InstructEval: Systematic Evaluation of Instruction Selection Methods

Anirudh Ajith, Chris Pan, Mengzhou Xia, Ameet Deshpande, Karthik Narasimhan


Abstract
In-context learning (ICL) performs tasks by prompting a large language model (LLM) using an instruction and a small set of annotated examples called demonstrations. Recent work has shown that precise details of the inputs used in the ICL prompt significantly impact performance, which has incentivized instruction selection algorithms. The effect of instruction-choice however is severely underexplored, with existing analyses restricted to shallow subsets of models and tasks, limiting the generalizability of their insights. We develop InstructEval, an ICL evaluation suite to conduct a thorough assessment of these techniques. The suite includes 13 open-sourced LLMs of varying scales from four model families, and covers nine tasks across three categories. Using the suite, we evaluate the relative performance of seven popular instruction selection methods over five metrics relevant to ICL. Our experiments reveal that using curated manually-written instructions or simple instructions without any task-specific descriptions often elicits superior ICL performance overall than that of automatic instruction-induction methods, pointing to a lack of generalizability among the latter. We release our evaluation suite (at https://github.com/princeton-nlp/InstructEval) for benchmarking instruction selection approaches and enabling more generalizable methods in this space.
Anthology ID:
2024.findings-naacl.270
Volume:
Findings of the Association for Computational Linguistics: NAACL 2024
Month:
June
Year:
2024
Address:
Mexico City, Mexico
Editors:
Kevin Duh, Helena Gomez, Steven Bethard
Venue:
Findings
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
4336–4350
Language:
URL:
https://aclanthology.org/2024.findings-naacl.270
DOI:
Bibkey:
Cite (ACL):
Anirudh Ajith, Chris Pan, Mengzhou Xia, Ameet Deshpande, and Karthik Narasimhan. 2024. InstructEval: Systematic Evaluation of Instruction Selection Methods. In Findings of the Association for Computational Linguistics: NAACL 2024, pages 4336–4350, Mexico City, Mexico. Association for Computational Linguistics.
Cite (Informal):
InstructEval: Systematic Evaluation of Instruction Selection Methods (Ajith et al., Findings 2024)
Copy Citation:
PDF:
https://aclanthology.org/2024.findings-naacl.270.pdf
Copyright:
 2024.findings-naacl.270.copyright.pdf