arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2402.07754cs.CLcs.AIcs.LG

Diffusion of Thoughts:扩散语言模型中的思维链推理

Diffusion of Thoughts: Chain-of-Thought Reasoning in Diffusion Language Models

The University of Hong Kong(香港大学);Huawei Noah’s Ark Lab(华为诺亚方舟实验室);Tencent AI Lab(腾讯AI Lab)

Jiacheng Ye, Shansan Gong, Liheng Chen, Lin Zheng, Jiahui Gao, Han Shi, Chuan Wu, Xin Jiang, Zhenguo Li, Wei Bi, Lingpeng Kong

AI总结:

本文提出 Diffusion-of-Thought(DoT),让推理步骤在扩散语言模型中逐步扩散,以更灵活地权衡计算与性能,并在算术、逻辑和数学题上以小模型超越大型自回归模型。

摘要:

近来,扩散模型因其相较于传统自回归模型的诸多潜在优势,在文本处理领域引起了广泛关注。在本研究中,我们提出了 Diffusion-of-Thought(DoT),这是一种将扩散模型与思维链相结合的新颖方法;思维链是一项成熟的技术,可用于提升自回归语言模型的推理能力。与以从左到右、逐 token 方式作出决策的自回归语言模型不同,DoT 允许推理步骤通过扩散语言模型随时间扩散,并在计算量与推理性能之间的权衡上提供更大的灵活性。我们的实验结果证明了 DoT 在多位数乘法、布尔逻辑和小学数学问题上的有效性:一个小型扩散模型在效率和准确率两方面均优于规模大得多的自回归模型。除此之外,DoT 展现出良好的自我纠错能力,并能从自洽性解码等现有推理增强技术中获益。我们的发现有助于理解和发展基于扩散语言模型的推理。

英文摘要:

Recently, diffusion models have garnered significant interest in the field of text processing due to their many potential advantages compared to conventional autoregressive models. In this work, we propose Diffusion-of-Thought (DoT), a novel approach that integrates diffusion models with Chain-of-Thought, a well-established technique for improving the reasoning ability of autoregressive language models. In contrast to autoregressive language models that make decisions in a left-to-right, token-by-token manner, DoT allows reasoning steps to diffuse over time through a diffusion language model and offers greater flexibility in trading-off computation for reasoning performance. Our experimental results demonstrate the effectiveness of DoT in multi-digit multiplication, boolean logic, and grade school math problems, with a small diffusion model outperforming a much larger autoregressive model in both efficiency and accuracy. In addition to that, DoT showcases promising self-correction abilities and benefits from existing reasoning-enhancing techniques like self-consistency decoding. Our findings contribute to the understanding and development of reasoning with diffusion language models.

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。

↑