Mathematical Reasoning on AMC 2023 (Tokens, ActRatio, Acc@all)
675TokensExtra-CoT (CHRPO)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Extra-CoT (CHRPO)Ratio=<POLICY>, Backbone=Qwen3-1.7B2026.02 | 675 | 0.32 | 50 | |
| Extra-CoTRatio=0.2, Backbone=Qwen3-1.7B2026.02 | 782 | 0.37 | 17.5 | |
| TokenSkipRatio=0.2, Backbone=Qwen3-1.7B2026.02 | 911 | 0.44 | 10 | |
| Extra-CoTRatio=0.4, Backbone=Qwen3-1.7B2026.02 | 945 | 0.45 | 25 | |
| TruncationRatio=0.6, Backbone=Qwen3-1.7B2026.02 | 1,334 | 0.64 | — | |
| Thinkless* (DeGRPO)Ratio=–, Backbone=Qwen3-1.7B2026.02 | 1,369 | 0.65 | 50 | |
| TokenSkipRatio=<POLICY>, Backbone=Qwen3-1.7B2026.02 | 1,437 | 0.69 | 12.5 | |
| Extra-CoTRatio=<POLICY>, Backbone=Qwen3-1.7B2026.02 | 1,515 | 0.72 | 42.5 | |
| Extra-CoTRatio=0.6, Backbone=Qwen3-1.7B2026.02 | 1,624 | 0.78 | 40 | |
| TokenSkipRatio=0.4, Backbone=Qwen3-1.7B2026.02 | 1,743 | 0.83 | 12.5 | |
| Extra-CoTRatio=0.8, Backbone=Qwen3-1.7B2026.02 | 1,845 | 0.88 | 42.5 | |
| LC-PromptRatio=0.6, Backbone=Qwen3-1.7B2026.02 | 2,015 | 0.96 | 45 | |
| TokenSkipRatio=0.6, Backbone=Qwen3-1.7B2026.02 | 2,077 | 0.99 | 20 | |
| Base ModelBackbone=Qwen3-1.7B2026.02 | 2,092 | — | 47.5 | |
| Extra-CoTRatio=1.0, Backbone=Qwen3-1.7B2026.02 | 2,153 | 1.03 | 42.5 | |
| TokenSkipRatio=0.8, Backbone=Qwen3-1.7B2026.02 | 2,302 | 1.1 | 27.5 | |
| TokenSkipRatio=1.0, Backbone=Qwen3-1.7B2026.02 | 2,333 | 1.12 | 37.5 |