Mathematical Reasoning on SVAMP (Accuracy, Avg Tok, Avg Steps)
92.33AccuracyMOC
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| MOCEdge Density=0.3, Backbone=Gemma-2-27B-Instruct2026.06 | 92.33 | — | — | |
| Vanilla MASEdge Density=1.0, Backbone=Gemma-2-27B-Instruct2026.06 | 92 | — | — | |
| MOCEdge Density=1.0, Backbone=Gemma-2-27B-Instruct2026.06 | 91.67 | — | — | |
| MOCEdge Density=0.7, Backbone=Gemma-2-27B-Instruct2026.06 | 91.33 | — | — | |
| Vanilla MASEdge Density=0.3, Backbone=Gemma-2-27B-Instruct2026.06 | 90.67 | — | — | |
| MOCEdge Density=0.5, Backbone=Gemma-2-27B-Instruct2026.06 | 90.67 | — | — | |
| Vanilla MASEdge Density=0.7, Backbone=Gemma-2-27B-Instruct2026.06 | 90.67 | — | — | |
| Vanilla MASEdge Density=0.5, Backbone=Gemma-2-27B-Instruct2026.06 | 90.33 | — | — | |
| Single LLMBackbone=Gemma-2-27B-Instruct2026.06 | 86.33 | — | — | |
| CoTModel=Qwen2.5-1.5B2026.03 | 59.3 | 29.09 | — | |
| CoTModel=Llama-3.2-1B2026.03 | 57.8 | 28.21 | — | |
| AdaAnchor (adaptive)Model=Qwen2.5-1.5B2026.03 | 55.2 | 2.23 | 4.12 | |
| iCoTModel=Llama-3.2-1B2026.03 | 54.2 | 2.43 | — | |
| AdaAnchor (adaptive)Model=Llama-3.2-1B2026.03 | 53.4 | 2.8 | 3.1 | |
| AdaAnchor (K=8)Model=Llama-3.2-1B2026.03 | 52 | 2.13 | 8 | |
| AdaAnchor (K=8)Model=Qwen2.5-1.5B2026.03 | 50.5 | 2.12 | 8 | |
| iCoTModel=Qwen2.5-1.5B2026.03 | 48.5 | 2.04 | — | |
| No CoTModel=Qwen2.5-1.5B2026.03 | 42 | 2.34 | — | |
| No CoTModel=Llama-3.2-1B2026.03 | 38.2 | 2.1 | — |