Mathematical Reasoning on MATH500 (Accuracy and Generation Length)
80.6AccuracyDeepSeek-R1-1.5B
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| DeepSeek-R1-1.5BReasoning Strategy=Long-CoT2026.01 | 80.6 | 67 | |
| CoDMethod Category=Prompt-guided2026.01 | 80.2 | 3,486 | |
| ACMMethod Category=Data-dependent Merging2026.01 | 78.8 | 1,638 | |
| RPAMMethod Category=Data-dependent Merging2026.01 | 78.8 | 1,427 | |
| Ada-R1Method Category=Training-based2026.01 | 76.8 | 887 | |
| AIMMethod Category=Data-dependent Merging2026.01 | 76 | 1,071 | |
| Average MergingMethod Category=Data-free Merging2026.01 | 74.8 | 934 | |
| TIES MergingMethod Category=Data-free Merging2026.01 | 72.6 | 1,619 | |
| Task ArithmeticMethod Category=Data-free Merging2026.01 | 62.8 | 692 | |
| DARE-LinearMethod Category=Data-free Merging2026.01 | 57.2 | 780 | |
| Qwen2.5-Math-1.5BReasoning Strategy=Short-CoT2026.01 | 36.2 | 411 |