Toxicity Detection on RealTox Standard
87.56AccuracyTrajectory (Raw)
Evaluation results
| Method | Links | |
|---|---|---|
| Trajectory (Raw)Backbone=Qwen2.5-14B2026.03 | 87.56 | |
| Trajectory (Raw)Backbone=Qwen3-32B2026.03 | 87.07 | |
| Trajectory (Raw)Backbone=Qwen3-30B MoE2026.03 | 86.57 | |
| TaT (Disp.)Backbone=Qwen2.5-14B2026.03 | 85.16 | |
| Trajectory (Raw)Backbone=Llama3.1-8b2026.03 | 82.16 | |
| TaT (Disp.)Backbone=Qwen3-30B MoE2026.03 | 79.43 | |
| TaT (Disp.)Backbone=Llama3.1-8b2026.03 | 79.35 | |
| Linear ProbeBackbone=Qwen3-30B MoE2026.03 | 77.87 | |
| Linear ProbeBackbone=Llama3.1-8b2026.03 | 77.86 | |
| TaT (Disp.)Backbone=Qwen3-32B2026.03 | 77.7 | |
| Linear ProbeBackbone=Qwen2.5-14B2026.03 | 76.46 | |
| Linear ProbeBackbone=Qwen3-32B2026.03 | 75.15 |