Mathematical Reasoning on Out-domain Aggregate SAT Math, AMC23, AIME24, OLYMPIAD Bench
52.98Avg Acc (A_bar)L1
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| L1Base Model=DeepSeek-1.5B2026.02 | 52.98 | 9,902.2 | 0.12 | 0.12 | |
| CRTBase Model=DeepSeek-1.5B2026.02 | 52.43 | 8,786.4 | 0.21 | 0.21 | |
| O1-PrunerBase Model=DeepSeek-1.5B, Ratio=4.02026.02 | 51.58 | 9,141 | 0.12 | 0.12 | |
| ACPOBase Model=DeepSeek-1.5B2026.02 | 51.33 | 8,772.2 | 0.14 | 0.14 | |
| DeepSeek-1.5BModel Type=Base Model2026.02 | 50.84 | 9,894.5 | 0 | 0 | |
| O1-PrunerBase Model=DeepSeek-1.5B, Ratio=1.02026.02 | 49.76 | 9,042.8 | -0.01 | -0.12 | |
| ShorterBetterBase Model=DeepSeek-1.5B2026.02 | 49.58 | 7,722.4 | 0.1 | -0.03 | |
| ThinkPruneBase Model=DeepSeek-1.5B, Constraint=1k2026.02 | 49.46 | 8,908.5 | -0.03 | -0.17 | |
| ThinkPruneBase Model=DeepSeek-1.5B, Constraint=5002026.02 | 48.72 | 9,123.2 | -0.13 | -0.33 |