Math Problem Solving on OlympiadBench
68.32AccuracyLESS
Evaluation Results
| Method | Links | |
|---|---|---|
| LESSBase Model=OpenR1-Distill-7B2026.06 | 68.32 | |
| DRIFTBase Model=OpenR1-Distill-7B2026.06 | 68.29 | |
| QuratingBase Model=OpenR1-Distill-7B2026.06 | 68.25 | |
| BM25Base Model=OpenR1-Distill-7B2026.06 | 67.62 | |
| IF (Impl. w. GraSS)Base Model=OpenR1-Distill-7B2026.06 | 67.62 | |
| OpenR1-Distill-7BBase Model=OpenR1-Distill-7B2026.06 | 67.51 | |
| RDSBase Model=OpenR1-Distill-7B2026.06 | 66.77 | |
| Random BaselineBase Model=OpenR1-Distill-7B2026.06 | 66.58 | |
| DSIRBase Model=OpenR1-Distill-7B2026.06 | 65.95 | |
| Self-DistillationBase Model=OpenR1-Distill-7B2026.06 | 60.24 | |
| TaH+Param.=4B2025.11 | 52 | |
| TaHParam.=4B2025.11 | 50.5 | |
| SoftThinkParam.=4B2025.11 | 50.1 | |
| StandardParam.=4B2025.11 | 47.7 | |
| RoutingParam.=4B2025.11 | 45 | |
| ACPORegime=N-OnP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.06 | 42.83 | |
| ACPORegime=OffP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.06 | 42.17 | |
| DAPORegime=N-OnP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.06 | 41.44 | |
| TaH+Param.=1.7B2025.11 | 41.2 | |
| CISPORegime=N-OnP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.06 | 40.61 | |
| CISPORegime=OffP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.06 | 39.1 | |
| TaHParam.=1.7B2025.11 | 38.8 | |
| DAPORegime=OffP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.06 | 37.2 | |
| Self-DistillationBase Model=Olmo3-7B-Instruct-SFT2026.06 | 36.31 | |
| DRIFTBase Model=Olmo3-7B-Instruct-SFT2026.06 | 35.61 | |
| BM25Base Model=Olmo3-7B-Instruct-SFT2026.06 | 35.5 | |
| Random BaselineBase Model=Olmo3-7B-Instruct-SFT2026.06 | 35.35 | |
| QuratingBase Model=Olmo3-7B-Instruct-SFT2026.06 | 35.27 | |
| RDSBase Model=Olmo3-7B-Instruct-SFT2026.06 | 34.9 | |
| Low-EntropyRegime=N-OnP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.06 | 34.69 | |
| High-EntropyRegime=OffP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.06 | 34.54 | |
| IF (Impl. w. GraSS)Base Model=Olmo3-7B-Instruct-SFT2026.06 | 34.31 | |
| Olmo3-7B-Instruct-SFTBase Model=Olmo3-7B-Instruct-SFT2026.06 | 34.27 | |
| AR-LoptiRegime=N-OnP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.06 | 33.6 | |
| High-EntropyRegime=N-OnP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.06 | 33.54 | |
| SoftThinkParam.=1.7B2025.11 | 33.4 | |
| StandardParam.=1.7B2025.11 | 33 | |
| LESSBase Model=Olmo3-7B-Instruct-SFT2026.06 | 32.34 | |
| RoutingParam.=1.7B2025.11 | 30.6 | |
| AlwaysThinkParam.=1.7B2025.11 | 30 | |
| Low-EntropyRegime=OffP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.06 | 26.61 | |
| DSIRBase Model=Olmo3-7B-Instruct-SFT2026.06 | 26.59 | |
| TaH+Param.=0.6B2025.11 | 24 | |
| AR-LoptiRegime=OffP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.06 | 24 | |
| TaHParam.=0.6B2025.11 | 23.9 | |
| SoftThinkParam.=0.6B2025.11 | 19.4 | |
| StandardParam.=0.6B2025.11 | 18.8 | |
| Base ModelRegime=N-OnP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.06 | 18.02 | |
| AlwaysThinkParam.=0.6B2025.11 | 10.2 | |
| RoutingParam.=0.6B2025.11 | 7.4 |