Multitask Language Understanding on MMLU (Score, Delta)
70.41MMLU ScoreIPROX
Evaluation results
| Method | Links | ||
|---|---|---|---|
| IPROXTarget Model=Qwen2-7B2026.02 | 70.41 | 0.1 | |
| Layer ExtractionTarget Model=Qwen2-7B2026.02 | 70.31 | — | |
| Influence ScorerTarget Model=Qwen2-7B2026.02 | 70.28 | — | |
| IPROXTarget Model=Qwen3-4B2026.02 | 70.15 | 0.29 | |
| Layer ExtractionTarget Model=Qwen3-4B2026.02 | 69.86 | — | |
| Influence ScorerTarget Model=Qwen3-4B2026.02 | 69.6 | — |