arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2304.09797cs.CLcs.LG
收藏

渐进提示提示法提升大语言模型的推理能力

Progressive-Hint Prompting Improves Reasoning in Large Language Models

CUHK(香港中文大学);Paris XI(巴黎第十一大学);HKU(香港大学);HKUST(香港科技大学)

Chuanyang Zheng, Zhengying Liu, Enze Xie, Zhenguo Li, Yu Li

AI总结:

针对现有大语言模型推理提示方法未充分利用生成答案的问题,提出渐进提示提示法PHP,通过多轮交互用历史答案引导推理,可结合CoT和自一致性,在多个基准上显著提升准确率并保持高效。

摘要:

大语言模型(LLMs)在推理任务中的表现高度依赖提示设计,思维链(CoT)和自一致性是增强该能力的关键方法。然而,这些方法并未充分利用大语言模型生成的答案来引导后续响应。本文提出一种名为Progressive-Hint Prompting(PHP,渐进提示提示法)的新型提示方法,它通过将先前生成的答案作为提示,逐步引导得出正确答案,实现用户与大语言模型之间的自动多次交互。PHP与CoT和自一致性是正交的,因此易于与前沿技术结合以进一步提升性能。我们在七个基准测试上开展了广泛全面的实验。结果表明,PHP在显著提升准确率的同时还保持了极高的效率。例如,在text-davinci-003模型上,采用贪心解码时,与Complex CoT相比,我们在GSM8K数据集上观察到4.2%的提升,而结合自一致性时,采样路径减少了46.17%。结合GPT-4和PHP,我们在SVAMP(89.1% -> 91.9%)、GSM8K(92% -> 95.5%)、AQuA(76.4% -> 79.9%)和MATH(50.3% -> 53.9%)上取得了当前最优性能。

英文摘要:

The performance of Large Language Models (LLMs) in reasoning tasks depends heavily on prompt design, with Chain-of-Thought (CoT) and self-consistency being critical methods that enhance this ability. However, these methods do not fully exploit the answers generated by the LLM to guide subsequent responses. This paper proposes a new prompting method, named Progressive-Hint Prompting (PHP), that enables automatic multiple interactions between users and LLMs by using previously generated answers as hints to progressively guide toward the correct answers. PHP is orthogonal to CoT and self-consistency, making it easy to combine with state-of-the-art techniques to further improve performance. We conducted extensive and comprehensive experiments on seven benchmarks. The results show that PHP significantly improves accuracy while remaining highly efficient. For instance, with text-davinci-003, we observed a 4.2% improvement on GSM8K with greedy decoding compared to Complex CoT, and a 46.17% reduction in sample paths with self-consistency. With GPT-4 and PHP, we achieve state-of-the-art performances on SVAMP (89.1% -> 91.9%), GSM8K (92% -> 95.5%), AQuA (76.4% -> 79.9%) and MATH (50.3% -> 53.9%).

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。

↑