Robustness Evaluation on GSM8K
0FPR (%)Master-RM
Evaluation results
| Method | Links | |
|---|---|---|
| Master-RMTarget Judge=Omni-Judge2025.12 | 0 | |
| Master-RMTarget Judge=General Verifier2025.12 | 0 | |
| Master-RMTarget Judge=Qwen2.5-RLVR2025.12 | 0 | |
| Master-RMTarget Judge=Master-RM Judge2025.12 | 0 | |
| AdvJudge-ZeroTarget Judge=General Verifier2025.12 | 0.01 | |
| AdvJudge-ZeroTarget Judge=Master-RM Judge2025.12 | 8.86 | |
| AdvJudge-ZeroTarget Judge=Qwen2.5-RLVR2025.12 | 9.18 | |
| AdvJudge-ZeroTarget Model=Llama-3.2-3B2025.12 | 69.34 | |
| Master-RMTarget Model=gemma-3-4b2025.12 | 92.2 | |
| Master-RMTarget Model=Qwen2.5-7B2025.12 | 92.96 | |
| Master-RMTarget Model=Qwen3-4B2025.12 | 97.9 | |
| Master-RMTarget Model=Qwen3-30B2025.12 | 98.29 | |
| Master-RMTarget Model=Llama-3.3-70B2025.12 | 98.65 | |
| Master-RMTarget Model=Llama-3.2-3B2025.12 | 99.16 | |
| AdvJudge-ZeroTarget Model=Llama-3.3-70B2025.12 | 99.56 | |
| AdvJudge-ZeroTarget Judge=Omni-Judge2025.12 | 99.79 | |
| AdvJudge-ZeroTarget Model=gemma-3-4b2025.12 | 99.92 | |
| AdvJudge-ZeroTarget Model=Qwen2.5-7B2025.12 | 99.96 | |
| AdvJudge-ZeroTarget Model=Qwen3-4B2025.12 | 100 | |
| AdvJudge-ZeroTarget Model=Qwen3-30B2025.12 | 100 |