arXivDaily arXiv每日学术速递 周一至周五更新
北美退伍军人节(Veterans Day),12 日无更新
arXiv 2203.07281cs.CLcs.AIcs.LG
收藏

GrIPS:用于提示大型语言模型的无梯度、基于编辑的指令搜索

GrIPS: Gradient-free, Edit-based Instruction Search for Prompting Large Language Models

UNC Chapel Hill(北卡罗来纳大学教堂山分校)

Archiki Prasad, Peter Hase, Xiang Zhou, Mohit Bansal

AI总结:

本文提出无梯度、基于编辑的提示指令搜索方法 GrIPS,可自动改写人类指令并支持 API 模型调优,在多模型和八类任务上显著提升性能。

摘要:

在零样本设置中,于提示里提供自然语言指令是提升大型语言模型任务性能的一种有用的新范式。近期工作旨在通过人工重写或基于梯度的调优来改进此类提示。然而,人工重写耗时且需要主观解读,而基于梯度的调优对大型模型而言可能计算需求极高,并且对于基于 API 的模型可能不可行。在本文中,我们提出了 Gradient-free Instructional Prompt Search(GrIPS),这是一种无梯度、基于编辑的搜索方法,用于改进大型语言模型的任务指令。GrIPS 接收为人类设计的指令,并自动返回经过编辑的改进提示,同时支持基于 API 的调优。在 InstructGPT 模型上,GrIPS 在 Natural Instructions 数据集的八个分类任务上将平均任务性能最高提升 4.30 个百分点(在 OPT、BLOOM 和 FLAN-T5 上也有相近提升)。我们在仅含指令的提示以及“指令 + k-shot 示例”提示中均观察到性能提升。值得注意的是,在控制可用计算和数据预算的情况下,GrIPS 优于人工重写和纯基于示例的提示。此外,GrIPS 的性能可与部分基于梯度的调优方法相媲美。在定性分析中,我们表明所做编辑可以简化指令,有时还会使指令变得不连贯,但仍能提高准确率。我们的代码可在 https://github.com/archiki/GrIPS 获取。

英文摘要:

Providing natural language instructions in prompts is a useful new paradigm for improving task performance of large language models in a zero-shot setting. Recent work has aimed to improve such prompts via manual rewriting or gradient-based tuning. However, manual rewriting is time-consuming and requires subjective interpretation, while gradient-based tuning can be extremely computationally demanding for large models and may not be feasible for API-based models. In this work, we introduce Gradient-free Instructional Prompt Search (GrIPS), a gradient-free, edit-based search approach for improving task instructions for large language models. GrIPS takes in instructions designed for humans and automatically returns an improved, edited prompt, while allowing for API-based tuning. With InstructGPT models, GrIPS improves the average task performance by up to 4.30 percentage points on eight classification tasks from the Natural Instructions dataset (with similar improvements for OPT, BLOOM, and FLAN-T5). We see improvements for both instruction-only prompts and instruction + k-shot examples prompts. Notably, GrIPS outperforms manual rewriting and purely example-based prompts while controlling for the available compute and data budget. Further, performance of GrIPS is comparable to select gradient-based tuning approaches. Qualitatively, we show our edits can simplify instructions and at times make them incoherent but nonetheless improve accuracy. Our code is available at: https://github.com/archiki/GrIPS

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。

↑