Multiple Choice Question Answering on MMLU (Cross-Benchmark Evaluation)
77.4AccuracyReal-target CSFT
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Real-target CSFTCondition=Real-target CSFT2026.04 | 77.4 | 61.6 | 470 | |
| Shuffled-targetCondition=Shuffled-target2026.04 | 56.1 | 52.3 | 66 | |
| BaselineCondition=Baseline2026.04 | 54.2 | 53.5 | 498 |