arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2405.16265cs.LG
收藏

MindStar:在推理时增强预训练大语言模型的数学推理能力

MindStar: Enhancing Math Reasoning in Pre-trained LLMs at Inference Time

Noah’s Ark Laboratory(诺亚方舟实验室);Huawei(华为)

Jikun Kang, Xin Zhe Li, Xi Chen, Amirreza Kazemi, Qianyi Sun, Boxing Chen, Dong Li, Xu He, Quan He, Feng Wen, Jianye Hao, Jun Yao

AI总结:

提出纯推理搜索方法MindStar,将数学推理视为搜索问题,显著提升开源模型推理能力,性能媲美GPT-3.5但成本更低。

摘要:

尽管大语言模型(LLMs)在各种任务中表现出色,但它们往往难以处理复杂的推理任务,例如回答数学问题。近期解决这一问题的努力主要集中在利用数学数据集进行监督微调或自我改进技术。然而,这些方法通常依赖于难以准备的高质量数据集,或者需要大量的计算资源进行微调。受LLMs知道如何产生正确答案但难以选择正确推理路径这一发现的启发,我们提出了一种纯推理的搜索方法——MindStar(M*)。该方法将推理任务形式化为搜索问题,并提出了两种搜索思路以识别最优推理路径。我们在GSM8K和MATH数据集上评估了M*框架,并将其性能与现有的开源和闭源LLMs进行了比较。我们的结果表明,M*显著增强了开源模型(如Llama-2-13B和Mistral-7B)的推理能力,并实现了与GPT-3.5和Grok-1相当的性能,但模型规模和计算成本大幅降低。

英文摘要:

Although Large Language Models (LLMs) achieve remarkable performance across various tasks, they often struggle with complex reasoning tasks, such as answering mathematical questions. Recent efforts to address this issue have primarily focused on leveraging mathematical datasets through supervised fine-tuning or self-improvement techniques. However, these methods often depend on high-quality datasets that are difficult to prepare, or they require substantial computational resources for fine-tuning. Inspired by findings that LLMs know how to produce the right answer but struggle to select the correct reasoning path, we propose a purely inference-based searching method -- MindStar (M*). This method formulates reasoning tasks as searching problems and proposes two search ideas to identify the optimal reasoning paths. We evaluate the M* framework on both the GSM8K and MATH datasets, comparing its performance with existing open and closed-source LLMs. Our results demonstrate that M* significantly enhances the reasoning abilities of open-source models, such as Llama-2-13B and Mistral-7B, and achieves comparable performance to GPT-3.5 and Grok-1, but with substantially reduced model size and computational costs.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑