Intent Leakage on BioQA
24.25ASR@1Soft-PO
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Soft-PO2026.04 | 24.25 | 42.84 | 33.65 | |
| GTKA2026.04 | 21.67 | 40.13 | 31.26 | |
| PP-TSTrusted Local Model=Qwen2.5-3B-Instruct, Untrusted External Model=Qwen-Turbo2026.04 | 21.48 | 39.65 | 31.6 | |
| Soft-POTrusted Local Model=Qwen2.5-3B-Instruct, Untrusted External Model=Qwen-Turbo2026.04 | 20.56 | 38.72 | 30.77 | |
| IOITrusted Local Model=Qwen2.5-3B-Instruct, Untrusted External Model=Qwen-Turbo2026.04 | 19.89 | 37.99 | 30.03 | |
| Hard-POTrusted Local Model=Qwen2.5-3B-Instruct, Untrusted External Model=Qwen-Turbo2026.04 | 19.33 | 37.02 | 29.21 | |
| GTKATrusted Local Model=Qwen2.5-3B-Instruct, Untrusted External Model=Qwen-Turbo2026.04 | 18.41 | 36.14 | 28.43 | |
| PP-TSTrusted Local Model=Llama-3.1-8B, Untrusted External Model=GPT-4o mini2026.04 | 14.16 | 31.98 | 25.76 | |
| Soft-POTrusted Local Model=Llama-3.1-8B, Untrusted External Model=GPT-4o mini2026.04 | 13.44 | 31.05 | 24.78 | |
| IOITrusted Local Model=Llama-3.1-8B, Untrusted External Model=GPT-4o mini2026.04 | 12.71 | 30.14 | 24.16 | |
| Hard-POTrusted Local Model=Llama-3.1-8B, Untrusted External Model=GPT-4o mini2026.04 | 11.99 | 29.37 | 23.26 | |
| GTKATrusted Local Model=Llama-3.1-8B, Untrusted External Model=GPT-4o mini2026.04 | 11.12 | 28.57 | 22.52 |