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 2404.13033cs.CL
收藏

样本设计工程:什么造就了 LLM 良好的下游微调样本的实证研究

Sample Design Engineering: An Empirical Study of What Makes Good Downstream Fine-Tuning Samples for LLMs

AI Lab, SIME, Shanghai University of Finance and Economics(上海财经大学SIME AI实验室);Guangdong Yunxi Technology(广东云溪科技);Ant Group(蚂蚁集团)

Biyang Guo, He Wang, Wenyilin Xiao, Hong Chen, Zhuxin Lee, Songqiao Han, Hailiang Huang

AI总结:

本研究提出样本设计工程(SDE),通过优化输入、输出和推理设计提升 LLM 微调性能,实验证明其在复杂任务中优于启发式方法。

摘要:

在诸如 ChatGPT 和 LLaMA 等大型语言模型(LLM)蓬勃发展的领域中,提示工程(PE)因通过提示修改增强零样本或上下文学习(ICL)而闻名。然而,对于下游微调的样本设计领域,这对任务特定的 LLM 适配至关重要,却 largely 未被探索。本文引入了样本设计工程(SDE),一种通过优化输入、输出和推理设计来提升 LLM 调优后性能的系统性方法。我们进行了一系列领域内(ID)和领域外(OOD)实验,以评估各种设计选项对 LLM 下游性能的影响,揭示了在不同 LLM 中一致成立的几个有趣模式。基于这些见解,我们提出了一个集成的 SDE 策略,结合最有效的选项,并验证了其在复杂下游任务(如多情感分析、事件抽取和嵌套实体识别)中相对于启发式样本设计的持续优越性。此外,对 LLM 固有的提示/输出困惑度、零样本和 ICL 能力的分析表明,好的 PE 策略可能并不总是转化为好的 SDE 策略。代码可在 https://github.com/beyondguo/LLM-Tuning 获取。

英文摘要:

In the burgeoning field of Large Language Models (LLMs) like ChatGPT and LLaMA, Prompt Engineering (PE) is renowned for boosting zero-shot or in-context learning (ICL) through prompt modifications. Yet, the realm of the sample design for downstream fine-tuning, crucial for task-specific LLM adaptation, is largely unexplored. This paper introduces Sample Design Engineering (SDE), a methodical approach to enhancing LLMs' post-tuning performance by refining input, output, and reasoning designs. We conduct a series of in-domain (ID) and out-of-domain (OOD) experiments to assess the impact of various design options on LLMs' downstream performance, revealing several intriguing patterns that hold consistently across different LLMs. Based on these insights, we propose an integrated SDE strategy, combining the most effective options, and validate its consistent superiority over heuristic sample designs in complex downstream tasks like multi-aspect sentiment analysis, event extraction, and nested entity recognition. Additionally, analyses of LLMs' inherent prompt/output perplexity, zero-shot, and ICL abilities illustrate that good PE strategies may not always translate to good SDE strategies. Code available at https://github.com/beyondguo/LLM-Tuning.

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。

↑