TroVE: Inducing Verifiable and Efficient Toolboxes for Solving Programmatic Tasks

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Wang, Zhiruo, Fried, Daniel, Neubig, Graham
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910306197831680
author Wang, Zhiruo
Fried, Daniel
Neubig, Graham
author_facet Wang, Zhiruo
Fried, Daniel
Neubig, Graham
contents Language models (LMs) can solve tasks such as answering questions about tables or images by writing programs. However, using primitive functions often leads to verbose and error-prone programs, and higher-level functions require expert design. To enable better solutions without human labor, we ask code LMs to curate reusable high-level functions, and use them to write solutions. We present TROVE, a training-free method of inducing a verifiable and efficient toolbox of functions, by generating via using, growing, and periodically trimming the toolbox. On 11 datasets from math, table question answering, and image reasoning tasks, TROVE consistently yields simpler solutions with higher accuracy than baselines using CODELLAMA and previous methods using GPT, while using 79-98% smaller toolboxes. TROVE further enables 31% faster and 13% more accurate human verification than baselines. With the same pipeline, it creates diverse functions for varied tasks and datasets, providing insights into their individual characteristics.
format Preprint
id arxiv_https___arxiv_org_abs_2401_12869
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle TroVE: Inducing Verifiable and Efficient Toolboxes for Solving Programmatic Tasks
Wang, Zhiruo
Fried, Daniel
Neubig, Graham
Artificial Intelligence
Language models (LMs) can solve tasks such as answering questions about tables or images by writing programs. However, using primitive functions often leads to verbose and error-prone programs, and higher-level functions require expert design. To enable better solutions without human labor, we ask code LMs to curate reusable high-level functions, and use them to write solutions. We present TROVE, a training-free method of inducing a verifiable and efficient toolbox of functions, by generating via using, growing, and periodically trimming the toolbox. On 11 datasets from math, table question answering, and image reasoning tasks, TROVE consistently yields simpler solutions with higher accuracy than baselines using CODELLAMA and previous methods using GPT, while using 79-98% smaller toolboxes. TROVE further enables 31% faster and 13% more accurate human verification than baselines. With the same pipeline, it creates diverse functions for varied tasks and datasets, providing insights into their individual characteristics.
title TroVE: Inducing Verifiable and Efficient Toolboxes for Solving Programmatic Tasks
topic Artificial Intelligence
url https://arxiv.org/abs/2401.12869