Husky: A Unified, Open-Source Language Agent for Multi-Step Reasoning

TL;DR

Husky is an open-source, unified language agent capable of multi-step reasoning across diverse tasks, outperforming prior models and matching larger models like GPT-4 on complex reasoning benchmarks.

Contribution

Husky introduces a holistic, open-source language agent with a unified action space and curated expert models, enabling versatile reasoning over multiple complex tasks.

Findings

Husky outperforms prior agents on 14 datasets.

Husky matches or exceeds GPT-4 on mixed-tool reasoning tasks.

Husky effectively handles numerical, tabular, and knowledge-based reasoning.

Abstract

Language agents perform complex tasks by using tools to execute each step precisely. However, most existing agents are based on proprietary models or designed to target specific tasks, such as mathematics or multi-hop question answering. We introduce Husky, a holistic, open-source language agent that learns to reason over a unified action space to address a diverse set of complex tasks involving numerical, tabular, and knowledge-based reasoning. Husky iterates between two stages: 1) generating the next action to take towards solving a given task and 2) executing the action using expert models and updating the current solution state. We identify a thorough ontology of actions for addressing complex tasks and curate high-quality data to train expert models for executing these actions. Our experiments show that Husky outperforms prior language agents across 14 evaluation datasets. Moreover, we introduce HuskyQA, a new evaluation set which stress tests language agents for mixed-tool reasoning, with a focus on retrieving missing knowledge and performing numerical reasoning. Despite using 7B models, Husky matches or even exceeds frontier LMs such as GPT-4 on these tasks, showcasing the efficacy of our holistic approach in addressing complex reasoning problems. Our code and models are available at https://github.com/agent-husky/Husky-v1.