Mathematical Reasoning on In-domain Reasoning Suite (AIME24, AIME25, AMC23, Math500, Minerva, Olympia) (test)
29.3AIME24 ScoreSFT+
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| SFT+Backbone=Qwen3-8b, Sampling=avg@16/avg@52026.03 | 29.3 | 24.7 | 67.3 | 84.8 | 42.5 | 54.5 | 50.5 | |
| SPOTBackbone=Qwen3-8b, Sampling=avg@16/avg@52026.03 | 28 | 27.3 | 71.5 | 87.4 | 39.7 | 58.5 | 52.1 | |
| RFTBackbone=Qwen3-8b, Sampling=avg@16/avg@52026.03 | 27.3 | 20 | 59 | 83 | 40.4 | 53.8 | 47.3 | |
| Qwen3-8bSampling=avg@16/avg@52026.03 | 22 | 19.3 | 66.5 | 82.8 | 37.9 | 52.3 | 46.8 | |
| SFTBackbone=Qwen3-8b, Sampling=avg@16/avg@52026.03 | 15.3 | 13.3 | 56 | 78.2 | 38.6 | 44.3 | 41 | |
| SPOTBackbone=Llama3.1-8b-Instruct, Sampling=avg@16/avg@52026.03 | 4 | 2 | 27 | 48.6 | 26.1 | 16.3 | 20.7 | |
| Llama3.1-8b-InstructSampling=avg@16/avg@52026.03 | 3.3 | 2.7 | 19.5 | 47 | 22.4 | 16.9 | 18.6 | |
| SFTBackbone=Llama3.1-8b-Instruct, Sampling=avg@16/avg@52026.03 | 3.3 | 2 | 21.5 | 46.8 | 19.5 | 15.1 | 18 | |
| RFTBackbone=Llama3.1-8b-Instruct, Sampling=avg@16/avg@52026.03 | 3.3 | 0 | 24 | 46 | 19.1 | 15.3 | 18 | |
| SFT+Backbone=Llama3.1-8b-Instruct, Sampling=avg@16/avg@52026.03 | 3.3 | 0.7 | 26 | 48.6 | 25.7 | 15.3 | 19.9 |