Multiple Choice Question Answering on MMLU (Performance Change Tracking)
58MMLU Accuracy Gain (Absolute)FLOATDOOR
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| FLOATDOORScenario=M12026.06 | 58 | — | — | |
| FLOATDOORScenario=M22026.06 | 35 | — | — | |
| FLOATDOORScenario=S32026.06 | 28 | — | — | |
| FLOATDOORScenario=S22026.06 | 7 | — | — | |
| FLOATDOORScenario=S12026.06 | 2 | — | — | |
| CRL-TokenBackbone=LLaMA-3.1 8B, Type=knowledge, Layer=20, Evaluation Protocol=0-shot2026.02 | 0.92 | — | 62.33 | |
| LLaMA-3.1 8BEvaluation Protocol=0-shot2026.02 | — | 61.41 | — |