Automatic Model Selection with Large Language Models for Reasoning

TL;DR

This paper presents a novel method using large language models to dynamically select the most effective reasoning approach—either Chain-of-Thought or Program-Aided Language Models—for improved accuracy and efficiency across multiple datasets.

Contribution

The paper introduces a model selection technique leveraging LLMs to choose between CoT and PAL methods, achieving state-of-the-art results and reducing computational costs.

Findings

Significant performance improvements on eight reasoning datasets.

Achieved new state-of-the-art accuracies on GSM8K and SVAMP.

Complementary to self-consistency, further enhancing performance.

Abstract

Chain-of-Thought (CoT) and Program-Aided Language Models (PAL) represent two distinct reasoning methods, each with its own strengths. CoT employs natural language, offering flexibility and interpretability, while PAL utilizes programming language, yielding more structured and rigorous logic. We introduce a model selection method to combine the best of both worlds by employing a large language model (LLM) to dynamically select between them. Our theoretical analysis underscores the feasibility of this method, which is further corroborated by empirical results. Our proposed method demonstrates significant performance improvements across eight reasoning datasets with Codex, ChatGPT, and GPT-4. Additionally, our method is complementary to self-consistency; when integrated, it can further enhance performance while significantly reducing computation costs. Moreover, we achieve new state-of-the-art results on GSM8K and SVAMP, with respective accuracies of 96.8% and 93.7%. Our code, data and prompts are available at https://github.com/XuZhao0/Model-Selection-Reasoning