You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Large language models (LLMs) can perform a wide range of tasks by followingnatural language instructions, without the necessity of task-specificfine-tuning. Unfortunately, the performance of LLMs is greatly influenced bythe quality of these instructions, and manually writing effective instructionsfor each task is a laborious and subjective process. In this paper, weintroduce Auto-Instruct, a novel method to automatically improve the quality ofinstructions provided to LLMs. Our method leverages the inherent generativeability of LLMs to produce diverse candidate instructions for a given task, andthen ranks them using a scoring model trained on a variety of 575 existing NLPtasks. In experiments on 118 out-of-domain tasks, Auto-Instruct surpasses bothhuman-written instructions and existing baselines of LLM-generatedinstructions. Furthermore, our method exhibits notable generalizability evenwith other LLMs that are not incorporated into its training process.
AkihikoWatanabe
changed the title
あ
Auto-Instruct: Automatic Instruction Generation and Ranking for
Black-Box Language Models, Zhihan Zhang+, N/A, arXiv'23
Oct 26, 2023
URL
Affiliations
Abstract
Translation (by gpt-3.5-turbo)
残念ながら、LLMsのパフォーマンスは、これらの指示の品質に大きく影響を受けます。
そして、各タスクの効果的な指示を手動で作成することは、手間がかかり主観的なプロセスです。
本論文では、LLMsに提供される指示の品質を自動的に向上させるための新しい手法であるAuto-Instructを紹介します。
私たちの手法は、LLMsの持つ生成能力を活用して、与えられたタスクに対して多様な候補の指示を生成し、
575の既存のNLPタスクでトレーニングされたスコアリングモデルを使用してそれらをランク付けします。
118のドメイン外のタスクにおける実験では、Auto-Instructは人間による指示と既存のLLM生成指示の両方を上回ります。
さらに、私たちの手法は、トレーニングプロセスに組み込まれていない他のLLMsでも顕著な汎化性能を示します。
Summary (by gpt-3.5-turbo)
この手法では、LLMsが生成する指示の品質を自動的に向上させるために、多様な候補の指示を生成し、スコアリングモデルでランク付けします。
実験結果では、Auto-Instructが人間による指示や既存のLLM生成指示を上回ることが示されています。
また、他のLLMsでも顕著な汎化性能を示すことも確認されています。
The text was updated successfully, but these errors were encountered: