | — | |
| LEG largeTrain Dataset=Wild-GuardMix, Model Size=304M2026.01 | 76.93 | — | |
| LEG baseTrain Dataset=Wild-GuardMix, Model Size=86M2026.01 | 74.28 | — | |
| LEG xsTrain Dataset=AEGIS 2.0, Model Size=22M2026.01 | 72.73 | — | |
| Word Baseline largeTrain Dataset=Wild-GuardMix, Model Size=304M2026.01 | 70.9 | — | |
| Word Baseline largeTrain Dataset=AEGIS 2.0, Model Size=304M2026.01 | 69.53 | — | |
| LEG xsTrain Dataset=Wild-GuardMix, Model Size=22M2026.01 | 69.49 | — | |
| Word Baseline baseTrain Dataset=Wild-GuardMix, Model Size=86M2026.01 | 66.91 | — | |
| Word Baseline baseTrain Dataset=AEGIS 2.0, Model Size=86M2026.01 | 64.06 | — | |
| LEG largeTrain Dataset=Toxic-Chat0124, Model Size=304M2026.01 | 52.77 | — | |
| Word Baseline largeTrain Dataset=Toxic-Chat0124, Model Size=304M2026.01 | 52.03 | — | |
| SHAP Baseline largeTrain Dataset=AEGIS 2.0, Model Size=-2026.01 | 45.94 | — | |
| LEG baseTrain Dataset=Toxic-Chat0124, Model Size=86M2026.01 | 45.91 | — | |
| Word Baseline baseTrain Dataset=Toxic-Chat0124, Model Size=86M2026.01 | 45.72 | — | |
| SHAP Baseline largeTrain Dataset=Wild-GuardMix, Model Size=-2026.01 | 43.99 | — | |
| SHAP Baseline baseTrain Dataset=AEGIS 2.0, Model Size=-2026.01 | 41.07 | — | |
| SHAP Baseline largeTrain Dataset=Toxic-Chat0124, Model Size=-2026.01 | 40.88 | — | |
| SHAP Baseline baseTrain Dataset=Wild-GuardMix, Model Size=-2026.01 | 37.21 | — | |
| SHAP Baseline baseTrain Dataset=Toxic-Chat0124, Model Size=-2026.01 | 33.12 | — | |
| LIME Baseline largeTrain Dataset=Wild-GuardMix, Model Size=-2026.01 | 32.74 | — | |
| LIME Baseline baseTrain Dataset=Wild-GuardMix, Model Size=-2026.01 | 32.53 | — | |
| LIME Baseline largeTrain Dataset=AEGIS 2.0, Model Size=-2026.01 | 31.74 | — | |
| LIME Baseline baseTrain Dataset=AEGIS 2.0, Model Size=-2026.01 | 31.18 | — | |
| LIME Baseline baseTrain Dataset=Toxic-Chat0124, Model Size=-2026.01 | 29.59 | — | |
| LIME Baseline largeTrain Dataset=Toxic-Chat0124, Model Size=-2026.01 | 29.34 | — | |
| LEG xsTrain Dataset=Toxic-Chat0124, Model Size=22M2026.01 | 26.48 | — | |
| GPT-4o-minizero-shot=true2026.01 | — | 53.28 | |
| LEG basesize=base2026.01 | — | 60.33 | |
| LEG largesize=large2026.01 | — | 66.11 |