Open-ended arithmetic reasoning on MATH (test)
96.9TDRCritical-CoT
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Critical-CoTModel=LLaMA-2-13B, Backdoor Attack Type=In-Context Learning-Based Backdoor2026.04 | 96.9 | 98.6 | 38.1 | 78 | 0 | |
| Critical-CoTModel=Qwen3-14B, Backdoor Attack Type=In-Context Learning-Based Backdoor2026.04 | 96.8 | 97.9 | 92.8 | 68 | 3 | |
| Critical-CoTModel=LLaMA-2-13B, Backdoor Attack Type=Fine-Tuning-Based Backdoor2026.04 | 95.9 | — | 32.4 | 42 | 20 | |
| Critical-CoTModel=GPT-OSS-20B, Backdoor Attack Type=In-Context Learning-Based Backdoor2026.04 | 92.7 | 93.8 | 91.7 | 68 | 9 | |
| Critical-CoTModel=Qwen3-14B, Backdoor Attack Type=Fine-Tuning-Based Backdoor2026.04 | 92.7 | — | 77.1 | 10 | 6 | |
| Critical-CoTModel=GPT-OSS-20B, Backdoor Attack Type=Fine-Tuning-Based Backdoor2026.04 | 84.6 | — | 79.6 | 65 | 0 |