Mathematical Reasoning on GSM8K (Acc, TNFT, Tokens, Delay)
91.85AccuracyBase
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| BaseBackbone=Qwen3-4B2026.05 | 91.85 | 117.03 | 1,340.49 | 41.94 | |
| VanillaBackbone=Qwen3-1.7B2026.05 | 90.6 | 93.3 | 660.64 | 14.93 | |
| BaseBackbone=Qwen3-1.7B2026.05 | 90.37 | 117.03 | 1,156.1 | 27.14 | |
| OursBackbone=Qwen3-1.7B2026.05 | 89.51 | 0 | 437.1 | 11.29 | |
| VanillaBackbone=Qwen3-4B2026.05 | 89.36 | 93.3 | 649.28 | 20.17 | |
| OursBackbone=Qwen3-4B2026.05 | 88.85 | 0 | 421.47 | 14.53 |