Mathematical Reasoning on GSM8K (Acc, Tok)
86.8AccuracyInfoDensity
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| InfoDensityBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.03 | 86.8 | 1,352 | |
| DSBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.03 | 86.5 | 1,551 | |
| GRPO-AccBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.03 | 86.4 | 1,757 | |
| GRPO-LPBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.03 | 86.4 | 1,560 | |
| OriginalBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.03 | 85.4 | 2,558 | |
| PEARBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.03 | 85.1 | 1,046 | |
| GRPO-AccBackbone=Qwen3-0.6B2026.03 | 82.9 | 2,310 | |
| PEARBackbone=Qwen3-0.6B2026.03 | 80.2 | 1,190 | |
| OriginalBackbone=Qwen3-0.6B2026.03 | 78.3 | 2,013 | |
| InfoDensityBackbone=Qwen3-0.6B2026.03 | 78.3 | 865 | |
| DSBackbone=Qwen3-0.6B2026.03 | 78 | 1,278 | |
| GRPO-LPBackbone=Qwen3-0.6B2026.03 | 77.8 | 1,212 |