Harmfulness Evaluation on HarmMetric Eval
89.6Score (Effectiveness)HarmClassifier
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| HarmClassifierOutputs=Binary (Yes or No), Backbone=Qwen2.5-7B-Instruct, Fine-tuning=LoRA2025.09 | 89.6 | 93.4 | 93.4 | 93.4 | 90.9 | 93.6 | 96.7 | |
| HarmJudgeOutputs=Three-way (Yes, Uncertain, or No)2025.09 | 78.6 | 94.9 | 94.9 | 94.5 | 88.7 | 96.4 | 67.9 |