Multiple-choice Commonsense Reasoning on CSQA (test)
98.2TDRCritical-CoT
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Critical-CoTModel=GPT-OSS-20B, Backdoor Attack Type=In-Context Learning-Based Backdoor2026.04 | 98.2 | 98.8 | 86.5 | 77 | 10 | |
| Critical-CoTModel=LLaMA-2-13B, Backdoor Attack Type=In-Context Learning-Based Backdoor2026.04 | 98 | 98.9 | 72.7 | 63 | 8 | |
| Critical-CoTModel=Qwen3-14B, Backdoor Attack Type=In-Context Learning-Based Backdoor2026.04 | 96.9 | 99 | 89.6 | 62 | 6 | |
| Critical-CoTModel=Qwen3-14B, Backdoor Attack Type=Fine-Tuning-Based Backdoor2026.04 | 94.5 | — | 89.2 | 54 | 22 | |
| Critical-CoTModel=GPT-OSS-20B, Backdoor Attack Type=Fine-Tuning-Based Backdoor2026.04 | 89.4 | — | 78.4 | 20 | 12 | |
| Critical-CoTModel=LLaMA-2-13B, Backdoor Attack Type=Fine-Tuning-Based Backdoor2026.04 | 86.5 | — | 45.6 | 65 | 15 |