Question Answering on MMLU STEM (Acc, Len, Refl/Ans)
92.22AccuracyPLE
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| PLEBase Model=Qwen2.5-7B, Mode=No-think2026.04 | 92.22 | 139 | 0 | |
| PLEBase Model=Qwen3-4B, Mode=Think2026.04 | 92.22 | 1,041 | 0 | |
| HybridBase Model=Qwen3-4B, Mode=Think2026.04 | 90.26 | 2,338 | 16.82 | |
| PLEBase Model=Qwen2.5-7B, Mode=Think2026.04 | 88.89 | 429 | 0 | |
| SFT-onlyBase Model=Qwen2.5-7B, Mode=Think2026.04 | 85.82 | 3,100 | 24.74 | |
| HybridBase Model=Qwen3-4B, Mode=No-think2026.04 | 84.49 | 612 | 0.01 | |
| PLEBase Model=Qwen3-4B, Mode=No-think2026.04 | 83.7 | 253 | 0.01 | |
| InstructBase Model=Qwen2.5-7B, Mode=No-think2026.04 | 62 | 211 | 0 | |
| SFT-onlyBase Model=Qwen2.5-7B, Mode=No-think2026.04 | 59.44 | 774 | 0.83 |