Mathematical Reasoning on MATH500, AMC23, GSM8K (test)
86.6Accuracy on MATH500HCC
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| HCCBackbone=Qwen2.5-Math-7B-Instruct, SFT Data Source=SFT on {T}R2026.05 | 86.6 | 65 | 96.2 | 82.6 | 455.9 | |
| EditorBackbone=Qwen2.5-Math-7B-Instruct, SFT Data Source=SFT on {T}R2026.05 | 86.4 | 67.5 | 96 | 83.3 | 443.2 | |
| VanillaBackbone=Qwen2.5-Math-7B-Instruct, SFT Data Source=SFT on {T}Q2026.05 | 85.8 | 62.5 | 95.8 | 81.4 | 519.6 | |
| VanillaBackbone=Qwen2.5-Math-7B-Instruct, SFT Data Source=SFT on {T}R2026.05 | 85.6 | 62.5 | 95.8 | 81.3 | 510.6 | |
| EditorBackbone=Qwen2.5-Math-7B-Instruct, SFT Data Source=SFT on {T}Q2026.05 | 84.4 | 70 | 96.2 | 83.5 | 499 | |
| HeuristicBackbone=Qwen2.5-Math-7B-Instruct, SFT Data Source=SFT on {T}R2026.05 | 84.4 | 62.5 | 95.7 | 80.9 | 498.5 | |
| HeuristicBackbone=Qwen2.5-Math-7B-Instruct, SFT Data Source=SFT on {T}Q2026.05 | 82.8 | 57.5 | 95.8 | 78.7 | 526.1 | |
| HCCBackbone=Qwen2.5-Math-7B-Instruct, SFT Data Source=SFT on {T}Q2026.05 | 82.6 | 62.5 | 96.1 | 80.4 | 497.9 | |
| EditorBackbone=LLaMA3.2-3B-Instruct, SFT Data Source=SFT on {T}R2026.05 | 51.2 | 22.5 | 75.9 | 49.8 | 1,942.1 | |
| HCCBackbone=LLaMA3.2-3B-Instruct, SFT Data Source=SFT on {T}R2026.05 | 47.6 | 22.5 | 77.8 | 49.3 | 1,934.4 | |
| HCCBackbone=LLaMA3.2-3B-Instruct, SFT Data Source=SFT on {T}Q2026.05 | 43.2 | 17.5 | 75.1 | 45.2 | 2,010.1 | |
| EditorBackbone=LLaMA3.2-3B-Instruct, SFT Data Source=SFT on {T}Q2026.05 | 41.8 | 20 | 75.4 | 45.7 | 1,906.1 | |
| HeuristicBackbone=LLaMA3.2-3B-Instruct, SFT Data Source=SFT on {T}R2026.05 | 40 | 15 | 70.1 | 41.7 | 5,172.1 | |
| VanillaBackbone=LLaMA3.2-3B-Instruct, SFT Data Source=SFT on {T}R2026.05 | 38.4 | 15 | 69.9 | 41.1 | 4,967.1 | |
| HeuristicBackbone=LLaMA3.2-3B-Instruct, SFT Data Source=SFT on {T}Q2026.05 | 34 | 15 | 71.5 | 40.2 | 3,430.8 | |
| VanillaBackbone=LLaMA3.2-3B-Instruct, SFT Data Source=SFT on {T}Q2026.05 | 29.8 | 10 | 69 | 36.3 | 3,478.8 |