General Language Understanding on MMLU-Pro (MMLU-Pro(%))
56.82MMLU-Pro ScoreStealth Fine-Tuning
Evaluation Results
| Method | Links | |
|---|---|---|
| Stealth Fine-TuningFine-Tuning=true2025.11 | 56.82 | |
| Stealth Fine-Tuning+Segment-level InterferenceFine-Tuning=true2025.11 | 56.82 | |
| no attackFine-Tuning=false2025.11 | 56.09 | |
| Fine-Tuning(pure bad with CoT)Fine-Tuning=true2025.11 | 53.6 | |
| Fine-Tuning(pure bad)Fine-Tuning=true2025.11 | 49.92 | |
| Fine-Tuning(AOA)Fine-Tuning=true2025.11 | 48.33 |