Scientific Reasoning on MMLU-Pro (Mean@16, Pass@16)
25.96Mean Score @16ADS
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ADSBackbone=Qwen2.5-Math-1.5B2026.06 | 25.96 | 74.73 | |
| ADSBackbone=Qwen3-4B-Base2026.06 | 20.96 | 63.8 | |
| GRPOBackbone=Qwen3-4B-Base2026.06 | 20.04 | 65.47 | |
| DOTS+RRBackbone=Qwen2.5-Math-1.5B2026.06 | 19.12 | 68.33 | |
| GRPOBackbone=Qwen2.5-Math-1.5B2026.06 | 17.37 | 69.8 | |
| ADSBackbone=Olmo3-7B-SFT2026.06 | 13.56 | 59.53 | |
| DOTS+RRBackbone=Qwen3-4B-Base2026.06 | 12.76 | 57.27 | |
| GRPOBackbone=Olmo3-7B-SFT2026.06 | 10.83 | 52.6 | |
| BaseBackbone=Qwen3-4B-Base2026.06 | 10.49 | 59.93 | |
| BaseBackbone=Qwen2.5-Math-1.5B2026.06 | 10.33 | 59.13 | |
| DOTS+RRBackbone=Olmo3-7B-SFT2026.06 | 9.2 | 51.87 | |
| BaseBackbone=Olmo3-7B-SFT2026.06 | 6.64 | 43 |