Mathematical Reasoning on AMC23, AIME24, AIME25, MATH500 (test)
43.91Average PerformanceOurs
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Oursbackbone=Qwen2.5-7B-Instruct, fine-tuning=true, training_dataset=OpenThoughts-Math, sample_size=10002026.06 | 43.91 | 59.17 | 21.11 | 18.15 | 77.2 | |
| Embedding Diversitybackbone=Qwen2.5-7B-Instruct, fine-tuning=true, training_dataset=OpenThoughts-Math, sample_size=10002026.06 | 43.5 | 57.86 | 22.38 | 16.67 | 77.2 | |
| Randombackbone=Qwen2.5-7B-Instruct, fine-tuning=true, training_dataset=OpenThoughts-Math, sample_size=10002026.06 | 43.12 | 56.88 | 20.83 | 16.67 | 78.1 | |
| S2Lbackbone=Qwen2.5-7B-Instruct, fine-tuning=true, training_dataset=OpenThoughts-Math, sample_size=10002026.06 | 42.72 | 51.25 | 16.67 | 26.67 | 76.3 | |
| Learnabilitybackbone=Qwen2.5-7B-Instruct, fine-tuning=true, training_dataset=OpenThoughts-Math, sample_size=10002026.06 | 41.73 | 53 | 20.67 | 15.67 | 77.6 | |
| Middle Perplexitybackbone=Qwen2.5-7B-Instruct, fine-tuning=true, training_dataset=OpenThoughts-Math, sample_size=10002026.06 | 39.03 | 51.07 | 13.33 | 16.19 | 75.5 |