notesum.ai
Published at November 26BPP-Search: Enhancing Tree of Thought Reasoning for Mathematical Modeling Problem Solving
cs.AI
Released Date: November 26, 2024
Authors: Teng Wang1, Wing-Yin Yu2, Zhenqi He1, Zehua Liu1, Xiongwei Han3, Hailei Gong3, Han Wu2, Wei Shi2, Ruifeng She2, Fangzhou Zhu3, Tao Zhong3
Aff.: 1Department of Mathematics, the University of Hong Kong, Hong Kong SAR, China; 2Noah's Ark Lab, Huawei, Hong Kong SAR, China; 3Noah's Ark Lab, Huawei, Shenzhen, China

| Model/Methods | CoT-BMLD | CoT-SPVOC | SC | ToT-Randomly-Chosen | ToT-Rethink | ToT-Fully-Traverse |
| GPT-4o | 19 | 19 | 21 | 21 | 23 | 30 |
| GPT-4o-mini | 13 | 14 | 19 | 12 | 19 | 21 |
| Llama-3-70B | 10 | 17 | 19 | 17 | 17 | 22 |
| Llama-3.1-70B | 18 | 10 | 21 | 5 | 14 | 17 |
| Llama-3.2-11B | 2 | 0 | 1 | 0 | 0 | 8 |
| Qwen-2-72B-Instruct | 2 | 2 | 4 | 3 | 1 | 5 |
| Qwen-2.5-MATH-72B-Instruct | 2 | 2 | 0 | 0 | 0 | 4 |
| Qwen-2.5-72B-Instruct | 6 | 5 | 9 | 2 | 2 | 2 |
| Mixtral-87B-v0.1 | 0 | 0 | 0 | 0 | 1 | 1 |