Privacy Probing and Leakage Assessment on PrivacyLens
99.5QA Probing AccuracyQwen3.5-2B
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Qwen3.5-2BCondition=SFT2026.04 | 99.5 | 25.8 | 60.5 | 1.7 | |
| CIRL-7BCondition=Zero-shot2026.04 | 99.2 | 24.1 | 83.2 | 2.4 | |
| Gemma-3-12BCondition=SFT2026.04 | 98.6 | 30.6 | 95.3 | 2.8 | |
| ContextReasoner-7BCondition=Zero-shot2026.04 | 98.5 | 30.6 | 34.5 | 1 | |
| Qwen3.5-2BCondition=Zero-shot2026.04 | 98.4 | 26.8 | 69.6 | 2 | |
| Gemma-3-12BCondition=Zero-shot2026.04 | 98.4 | 30.7 | 97.2 | 2.8 | |
| Qwen3.5-9BCondition=SFT2026.04 | 97.6 | 27.4 | 97.8 | 2.8 | |
| Qwen3.5-9BCondition=GRPO2026.04 | 97.6 | 26.7 | 97.4 | 2.8 | |
| Phi-4Condition=SFT2026.04 | 97 | 22 | 99.6 | 2.9 | |
| Phi-4Condition=Zero-shot2026.04 | 96.9 | 23.1 | 99.4 | 2.9 | |
| OpenThinker3-7BCondition=SFT2026.04 | 96.2 | 26.5 | 75.9 | 2.2 | |
| Qwen3.5-4BCondition=SFT2026.04 | 95.3 | 29.2 | 95.1 | 2.7 | |
| Qwen3.5-9BCondition=Zero-shot2026.04 | 95.1 | 27.1 | 97.2 | 2.8 | |
| Qwen3.5-4BCondition=Zero-shot2026.04 | 90.5 | 31.9 | 96.5 | 2.8 | |
| GPT-OSS-20BCondition=Zero-shot2026.04 | 88.4 | 12.1 | 98.8 | 2.9 | |
| GPT-OSS-20BCondition=SFT2026.04 | 85.9 | 17.6 | 93.3 | 2.8 | |
| OpenThinker3-7BCondition=Zero-shot2026.04 | 85.9 | 27.1 | 44.2 | 1.3 |