Mathematical Reasoning on MATH 500 (Acc, Tokens, CR, TE)
75.8AccuracyOriginal
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| OriginalModel=Qwen2.5-14B, Compression Setting=None2026.01 | 75.8 | 583.66 | 100 | 12.99 | |
| OriginalModel=Qwen2.5-7B, Compression Setting=None2026.01 | 71.2 | 574.58 | 100 | 7.7 | |
| CtrlCoTModel=Qwen2.5-14B, Compression Setting=High+2026.01 | 66.6 | 270.8 | 46 | 24.59 | |
| OriginalModel=Qwen2.5-3B, Compression Setting=None2026.01 | 63.2 | 575.66 | 100 | 9.11 | |
| CtrlCoTModel=Qwen2.5-14B, Compression Setting=High++2026.01 | 60.2 | 188.61 | 32 | 31.92 | |
| CtrlCoTModel=Qwen2.5-7B, Compression Setting=High+2026.01 | 58 | 225.61 | 39 | 25.71 | |
| CtrlCoTModel=Qwen2.5-14B, Compression Setting=High+++2026.01 | 58 | 137.55 | 24 | 42.17 | |
| TokenSkipModel=Qwen2.5-14B, Compression Setting=High+2026.01 | 54.8 | 285.42 | 49 | 19.2 | |
| CtrlCoTModel=Qwen2.5-7B, Compression Setting=High++2026.01 | 54 | 187.34 | 33 | 28.82 | |
| CtrlCoTModel=Qwen2.5-7B, Compression Setting=High+++2026.01 | 49.4 | 140.17 | 24 | 35.24 | |
| TokenSkipModel=Qwen2.5-14B, Compression Setting=High++2026.01 | 44.4 | 227.86 | 39 | 19.49 | |
| TokenSkipModel=Qwen2.5-7B, Compression Setting=High+2026.01 | 43.6 | 258.58 | 45 | 16.86 | |
| CtrlCoTModel=Qwen2.5-3B, Compression Setting=High++2026.01 | 42 | 245.06 | 43 | 17.14 | |
| CtrlCoTModel=Qwen2.5-3B, Compression Setting=High+2026.01 | 39.8 | 249.26 | 43 | 15.97 | |
| TruncationModel=Qwen2.5-7B, Compression Setting=High+2026.01 | 36.6 | 379.45 | 66 | 9.65 | |
| TokenSkipModel=Qwen2.5-7B, Compression Setting=High++2026.01 | 35.8 | 199.87 | 35 | 17.91 | |
| CtrlCoTModel=Qwen2.5-3B, Compression Setting=High+++2026.01 | 34.8 | 140.06 | 24 | 24.85 | |
| TruncationModel=Qwen2.5-14B, Compression Setting=High+2026.01 | 33.6 | 381.41 | 65 | 8.81 | |
| TokenSkipModel=Qwen2.5-14B, Compression Setting=High+++2026.01 | 33.2 | 164.98 | 28 | 20.12 | |
| TruncationModel=Qwen2.5-3B, Compression Setting=High+2026.01 | 33 | 377.83 | 66 | 8.73 | |
| TokenSkipModel=Qwen2.5-3B, Compression Setting=High+2026.01 | 31.4 | 284.91 | 49 | 11.02 | |
| TokenSkipModel=Qwen2.5-7B, Compression Setting=High+++2026.01 | 28.2 | 142.76 | 25 | 19.75 | |
| TokenSkipModel=Qwen2.5-3B, Compression Setting=High++2026.01 | 22.8 | 222 | 39 | 10.27 | |
| TokenSkipModel=Qwen2.5-3B, Compression Setting=High+++2026.01 | 18.6 | 167.11 | 29 | 11.13 | |
| TruncationModel=Qwen2.5-3B, Compression Setting=High++2026.01 | 18.2 | 299.73 | 52 | 6.07 | |
| TruncationModel=Qwen2.5-7B, Compression Setting=High++2026.01 | 18 | 300.29 | 52 | 5.99 | |
| TruncationModel=Qwen2.5-14B, Compression Setting=High++2026.01 | 16.6 | 301.09 | 52 | 5.51 | |
| TruncationModel=Qwen2.5-3B, Compression Setting=High+++2026.01 | 6.4 | 185.11 | 32 | 3.46 | |
| TruncationModel=Qwen2.5-14B, Compression Setting=High+++2026.01 | 5.4 | 204.71 | 35 | 2.64 | |
| TruncationModel=Qwen2.5-7B, Compression Setting=High+++2026.01 | 5 | 165.23 | 29 | 3.03 |