Mathematical Reasoning on GSM8K (Acc, Peak, Dep)
87.92Accuracy (GSM8K)TokenSkip
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| TokenSkipModel Series=Qwen2.5-7B, Budget setting=true2026.04 | 87.92 | 775 | 1 | |
| LThinker++Model Series=Qwen2.5-7B, Budget setting=true2026.04 | 87.87 | 471 | 0.4 | |
| VanillaModel Series=Qwen2.5-7B, Budget setting=true2026.04 | 87.62 | 782 | 1.1 | |
| CoTModel Series=Qwen2.5-7B, Budget setting=true2026.04 | 86.12 | 513 | 0.1 | |
| CoTModel Series=Llama3.1-8B, Budget setting=true2026.04 | 85.14 | 550 | 0.2 | |
| LThinker*Model Series=Qwen2.5-7B, Budget setting=true2026.04 | 84.61 | 444 | 0.7 | |
| Distill-R1Model Series=Qwen2.5-7B, Budget setting=true2026.04 | 81.88 | 844 | 1.1 | |
| TokenSkipModel Series=Llama3.1-8B, Budget setting=true2026.04 | 79.4 | 838 | 1.2 | |
| VanillaModel Series=Llama3.1-8B, Budget setting=true2026.04 | 79.38 | 740 | 0.9 | |
| LThinker++Model Series=Llama3.1-8B, Budget setting=true2026.04 | 77.69 | 528 | 0.7 | |
| LThinker*Model Series=Llama3.1-8B, Budget setting=true2026.04 | 76.72 | 522 | 0.9 | |
| Distill-R1Model Series=Llama3.1-8B, Budget setting=true2026.04 | 73.62 | 395 | 0.1 |