Toxicity Detection on RealTox Challenging
98.92AccuracyTrajectory (Raw)
Evaluation Results
| Method | Links | |
|---|---|---|
| Trajectory (Raw)Backbone=Qwen3-30B MoE2026.03 | 98.92 | |
| Trajectory (Raw)Backbone=Qwen2.5-14B2026.03 | 98.83 | |
| Trajectory (Raw)Backbone=Qwen3-32B2026.03 | 98.83 | |
| TaT (Disp.)Backbone=Qwen2.5-14B2026.03 | 98.58 | |
| Trajectory (Raw)Backbone=Llama3.1-8b2026.03 | 97.91 | |
| TaT (Disp.)Backbone=Llama3.1-8b2026.03 | 96 | |
| Linear ProbeBackbone=Llama3.1-8b2026.03 | 95.83 | |
| Linear ProbeBackbone=Qwen2.5-14B2026.03 | 95.16 | |
| TaT (Disp.)Backbone=Qwen3-32B2026.03 | 95.08 | |
| Linear ProbeBackbone=Qwen3-30B MoE2026.03 | 94.5 | |
| Linear ProbeBackbone=Qwen3-32B2026.03 | 91.16 | |
| TaT (Disp.)Backbone=Qwen3-30B MoE2026.03 | 87.66 |