General Language Understanding on MMLU (Average accepted length (m), Accuracy)
7.576Average Accepted Length (m)SelfJudge
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SelfJudgeTarget model=Llama-3.1-70B, Draft model=Llama-3.1-8B, Threshold (theta)=0.52025.09 | 7.576 | 77.2 | |
| SelfJudge-FTarget model=Llama-3.1-70B, Draft model=Llama-3.1-8B, Threshold (theta)=0.382025.09 | 6.955 | 79.8 | |
| SelfJudge-RTarget model=Llama-3.1-70B, Draft model=Llama-3.1-8B, Threshold (theta)=0.282025.09 | 6.262 | 81.4 | |
| SDTarget model=Llama-3.1-70B, Draft model=Llama-3.1-8B2025.09 | 5.668 | 80.6 |