Mathematical Reasoning on MATH 500 (Accuracy, Tokens, ActRatio)
48.8AccuracyASCoT
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| ASCoTBackbone=LLaMA-3.1-8B-Instruct, Ratio=1.02025.08 | 48.8 | 478.8 | 95 | |
| OriginalBackbone=LLaMA-3.1-8B-Instruct, Ratio=1.02025.08 | 48.6 | 502.6 | 100 | |
| PromptBackbone=LLaMA-3.1-8B-Instruct, Ratio=0.92025.08 | 48.6 | 468.04 | 93 | |
| PromptBackbone=LLaMA-3.1-8B-Instruct, Ratio=0.72025.08 | 48.4 | 472.13 | 94 | |
| PromptBackbone=LLaMA-3.1-8B-Instruct, Ratio=0.52025.08 | 47.8 | 471.11 | 94 | |
| TruncationBackbone=LLaMA-3.1-8B-Instruct, Ratio=0.92025.08 | 47.8 | 440.33 | 88 | |
| ASCoTBackbone=LLaMA-3.1-8B-Instruct, Ratio=0.92025.08 | 47.6 | 427.15 | 84 | |
| ASCoTBackbone=LLaMA-3.1-8B-Instruct, Ratio=0.82025.08 | 47.2 | 376.72 | 74 | |
| ASCoTBackbone=LLaMA-3.1-8B-Instruct, Ratio=0.72025.08 | 46.8 | 334.26 | 67 | |
| TruncationBackbone=LLaMA-3.1-8B-Instruct, Ratio=0.72025.08 | 45 | 386.89 | 77 | |
| ASCoTBackbone=LLaMA-3.1-8B-Instruct, Ratio=0.62025.08 | 41.8 | 309.69 | 61 | |
| ASCoTBackbone=LLaMA-3.1-8B-Instruct, Ratio=0.52025.08 | 40.1 | 288.94 | 57 | |
| TruncationBackbone=LLaMA-3.1-8B-Instruct, Ratio=0.52025.08 | 27.4 | 283.7 | 56 |