Mathematical Reasoning on MATH 500 (Pass@1, FLOPS)
38.2Pass@1ϕ-Decoding
Evaluation results
| Method | Links | ||
|---|---|---|---|
| ϕ-DecodingBackbone=LLaMA3.1-8B-Instruct2026.01 | 38.2 | — | |
| MFS (Ours)Backbone=LLaMA3.1-8B-Instruct2026.01 | 38.2 | — | |
| MCTSBackbone=LLaMA3.1-8B-Instruct2026.01 | 34.4 | — | |
| Predictive DecodingBackbone=LLaMA3.1-8B-Instruct2026.01 | 34 | — | |
| Tree-of-ThoughtsBackbone=LLaMA3.1-8B-Instruct2026.01 | 31.6 | — | |
| Guided DecodingBackbone=LLaMA3.1-8B-Instruct2026.01 | 31.2 | — | |
| Auto-Regressive (CoT)Backbone=LLaMA3.1-8B-Instruct2026.01 | 31 | — | |
| MFS (Ours)Backbone=Mistral-v0.3-7B-Instruct2026.01 | 17.6 | — | |
| ϕ-DecodingBackbone=Mistral-v0.3-7B-Instruct2026.01 | 16.4 | — | |
| Auto-Regressive (CoT)Backbone=Mistral-v0.3-7B-Instruct2026.01 | 12.2 | — | |
| Predictive DecodingBackbone=Mistral-v0.3-7B-Instruct2026.01 | 11 | — | |
| Tree-of-ThoughtsBackbone=Mistral-v0.3-7B-Instruct2026.01 | 10.8 | — | |
| MCTSBackbone=Mistral-v0.3-7B-Instruct2026.01 | 10.8 | — | |
| Guided DecodingBackbone=Mistral-v0.3-7B-Instruct2026.01 | 10.8 | — |