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

Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics, pages 3845–3864 May 2-6, 2023 ©2023 Association for Computational Linguistics GRIPS: Gradient-free, Edit-based Instruction Search for Prompting Large Language Models Archiki Prasad Peter Hase Xiang Zhou Mohit Bansal UNC Chapel Hill {archiki, peter, xzh, mbansal}@cs.unc.edu Abstract Providing natural language instructions in prompts is a useful new paradigm for improv- ing task performance of large language models in a zero-shot setting. Recent work has aimed to improve such prompts via manual rewrit- ing or gradient-based tuning. However, man- ual rewriting is time-consuming and requires subjective interpretation, while gradient-based tuning can be extremely computationally de- manding for large models and may not be fea- sible 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 in- structions designed for humans and automati- cally returns an improved, edited prompt, while allowing for API-based tuning. With Instruct- GPT models, GRIPS improves the average task performance by up to 4.30 percentage points on eight classification tasks from the NATU- RAL-INSTRUCTIONS dataset (with similar im- provements 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 ap- proaches. Qualitatively, we show our edits can simplify instructions and at times make them incoherent but nonetheless improve accuracy.1 1 Introduction Recent advancements in prompting large language models (LMs) such as GPT-3 show that models can perform NLP tasks without any task-specific tuning (Brown et al., 2020). Most of the work in this area focuses on few-shot learning, where mod- els rely on textual prompts containing input-output 1Code: https://github.com/archiki/GrIPS example pairs (exemplar prompts). However, hu- mans are often able to perform a new task when provided with a relevant set of instructions or a task description, not necessarily including any ex- amples. In this direction, past works explore a new paradigm of instructional prompts where a prompt is tailored for a particular task by includ- ing natural language instructions (Efrat and Levy, 2020; Mishra et al., 2022a,b). Following Webson and Pavlick (2021), we characterize instructions as a natural language description of the task that includes what is required for a person to complete the task correctly.2 Demonstrative examples of the task are not considered a part of the instructions. For purposes of improving task performance via instructional prompts, Mishra et al. (2022b) pro- vide a set of guidelines to manually rewrite raw instructions. Yet this kind of rewriting process requires substantial manual effort and subjective interpretation of the guidelines. In addition, an un- derlying assumption in Mishra et al. (2022b) is that instructions should be semantically coherent to hu- mans. However, it is possible that the prompts that most improve model performance are semantically confusing to humans …

Content truncated to 500 words