Science on MMLU-Pro (test)
41.9AccuracyBase
Evaluation results
| Method | Links | |
|---|---|---|
| BaseBackbone=Qwen2.5-3B-Instruct2026.05 | 41.9 | |
| BaseBackbone=Qwen2.5-3B-Instruct2026.05 | 41.9 | |
| Low-SFTBackbone=Qwen2.5-3B-Instruct2026.05 | 41.7 | |
| DFTBackbone=Qwen2.5-3B-Instruct2026.05 | 41.3 | |
| Self-SFTBackbone=Qwen2.5-3B-Instruct2026.05 | 41.1 | |
| Anchored LearningBackbone=Qwen2.5-3B-Instruct2026.05 | 40.9 | |
| Self-sftBackbone=Qwen2.5-3B-Instruct2026.05 | 39.4 | |
| KL-SFTBackbone=Qwen2.5-3B-Instruct2026.05 | 38.3 | |
| SFTBackbone=Qwen2.5-3B-Instruct2026.05 | 37.5 | |
| STMBackbone=Qwen2.5-3B-Instruct2026.05 | 36.1 | |
| Iter-SFTBackbone=Qwen2.5-3B-Instruct2026.05 | 35.8 | |
| Anchored LearningBackbone=Qwen2.5-3B-Instruct2026.05 | 35.6 | |
| Iter-SFTBackbone=Qwen2.5-3B-Instruct2026.05 | 33.8 | |
| Low-SFTBackbone=Qwen2.5-3B-Instruct2026.05 | 31.7 | |
| DFTBackbone=Qwen2.5-3B-Instruct2026.05 | 29.7 | |
| STMBackbone=Qwen2.5-3B-Instruct2026.05 | 27.7 | |
| SFTBackbone=Qwen2.5-3B-Instruct2026.05 | 10.9 | |
| KL-SFTBackbone=Qwen2.5-3B-Instruct2026.05 | 9.6 |