Mathematical Reasoning on MATH 500 (Tokens, ActRatio, Acc@all)
452Tokens UsedExtra-CoT (CHRPO)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Extra-CoT (CHRPO)Ratio=<POLICY>, Backbone=Qwen3-1.7B2026.02 | 452 | 0.27 | 64.8 | |
| Extra-CoTRatio=0.2, Backbone=Qwen3-1.7B2026.02 | 481 | 0.29 | 47.8 | |
| TokenSkipRatio=0.2, Backbone=Qwen3-1.7B2026.02 | 660 | 0.39 | 23.4 | |
| Thinkless* (DeGRPO)Ratio=–, Backbone=Qwen3-1.7B2026.02 | 888 | 0.53 | 63.6 | |
| Extra-CoTRatio=0.4, Backbone=Qwen3-1.7B2026.02 | 920 | 0.54 | 54.2 | |
| TokenSkipRatio=<POLICY>, Backbone=Qwen3-1.7B2026.02 | 1,182 | 0.71 | 35.8 | |
| TokenSkipRatio=0.4, Backbone=Qwen3-1.7B2026.02 | 1,259 | 0.75 | 34.4 | |
| TruncationRatio=0.6, Backbone=Qwen3-1.7B2026.02 | 1,271 | 0.76 | 49.8 | |
| Extra-CoTRatio=0.6, Backbone=Qwen3-1.7B2026.02 | 1,304 | 0.77 | 59.6 | |
| Extra-CoTRatio=<POLICY>, Backbone=Qwen3-1.7B2026.02 | 1,312 | 0.78 | 63 | |
| LC-PromptRatio=0.6, Backbone=Qwen3-1.7B2026.02 | 1,466 | 0.88 | 60.4 | |
| TokenSkipRatio=0.6, Backbone=Qwen3-1.7B2026.02 | 1,517 | 0.9 | 47.4 | |
| Extra-CoTRatio=0.8, Backbone=Qwen3-1.7B2026.02 | 1,520 | 0.9 | 64.2 | |
| TokenSkipRatio=0.8, Backbone=Qwen3-1.7B2026.02 | 1,527 | 0.91 | 53 | |
| Base ModelBackbone=Qwen3-1.7B2026.02 | 1,675 | — | 64.2 | |
| TokenSkipRatio=1.0, Backbone=Qwen3-1.7B2026.02 | 1,696 | 1.01 | 54.4 | |
| Extra-CoTRatio=1.0, Backbone=Qwen3-1.7B2026.02 | 1,698 | 1.01 | 64 |