Introducing MAPO: Momentum-Aided Gradient Descent Prompt Optimization

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Cui, Anthony, Nandyalam, Pranav, Rufail, Andrew, Cheung, Ethan, Lei, Aiden, Zhu, Kevin, O'Brien, Sean
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908424549171200
author Cui, Anthony
Nandyalam, Pranav
Rufail, Andrew
Cheung, Ethan
Lei, Aiden
Zhu, Kevin
O'Brien, Sean
author_facet Cui, Anthony
Nandyalam, Pranav
Rufail, Andrew
Cheung, Ethan
Lei, Aiden
Zhu, Kevin
O'Brien, Sean
contents Momentum-Aided Prompt Optimization (MAPO) enhances the efficiency and efficacy of prompt optimization for Large Language Models (LLMs). Building on ProTeGi, MAPO uses positive natural language "gradients" and a momentum-based extension to refine prompts effectively. By tracking gradient history, MAPO avoids local minima and oscillations. It also utilizes beam search and an Upper Confidence Bound (UCB) algorithm for balanced candidate expansion and selection. Benchmark testing shows that MAPO achieves faster convergence time with fewer API calls and higher F1 scores than ProTeGi, proving it as a robust and scalable solution for automated prompt engineering in LLMs.
format Preprint
id arxiv_https___arxiv_org_abs_2410_19499
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Introducing MAPO: Momentum-Aided Gradient Descent Prompt Optimization
Cui, Anthony
Nandyalam, Pranav
Rufail, Andrew
Cheung, Ethan
Lei, Aiden
Zhu, Kevin
O'Brien, Sean
Computation and Language
Momentum-Aided Prompt Optimization (MAPO) enhances the efficiency and efficacy of prompt optimization for Large Language Models (LLMs). Building on ProTeGi, MAPO uses positive natural language "gradients" and a momentum-based extension to refine prompts effectively. By tracking gradient history, MAPO avoids local minima and oscillations. It also utilizes beam search and an Upper Confidence Bound (UCB) algorithm for balanced candidate expansion and selection. Benchmark testing shows that MAPO achieves faster convergence time with fewer API calls and higher F1 scores than ProTeGi, proving it as a robust and scalable solution for automated prompt engineering in LLMs.
title Introducing MAPO: Momentum-Aided Gradient Descent Prompt Optimization
topic Computation and Language
url https://arxiv.org/abs/2410.19499